YALineo
EN
Install

Tutorials

From installing to the cost of every table, step by step. Each tutorial says what to have first, every click, what you should see, and what to do when you do not.

The video tutorial, step by step

One business case, followed from the first setting to the budget.

Part 0 · The need · 1 min 26 s Eight narrated and subtitled parts, recorded on an earlier version of the studio: some screens have changed since.

Where to start

Getting startedData and lineageCostsSimulationsGoing furtherGovernance

Contents

57 tutorials. The screens show the example platform shipped with the extension (fictional data; on the screenshots, the workspace address is masked).

Getting started 11 tutorials

  1. Install YALineo2 min
  2. Connect to your workspace2 min
  3. Choose the scope and load the map3 min
  4. Pick your profile3 min
  5. The settings: find everything, choose your language2 min
  6. Pick your style1 min
  7. The chip and the costs in Databricks3 min
  8. Find your way in the studio4 min
  9. Right click, Ctrl K and your shortcuts3 min
  10. To review: findings and your follow-up3 min
  11. Learn: a path per profile and the dictionary3 min

Data and lineage 14 tutorials

  1. Read the platform map7 min
  2. Change the period of the map2 min
  3. Isolate a station (focus)3 min
  4. Find a catalog, a schema or follow a table4 min
  5. The dependencies of an asset4 min
  6. The sheet of an asset3 min
  7. Why no link, why this cost3 min
  8. Impact analysis: before changing a table or a job5 min
  9. The health check3 min
  10. The data catalog: search, filter, read the sheet4 min
  11. The business glossary: extract, import, validate, publish6 min
  12. Descriptions and tags: propose, validate, apply5 min
  13. Welcome a new employee2 min
  14. Export the maps, pages, a view, see what changed5 min

Costs 11 tutorials

  1. Read the cost overview4 min
  2. Choose the period and read the changes3 min
  3. Filter: a team, a catalog, a product3 min
  4. What each catalog and schema costs4 min
  5. Find the tables with no read observed2 min
  6. Jobs and compute: who uses what3 min
  7. SQL warehouses, AI and Genie, teams and tags4 min
  8. Projections: end of month, 30, 90 or 365 days3 min
  9. Sort and adjust the tables2 min
  10. What a dashboard costs2 min
  11. Cost analysis: compose, save, explore5 min

Simulations 14 tutorials

  1. Learn with the worked cases4 min
  2. Price a workload before you build it6 min
  3. Add a service: AI, models, data and platform4 min
  4. Cost a data project5 min
  5. Size and tune: the fourteen calculations7 min
  6. Data engineering: ingest, finish on time, tune a job, a stream, a Delta table6 min
  7. Data science: shared cluster, training run, model endpoint4 min
  8. Analytics and dashboards: warehouse at peak, what a dashboard costs4 min
  9. Architecture: storage, Photon, spot4 min
  10. AI: Genie and agents, build and price6 min
  11. Compare scenarios and share them3 min
  12. Make a measured workload vary4 min
  13. Prices, discounts and the price snapshot3 min
  14. SKU prices, token prices, your account's prices3 min

Going further 7 tutorials

  1. Grant access (for admins)3 min
  2. Without system tables: analyst mode2 min
  3. Large platforms: a lighter load3 min
  4. Keep or erase your data2 min
  5. Pre-flight in pull requests3 min
  6. Troubleshooting4 min
  7. Under the hood2 min

Getting started · 2 min

Install YALineo

Contents

YALineo installs like any extension, from your browser's store. It runs on Chrome, Edge and Firefox, with any Databricks workspace on Azure, AWS or GCP.

Before you start

  • An up-to-date Chrome, Edge or Firefox.
  • A Databricks workspace with Unity Catalog that you can sign in to.
  1. Open the YALineo listing. From this site's home page, "Install" section, pick your browser: Chrome Web Store, Microsoft Edge Add-ons or Firefox Add-ons.
  2. Add the extension. Click "Add" (or "Get" on Edge). The browser asks for access to the Databricks sites (azuredatabricks.net, cloud.databricks.com, gcp.databricks.com). Those are the only sites YALineo can read.
  3. Pin the icon. Chrome: the puzzle-piece icon, then the pin next to YALineo. Edge: the "Extensions" icon, then "Show in toolbar". Firefox: the extensions button, YALineo's gear, "Pin to Toolbar".
  4. Let the settings page open. It opens on its own after installing, on "Settings". Move on to the next tutorial.

To check The YALineo icon (a station on a line) shows in the browser toolbar, and the "Settings" page is open.

If it does not work1 case
The settings page did not open.
Right-click the YALineo icon, then choose "Options" (Chrome, Edge) or "Manage Extension", "Options" tab (Firefox).

Good to know Only install YALineo from the official stores or the links on this site, and check that the publisher is YALineo. A copy from elsewhere may have been modified.

Getting started · 2 min

Connect to your workspace

Contents

YALineo reads your workspace with your own permissions. It sees only what Databricks already shows you. With the Databricks session, there is nothing to create or paste.

Before you start

  • A tab open on your Databricks workspace, where you are signed in.
  1. Step 1, "Workspace". YALineo picks up the workspace address from the open Databricks tab ("Open Databricks tabs" list). You can also type it. It is the address your browser shows when you are in Databricks.
  2. Step 2, "Connection": keep "Databricks session". This is the recommended choice. YALineo reads through the Databricks tab, with your session. Keep that tab open while loading.
  3. Or choose "Personal access token". In Databricks: your profile, "Settings", "Developer", "Access tokens", "Generate new token". Paste it in the "Token" field of the extension's settings, and nowhere else. Tick "Remember the token on this device" if you do not want to paste it again at each browser start.
  4. Click "Test the connection". YALineo makes a single read and shows how many SQL warehouses are visible.

To check The message "Connected: N SQL warehouses visible." shows under the button.

If it does not work4 cases
"The session was refused".
Show the Databricks tab for a few seconds (it renews its token while it works) or reload it with F5, then click "Test the connection" again. YALineo shows these steps, in order.
"The browser paused the Databricks tab".
YALineo says so at once instead of staying on "Loading". With its energy saver, Chrome freezes a busy tab left a long time in the background. Click "Show the Databricks tab and try again". YALineo shows the tab for a moment to wake it, comes back and tries again. During a load of the map, YALineo pauses the load and resumes it as soon as the tab is visible.
"The Databricks tab does not answer YALineo".
The tab is busy (a dialog open, a page that no longer answers). The test stops after 75 s instead of staying on "Loading". Show the tab, close what keeps it busy or reload it, then try again.
It is still refused.
Switch to the personal token with "Switch to a token". It always works. "Send the diagnostic" prepares an e-mail to support with no secret value.

Good to know A remembered token stays on this device, in the extension's storage. "Forget", at the top of the settings, erases the workspace, its token and its maps. YALineo will never ask for your Databricks token, password or sign-in codes, on this site or by e-mail.

See the screenshot
Settings, steps 1 and 2: the workspace address picked up from the open tab, the Databricks session, and a successful test.
Settings, steps 1 and 2: the workspace address picked up from the open tab, the Databricks session, and a successful test.

Getting started · 3 min

Choose the scope and load the map

Contents

The scope sets what YALineo reads: which catalogs, how much history, with or without costs. A small serverless SQL warehouse is enough.

Before you start

  • The connection tested (previous tutorial).
  • A SQL warehouse you can use.
  • For the full map, read access to system.access; for costs, to system.billing ("Grant access" tutorial). Without them, analyst mode still draws the map.
  1. Pick the SQL warehouse. A small serverless one starts fast and is enough. YALineo only uses it during a load.
  2. Click "List catalogs", then tick the ones to draw. With none ticked, YALineo draws them all. "Catalogs shown": "This workspace's catalogs" (recommended), or the whole metastore.
  3. Set the history. "Lineage history loaded": one year (recommended), or less for a lighter map. "Activity period (days)": the window for reads, delays and faded links (90 by default). You change them later from the map's calendar, without coming back here ("Change the period of the map" tutorial).
  4. Choose the lineage source. "System tables" for the full platform map, or "Analyst mode" if you have no access to them.
  5. FinOps side. Keep "Load costs with the map" ticked, and enter your negotiated discount on the list price if you have one (in %).
  6. Click "Check lineage access". The message "Access to system.access.table_lineage confirmed." means all is ready. On a large platform, also click "Estimate the volume" ("Large platforms" tutorial).
  7. Step 4, "Your profile". "Full" (the default) opens every page. Another profile puts the pages of your role first and a padlock on the others ("Pick your profile" tutorial).
  8. Step 5, "Privacy". "Keep nothing after the browser closes" (default), or "Keep the map and its history on this device" to compare versions of the map over time.
  9. Click "Save and load the map". A pipe fills step by step, with a percentage that only goes up. Count a few seconds to a few minutes depending on the platform.
  10. Open the studio. While the map loads, YALineo locks the settings. "Cancel and return to the settings" stops the load and keeps the previous map. Once the map is in, "Open YALineo Studio" opens the studio on this workspace, and "Edit the settings" opens the form again.

To check On the Databricks tab, the YALineo chip turns orange while loading, then green. The settings show "Settings saved".

If it does not work2 cases
"No access to system tables".
An admin must run the grant script ("Grant access" tutorial), or switch to "Analyst mode".
"Incomplete load" after loading.
A query reached its limit or its time-out. Follow the tips shown ("Large platforms" tutorial).

Good to know Every query is a SELECT tagged "yalineo". Your admins find them in the Databricks query history.

See the screenshot
Settings, step 3: the SQL warehouse, the catalogs, the lineage source, the FinOps side and the access check.
Settings, step 3: the SQL warehouse, the catalogs, the lineage source, the FinOps side and the access check.

Getting started · 3 min

Pick your profile

Contents

Eight profiles: "Full", the default, which includes everything, and seven roles. A role puts its pages and calculations first, and a padlock on the others. The profile helps you focus on your job. It protects nothing. Anyone can change it at any time, and YALineo still reads the data with your own Databricks permissions.

Before you start

  • The YALineo settings open.
  1. Go to step 4, "Your profile". In the settings, between "Scope" and "Privacy". From the studio, "Change", on the home page, leads straight there.
  2. Keep "Full", or pick your role. "Full": every page and every calculation, no padlock. The roles: Business, Data steward or owner, Data developer, Data architect, FinOps lead, Data scientist, BI analyst. YALineo saves the choice on the click ("Profile saved: …") and adapts the studio at once.
  3. Read each card. Who the profile is for, then "Included:" (each section with the number of its pages included, and how many of the fourteen calculations of "Size and tune") and "Locked:". The Business profile does not estimate. The whole Simulations section is locked for it.
  4. Find it on the studio's home page. The line "View adapted to the … profile" (or "Full profile: every page"), with "Change". With a role, the "For you" block asks its five questions, and each opens the page that answers it. "Your pages" gives its first pages. In "To review first", the findings of its kinds come first, and in each section of the menu, its pages come first.
  5. Recognise a locked page. It keeps its place in the menu, a padlock in place of its icon. If you open it from the menu, a link, the address or the arrows, the studio shows "Page not included in your profile" in its place: the page asked for, why the profile leaves it out, then "Change the profile in the settings", "See the profiles of use" and "Back to the home". A calculation outside the profile does the same: "Calculation not included in your profile".
  6. Compare the profiles. "See the table of profiles", in the settings, or "Profiles of use" in the studio's help: a tick for what opens, a padlock for the rest, page by page and calculation by calculation. Then "Why these choices", and each role in detail: its questions, its pages, its calculations, the findings it sees first.

To check On the studio's home page, the profile line names the one you picked, and the menu shows a padlock on each page it leaves out.

Good to know Always included, whatever the profile: Home, Learn (paths and dictionary), Sources and sync, Settings, Diagnostics, Export, the help and privacy. You choose the profile in the settings only, and YALineo keeps it with your display preferences.

See the screenshot
Settings, step 4: "Full" by default, and the seven roles, each with what it includes (by section, and how many of the fourteen calculations) and what it locks.
Settings, step 4: "Full" by default, and the seven roles, each with what it includes (by section, and how many of the fourteen calculations) and what it locks.

Getting started · 2 min

The settings: find everything, choose your language

Contents

The settings page puts what to set before opening the studio on one side, and your preferences on the other. Each section folds, and a search finds a setting from one word.

Before you start

  • The extension installed.
  1. Open the settings. They open on their own after installing. After that, right-click the YALineo icon, then choose "Options" (Chrome, Edge) or "Manage Extension", "Options" tab (Firefox).
  2. Find the two families. "Connection and loading", to set before the studio: Workspace, Connection, Scope, Your profile, Data retention, Loading the map, Configured workspaces and Access for an admin. "Preferences", to change at any time without loading the map again: Language, Appearance, Map and studio, Databricks pages, Ctrl K shortcuts, Privacy and data, Help and about.
  3. Go straight to a section. The navigation on the left stays in sight as the page scrolls; a click brings the section under the header and marks it as the one you are in.
  4. Search for a setting. "Search the settings": type a word, with or without accents, in English or French (theme, language, chip, Ctrl K…). Only the sections that match stay; Escape clears the search.
  5. Choose your language. "Language": French, English, or the browser's language. It applies at once to these settings and to the open studio, to the panel the next time it opens, and to the chip and the badges on Databricks pages.
  6. Fold what you do not use. A click on a section's title folds it without hiding its sentence. It stays folded next time. "Fold all", under the navigation, folds the page at once. Loading the map does not fold, because its buttons are the next step.
  7. Open the studio. When a map of this workspace is already loaded, the settings say so (date, tables, links), and "Open YALineo Studio" stays beside "Save and load the map", even after the page is reloaded. "Open the studio", at the top right, leads there too.

To check After choosing "Français", the page's title becomes "Réglages", and the open studio switches to French.

Good to know YALineo keeps the language and the folded sections with your preferences, on this device, until "Erase everything". Nothing is sent.

See the screenshot
The settings: on the left, the search and the navigation in two families, "Connection and loading" then "Preferences"; on the right, the preferences, starting with the language.
The settings: on the left, the search and the navigation in two families, "Connection and loading" then "Preferences"; on the right, the preferences, starting with the language.

Getting started · 1 min

Pick your style

Contents

Eight styles: "YALineo", the default, and seven others, each in light and dark. A style changes the frame (band, surfaces, inks, corners, shadows), never the colors of the lines, series or states, nor the drawing of the map.

Before you start

  • The YALineo settings open.
  1. Open "Appearance". In the settings, "Preferences" family. The settings search also finds the word "style".
  2. Pick a style. Horizon (lifted cards), Prestige (night blue and champagne hairlines), Graphite (monochrome), Desert (sand by day, indigo by night), Quartz (neutral grays, in the spirit of SAP Fiori), Frosted glass and Deep glass (see-through panes). It applies at once to the settings, the open studio and the panel, without a reload.
  3. Keep your theme. You set the theme just above: light, dark or "Follow Databricks". It holds for every style, and the sun or moon button of the studio changes it too.
  4. On a modest computer, prefer a color style. The Glass styles only blur the menus and the layers that open for a moment. Tables, the palette, the search results, the cost filters and what lies on the map are opaque; cards and the side bar stay see-through, without blur. The page still rests on a gradient scene. Without a graphics card, in Firefox above all, scrolling a long table can then stutter (measured: one frame in five to eight comes late). If the browser cannot blur, or if your system asks for less transparency, the glass becomes a plain ground.

To check The band and the panes of the studio take the colors of the chosen style; the map keeps its lines and their colors.

Good to know YALineo keeps the style with your display preferences, on this device, until "Erase everything". At each version, each style is measured in light and dark on the settings, the panel, the screens of each section and the floating layers: every word at 4.5:1 at least.

See the 2 screenshots
Settings, Preferences, Appearance: the theme, then the style just below.
Settings, Preferences, Appearance: the theme, then the style just below.
The studio in Desert: the frame changes, the map and its colors stay the same.
The studio in Desert: the frame changes, the map and its colors stay the same.

Getting started · 3 min

The chip and the costs in Databricks

Contents

On every table you open in Databricks, the YALineo chip shows its line, its freshness, its neighbors, its reads and what its refresh costs, without leaving the page.

Before you start

  • The map loaded (green chip).
  1. Open a table in Catalog Explorer. The chip appears in the top bar, to the right of the search, in a free spot. In a crowded bar, it shrinks to the YALineo logo.
  2. Hover it. Its card opens: its line, its freshness, its direct neighbors upstream and downstream ("direct links", like Catalog Explorer), then "Along the whole chain", its reads and what its refresh costs.
  3. Open the map. "Open the map" opens the side panel on this table; the icon next to it opens the studio, the large window. Panel shortcut: Alt+Shift+L.
  4. Look at the small amounts. In the jobs list, Catalog Explorer and the workspace browser, each link to an asset of the loaded map (job, pipeline, SQL warehouse, table, view, dashboard, app…) has its badge. The badge gives what it cost over the period, "< $0.01" under a cent, or a short note with its reason ("$0 · view", "not attributed"…), framed with dashes. Only an asset absent from the map has none. In Catalog Explorer, the list of catalogs and a catalog's list of schemas have theirs too, which is the refresh of their tables, shown without "≈" when it is exact ("What each catalog and schema costs" tutorial).
  5. Pause YALineo if needed. The switch in the card stops the badges and the card on Databricks pages, without erasing anything. One click turns it back on.

To check The chip is green and its card names the open table.

If it does not work2 cases
The chip does not show.
Check in the settings, "Chip in Databricks", that it is not "Hidden". YALineo adds the chip to the open Databricks tabs, at install and at every update; if it is still missing, reload the tab.
The chip is red.
The last load failed. Hover it to read the reason, then open the panel to load again.

Good to know Its color shows where the map stands: orange while loading, green when ready, red if the last load failed. It can also sit in the bottom-right corner (settings, "Databricks pages").

See the 3 screenshots
The chip's card on a table: its line, its freshness, its direct neighbors then the whole chain, its reads and what its refresh costs.
The chip's card on a table: its line, its freshness, its direct neighbors then the whole chain, its reads and what its refresh costs.
In Jobs & Pipelines, what each job and pipeline cost over the period, next to its name.
In Jobs & Pipelines, what each job and pipeline cost over the period, next to its name.
In Catalog Explorer, the refresh cost of each table of the schema (≈: the table is not billed alone, its share comes from the published attribution rule).
In Catalog Explorer, the refresh cost of each table of the schema (≈: the table is not billed alone, its share comes from the published attribution rule).

Getting started · 4 min

Find your way in the studio

Contents

The studio is YALineo's large window. On the left, six sections: Home, Data and lineage, Costs, Simulations, To review and Learn. At the top, the workspace, the search and the tools.

Before you start

  • The map loaded.
  1. Open the studio. From the chip's card (the icon next to "Open the map"), from the side panel, or "Open YALineo Studio" in the settings once the map is loaded.
  2. Start from the home page. At the top, the profile line, with "Change" ("Pick your profile" tutorial) and, for a role, "For you" and its five questions. Then a search, and "Your day at a glance", which counts what calls for attention on the map read: the findings to review, the tables late in writing, the tables refreshed with no read observed, with the same figures as the badges. "See" opens the page that lists them. Next, "Get started", four first steps ticked as you take them ("Hide" removes the card), and four ways in that lead to the right page ("Find a piece of data", "Understand a dependency", "Explain a spend", "Prepare an estimate"). Last, "To review first", "Data read" (source, date, assets, links, state) and the "Start here" guide.
  3. Move from one section to another. Rest the mouse on an entry (or tab onto it). A bubble shows beside it, with its name and what the page is for. A click on a section opens its main page, or the last view opened in it; its chevron unfolds it. "Unfold all", at the top of the bar, opens them all, and the studio remembers which sections are open. Lower: "Sources and sync", "Export", "Diagnostics" and "Settings".
  4. Know where you are. The trail, under the top bar, names the section, then the page you came from, then the open page, and each step is a button. After a click in a page (a station, a line, a search result), that page's step opens it again as you left it: its followed table, its filters. The last step goes back to the top of the page. "Back" reopens a long page at the position where you left it. The YALineo logo, top left, goes back to the home page. On a long page, two arrows appear at the bottom right: one goes back to the top, the other down to the bottom.
  5. Find anything from the keyboard. Ctrl K (Cmd K on a Mac), or the key shown in the search field, opens the command palette. It shows your shortcuts first; then type the name of a table, a page, a "Size and tune" calculation or an action (copy the context, theme, language, read the workspace again, settings, export). The arrows choose, Enter opens, Escape closes.
  6. Right-click. On a table, a station, a row of a list, a calculation, an entry of the side bar or the page itself, a menu offers the actions for that element. The next tutorial goes through them, with the Ctrl K shortcuts.
  7. Switch workspace or search. The workspace card, top left, gives its cloud and its state (green dot: map loaded; orange: reading) and opens the list of sources. The "Search a table, view or report" field opens any asset: first the names that start with what you type ("dim" gives dim_promotion, dim_staff, dim_store), then the names that hold it. At the bottom of the list, "N results here · see them all in the catalog" opens the whole list.
  8. Use the top bar. The "Back" and "Forward" arrows retrace your path, where each view, table or schema opened counts as a step (Alt+← on the keyboard). Then: reload, the light or dark theme, FR / EN applied at once, the help (tutorials, how it is calculated, feedback, support, diagnostics, profiles of use) and the settings.
  9. Read the status line. Under the bar: the map's date, the link history, the activity period (a click on a period opens the map's calendar), the catalog scope. "Copy the context" copies the name of the page, the workspace, the date of the data and the periods, then the text of what the page shows, for a ticket or a question to a colleague. At the bottom of the window: privacy, tutorials, support and "Erase now".
  10. Fold the side bar. "Fold the bar", at the bottom, folds it to its icons, and the pages take the whole width. Rest the mouse on the icons (or tab onto them) to open it over the page. "Keep the bar open" pins it. The studio remembers the choice.

To check The side bar marks the section of the open page and highlights its view.

Good to know The studio uses your system's font, like the Databricks workspace (Segoe UI on Windows, SF Pro on a Mac). It follows the size of the window and the browser zoom, and charts redraw when their frame changes. The right panel of the map, the catalog and the dependencies folds to a rail and resizes with the mouse or the keyboard.

See the 3 screenshots
The home of the studio: "Your day at a glance" and "Get started", then four ways in, the first findings and the data read; on the left, the six sections.
The home of the studio: "Your day at a glance" and "Get started", then four ways in, the first findings and the data read; on the left, the six sections.
The command palette (Ctrl K): a table, a page or an action, from the keyboard.
The command palette (Ctrl K): a table, a page or an action, from the keyboard.
The side bar folded to its icons: the map takes the whole width of the window.
The side bar folded to its icons: the map takes the whole width of the window.

Getting started · 3 min

Right click, Ctrl K and your shortcuts

Contents

Everywhere in the studio and in the side panel, a right click opens a menu of actions for the element under the mouse. Ctrl K finds every page and every calculation, and shows first the shortcuts you chose.

Before you start

  • The map loaded, for the menus of a table or a station.
  1. Right-click a table. A table or a view, wherever it is: a row of the catalog or of a list, a station of "Dependencies of an asset", a sheet. The menu, titled with its name, offers "Open the sheet", "See on the map, with its lineage", "See its dependencies", "Analyze the impact of a change", "See its use and its cost", "Open in Databricks", "Copy the full name", "Copy a SELECT query (100 rows)" and "Watch this table", which shows its delay on the home page.
  2. On a station of the map. A schema: "See its N items" opens its card on the right, "Isolate on the map" puts it in focus, "See the dependencies of this schema" opens its tables in "Dependencies of an asset", and "Copy the full name".
  3. On a calculation, a side bar entry, a link, words. A "Size and tune" calculation: open it, open it with its example filled in, read its course ("Learn more") or add it to the Ctrl K shortcuts. A side bar entry: open it, open it in a new tab, read the page's course when it has one, add it to the shortcuts. A link: open it in a new tab or copy its address. Selected words: copy them, or look them up in the map and the pages.
  4. Anywhere else on the page. Back to the previous page or forward to the next, open the palette, read the page's course when it has one, copy the link of the page or its context, open the export center, add the page to the Ctrl K shortcuts, switch the theme, open the settings.
  5. From the keyboard. The menu key, or Shift F10, opens the menu of the element that has the focus. The arrows choose, Home and End go to the first and last choice, Enter confirms, Escape or Tab closes the menu.
  6. Keep the browser's menu when needed. Shift and a right click open the browser's menu, as the foot of the menu reminds you. In a typing field, the right click always opens the browser's.
  7. In the side panel. On a table: "See its route here", "Open its sheet in the studio", "Open in Databricks", copy its full name or a SELECT query. Links and selected words have their menu too.
  8. Choose your Ctrl K shortcuts. Settings, "Ctrl K shortcuts" card: what the palette shows first, in order. Drag a row to move it, or use its arrows from the keyboard. The cross removes it. With the list and "Add", you add a page, a "Size and tune" calculation or an action. You can keep up to 12 shortcuts. "Back to the 8 default shortcuts" brings back the platform map, the overview, the cost map, "Size and tune", "Cost a project", the findings, the catalog and the export center. YALineo saves every change at once.
  9. Or add them with a right click. "Add to the Ctrl K shortcuts", on a calculation or a side bar entry, and "Add this page to the Ctrl K shortcuts", on the page. The shortcut goes to the end of the list, and you set its order in the settings.
  10. Open the palette. Ctrl K (Cmd K on a Mac) opens the palette on your shortcuts, with nothing typed. Type a few letters to find a table, a page, a calculation or an action: "Photon" finds the Photon break-even calculation, and Enter opens it.

To check A right click on a row of the catalog opens the table's menu, titled with its name; Escape closes it. After "Add to the Ctrl K shortcuts", the palette opened with nothing typed shows that shortcut under "Your shortcuts".

Good to know The menus write nothing in Databricks. "Copy a SELECT query" puts the query on the clipboard, and you run it yourself in the SQL editor. YALineo keeps the shortcuts with your display preferences, on this device.

See the 3 screenshots
A right click on a table of the catalog: its sheet, the map, its dependencies, the impact, its cost, Databricks, its name or a SELECT query to copy, and watching it.
A right click on a table of the catalog: its sheet, the map, its dependencies, the impact, its cost, Databricks, its name or a SELECT query to copy, and watching it.
A right click on a station of the map: its items, the focus, the schema's dependencies, its name.
A right click on a station of the map: its items, the focus, the schema's dependencies, its name.
Settings, "Ctrl K shortcuts": what the palette shows first, in order, to drag or to move with the arrows.
Settings, "Ctrl K shortcuts": what the palette shows first, in order, to drag or to move with the arrows.

Getting started · 3 min

To review: findings and your follow-up

Contents

The "To review" section lists the situations worth a check. YALineo recomputes them at each read, with no new query. Each gives the asset, the possible effect, the evidence and a next action.

Before you start

  • The map loaded; for cost findings, the FinOps side.
  1. Open "Findings". "To review" section; its badge gives the number of findings. The home page shows the first ones, and "All findings" opens the list.
  2. Filter by kind. "Freshness", "Failures", "Usage", "Sensitive data", "Documentation", "Architecture"… each with its count; priority findings come first.
  3. Read the evidence. The date, the threshold, the measured values and the system table they come from. An amount states its nature: observed, estimated (discount taken off) or potential (a saving to confirm). Without readable billing, cost findings are "not assessed", never 0.
  4. Act. The finding's button opens the right page with the right context, for example "Analyze the impact of the delay".
  5. Follow it up. "My follow-up": "To review", "Checked by me" or "Ignored", with a note. A finding comes back to "To review" when its evidence changes. The list at the top right filters by your follow-up.

To check The sentence under the kinds says how many findings are to review, and how many of them are priority.

Good to know The follow-up is personal. YALineo keeps it on this device, workspace by workspace, under the map's retention rule. It is neither a team sign-off nor a technical fix. The health check is just below, under the name "Health check".

See the screenshot
The findings: sorted by kind, each with the asset, the possible effect, the evidence, an action and your follow-up.
The findings: sorted by kind, each with the asset, the possible effect, the evidence, an action and your follow-up.

Getting started · 3 min

Learn: a path per profile and the dictionary

Contents

The "Learn" section, the sixth of the side bar, teaches Databricks and the platform: a path of modules per profile, each ending with a quiz, and the dictionary of the platform's notions. It is open in every profile.

Before you start

  • The studio open. The courses are bundled in the extension. Reading them sends no query to your workspace.
  1. Open "Learn". The section, in the side bar, under "To review": "Paths" and "Dictionary".
  2. Choose your path. "Your path": "The whole course", or one of the seven profiles. The page starts from the profile chosen in the settings; with "Full", it is the whole course. A bar says how many modules are done, and the "Start" or "Continue" button opens the first module left.
  3. Go through the modules. Each gives its title, what it brings, its length, its level ("first steps", "the essentials" or "in depth") and, once done, your quiz score. "Start" opens it; a module done opens again with "Read again".
  4. Read a module. Its parts, its words and its official sources, with the date they were read. "Previous module" and "Next module" follow the order of the path; "All the modules" goes back to the list.
  5. Take the quiz. "Check what you remember", then "Check my answers": each answer says "Right." or "Not quite.", and why. If all are right, the module counts as done; "Mark as done" does it by hand, "Mark as to review" undoes it. YALineo keeps the best score.
  6. Read the word that follows. When all is right, congratulations show, with "Next module". Otherwise, an encouragement says what to read again, with "Try again". YALineo keeps only your best score.
  7. Ask for a training session. At the bottom of the path page, "Get support": getting started with YALineo, learning Databricks, or a team trained on its own path. "Ask for a session" opens your mail app to contact@yalineo.com with a message to complete (topic, number of people, availability) and your progress; you decide whether to send it.
  8. Look up a notion. "Dictionary": type a word (DBU, Unity Catalog, Photon…) or choose a theme (Platform, Governance, Data, Compute, Flows and jobs, Analytics and BI, AI, Costs, Security, YALineo). A click on a notion opens it: its definition, an example, where to see it in YALineo, the notions next to it and its official pages.

To check After a quiz all right, the bar of "Your path" counts one more module done, and keeps it when you come back to the studio.

Good to know YALineo keeps your progress (modules done, best scores) on this device with your preferences. Nothing is sent. Links to the official pages open in a new tab. For the calculations, each tool's course is behind "Learn more" ("Size and tune" tutorial).

See the 5 screenshots
"Learn", the path of the Data developer profile: the progress, "Start" on the first module left, then each module with its length and its level.
"Learn", the path of the Data developer profile: the progress, "Start" on the first module left, then each module with its length and its level.
A module's quiz, checked: each answer says whether it is right and why.
A module's quiz, checked: each answer says whether it is right and why.
Once checked, a word by the score: congratulations when all is right, encouragement otherwise, with "Try again" and "Next module".
Once checked, a word by the score: congratulations when all is right, encouragement otherwise, with "Try again" and "Next module".
"Get support": ask for a training session with the YALineo team, for you or your team.
"Get support": ask for a training session with the YALineo team, for you or your team.
The platform dictionary, searched for "DBU": each notion with its short definition and its theme.
The platform dictionary, searched for "DBU": each notion with its short definition and its theme.

Data and lineage · 7 min

Read the platform map

Contents

Each data domain is a colored line, each schema a station, from sources to dashboards. The map reads from top to bottom, the way data flows.

Before you start

  • The map loaded.
  1. Open "Platform map". A click on the studio's "Data and lineage" section opens it; it is also in the side panel. At the top, the figures, each on one line: catalogs, schemas, tables, links, consumers, share of tables on time and late. They count what is shown. Under a filter, the workspace total shows under each figure, with "Back to the whole workspace".
  2. Give the drawing room. The filters fold under "Filters", which counts the ones that are on. Once folded, a line names them. The studio keeps this choice. If you turn the wheel down over the drawing, it grows to the whole height of the window, then zooms. When you come back to the top of the page, it fits again.
  3. Read the map another way. "Read by", above the drawing: "Domains" (the lines, as usual), "Costs" (the cost map, where a station's halo shows its share of the map's cost), "Freshness" (the halo counts the tables with no write for more than 8 days) or "Use" (its reads over the period read, and its readers' events). A line at the bottom of the drawing says what the halo measures. The figures are the ones the station shows on hover. The studio keeps the choice, and the halos never go into the downloaded picture or page.
  4. Follow a line. When you hover it, it thickens. Click it, on the map or in the "Lines" list on the right, to open its card (tables, stations, delays, reads, cost). "Focus on this line" keeps it alone, with every station that holds its tables, even when another line feeds it.
  5. Read why a line shows nothing. If the filters keep the line in focus out of the drawing, the map says so: how many tables the map holds on this line, and what keeps them out (the catalogs or schemas chosen, the period of the links, the kinds of links removed, the followed table). "Show everything, keeping the focus on this line" removes these filters; "Leave the focus" brings the whole map back.
  6. Select a station. A click keeps it selected, its lineage lit, and its card opens on the right. A second click (or Esc) lets it go. A right click opens its menu ("Right click, Ctrl K and your shortcuts" tutorial). The next tutorial shows how to isolate it.
  7. Move the drawing. Grab the background to pan the map, or a station to put it elsewhere. Its links follow, and it slides to the nearest free place in its band. This works with a mouse, a pen or touch, and with Shift + arrows on the keyboard. A click without a drag moves nothing and keeps its usual effect. "Reset the layout" brings back the original drawing. Exports keep the positions.
  8. Hover any element for its figures. A tooltip opens on every station and every link. A schema: its tables, layer, lines, freshness, reads over the period, last write, owners. "External sources": what they are (cloud storage files and folders, outside Unity Catalog), how many paths, their last use and a few names. Dashboards, queries and notebooks: how many, their reads and examples. A link: its line, both stations and its links.
  9. Switch the drawing. Among the tools at the bottom right of the map, "Show as a graph, like Databricks" draws the same lineage as boxes and arrows; "Show as a metro map" goes back. The studio keeps the choice, in "Dependencies of an asset" and the side panel too.
  10. Open the legend. "Legend", next to "Full screen", shows each symbol of both drawings and what it means. Esc closes it.
  11. Change the bands. "Bands": "Medallion layers" (source, bronze, silver, gold, platinum, consumption) or "Catalogs", in the order data flows.
  12. Filter the links. The "Transformations", "Loads", "Exports", "Consumers" boxes choose the kinds of links. "Show only the links seen" (24 h, 7 days… or the whole history) sits in the calendar above the map, with the period (next tutorial). The "Scope" bar gathers the filters set, each removable with one click, with the count of links and their CSV.
  13. Use the drawing tools. Bottom right: "Follow the route" animates the path, "Continuous flow" makes data run, then "Zoom out", "Zoom in" and "Fit". "Full screen" spreads the map and its panel over the whole window.
  14. Keep the overview. When the drawing is zoomed in or moved, a small map of the whole drawing appears in its bottom left corner, without covering the scope or the figures. A frame on it marks the part seen, and a click or a drag in it moves the view there. When the map shows only a part of the platform (a followed table, a station in focus, chosen catalogs, one line), the overview shows the whole platform and marks the stations drawn. Its cross removes it ("Undo" in the message). The "Overview of the map" box, in the settings, "Map and studio" section, removes it or brings it back. It is in "Dependencies of an asset" too.
  15. Go down into a station. Its card lists its tables, counted per line and freshness (each count filters the list), with a search and a sort. A click on a table shows only its lineage; the arrow opens it in "Dependencies of an asset".

To check Hovering a line thickens it; clicking a station rings it and opens its card on the right; hovering "External sources" explains what they are.

If it does not work1 case
"No map loaded".
Load the map from the settings or the panel ("Load the map"), or open the fictional example in "Sources and sync".

Good to know A faded link was not used during the activity period. The pipeline may have stopped. Hover it to see its last date. The privacy band at the bottom of the studio folds to one line after a minute, and "Details" opens it again.

See the 6 screenshots
The platform map in the studio, with the figures of what is shown: data lines, stations per schema, medallion bands, consumers at the bottom. A tooltip gives the figures of the station hovered.
The platform map in the studio, with the figures of what is shown: data lines, stations per schema, medallion bands, consumers at the bottom. A tooltip gives the figures of the station hovered.
"Read by: Freshness": a station's halo counts its tables late in writing.
"Read by: Freshness": a station's halo counts its tables late in writing.
The "Other" line in focus, the semantic catalog chosen: the map says how many tables the line holds and what keeps them out of the drawing.
The "Other" line in focus, the semantic catalog chosen: the map says how many tables the line holds and what keeps them out of the drawing.
Zoomed in on a part: the overview, at the bottom left, shows the whole drawing and the part seen.
Zoomed in on a part: the overview, at the bottom left, shows the whole drawing and the part seen.
The same lineage as a graph, like Databricks: a box per schema or reader, an arrow per pair of stations.
The same lineage as a graph, like Databricks: a box per schema or reader, an arrow per pair of stations.
The same map in the side panel.
The same map in the side panel.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Data and lineage · 2 min

Change the period of the map

Contents

You first set the period of the map in the settings. The calendar above the map changes it without going back there: how many days of lineage to load, and over how many days to read activity.

Before you start

  • The map loaded from a connected workspace. On the fictional example, the calendar opens but reloads nothing.
  1. Open the calendar. The button with the calendar icon, above the map, after "Bands", shows the current period, for example "365 d of links · 90 d of activity". The periods in the status line, under the studio header, open it too.
  2. Choose the period of links. "Links": 7, 30, 90, 180 or 365 days of lineage loaded. A longer period loads more links and takes longer, and the calendar says so.
  3. Choose the activity period. "Activity": 7, 30, 90, 180 or 365 days for freshness, reads and delays.
  4. Filter the links seen, without reloading. "Show only the links seen" keeps, among the links already loaded, those used recently (24 h, 7 days…). It is a filter of the map, and YALineo reads nothing again.
  5. Click "Reload the map". The map reloads from the system tables, with the same read-only queries, and keeps your filters. YALineo saves the new period for this workspace, and the panel and the chip use it too.
  6. In analyst mode. Lineage comes from the API, with no date limit. You can change only the activity period.

To check After loading, the calendar button and the status line show the new period.

If it does not work1 case
"Reload the map" stays grayed out.
Nothing changed, a load is running, or it is the fictional example, which does not reload.

Good to know This calendar sets lineage and activity. You choose the period of the costs separately, in the "Costs" section ("Choose the period and read the changes" tutorial). "Cancel" or Esc closes the calendar without touching the selected station.

See the screenshot
The calendar open above the map: the days of links, the days of activity, the filter of links seen, then "Reload the map".
The calendar open above the map: the days of links, the days of activity, the filter of links seen, then "Reload the map".

Data and lineage · 3 min

Isolate a station (focus)

Contents

The focus keeps on the map only the chosen station and the lineage of its tables. Everything else leaves the drawing.

Before you start

  • The map open (previous tutorial).
  1. Click a station. It stays selected and its card opens on the right.
  2. Click "Isolate (focus)" in its card. The map redraws: the station, its tables, and the links that go through them. A "Focus: station name" pill shows above the map.
  3. Choose the lineage shown. "Direct links" (default): what feeds its tables directly and what they feed directly. "The whole lineage": step by step, up to the sources and down to the reports, keeping only the paths that go through its tables.
  4. Explore its tables. The card lists the station's tables: counts per line and freshness, a search, a sort (most read, by name, most costly…) and the ↓ ↑ arrow to reverse the order.
  5. Go further. "Explore this schema" opens its tables in "Dependencies of an asset", "Analyze" in "Cost analysis", "See the N tables" in the catalog.
  6. Leave the focus. Use "Leave the focus" in the card, a click on the pill, or Esc. The whole map comes back, and the station stays selected.

To check The map shows only a few stations and the "Focus" pill is visible. The "Catalog" choice is grayed out, because a station's lineage crosses catalogs.

Good to know A station that feeds no dashboard directly shows none with "Direct links". Switch to "The whole lineage" to see the reports it feeds further down.

See the screenshot
Focus on lake.gold, direct links: only its tables, what feeds them directly and what they feed.
Focus on lake.gold, direct links: only its tables, what feeds them directly and what they feed.

Data and lineage · 4 min

Find a catalog, a schema or follow a table

Contents

At the top of the map, a field finds a catalog, a schema or a table. If you choose a catalog or a schema, it becomes the map shown. If you choose a table, the map follows it and shows only its lineage, across catalogs.

Before you start

  • The map open in the studio.
  1. Type in "Catalog, schema or table…". Matching catalogs come first ("Catalog · 48 tables"), then schemas ("Schema · 12 tables"), then tables, views and reports, with their full name. A name typed in full always comes first.
  2. Choose a catalog. Arrows and Enter, or a click. The map shows only that catalog and the stations directly linked to it. The "Catalog" list takes it up.
  3. Choose a schema in the "Schema" list. Once a catalog is chosen, this list gives its schemas with their number of tables. The map keeps the schema's station and the stations directly linked to it, and the "Table" list its tables only. Each list takes several values, with a search and "Clear all".
  4. Choose a table in the "Table" list. Once a catalog is chosen, this list gives its tables, grouped by schema (lake.gold, lake.silver…), or those of the schema chosen. Without a catalog, it stays inactive. Search for the table in the field instead.
  5. Read its lineage. A "Table: name" pill shows. On the right: its direct upstream and downstream tables and its direct uses, then "Along the whole chain", and the lists "Upstream: what feeds it" and "Downstream: what it feeds", nearest first ("direct link", "2 levels away"…).
  6. Follow step by step. A click on a table of these lists follows it in turn. The "Back" arrow returns to the previous one.
  7. Set the depth. "Lineage shown": "Direct links" or "The whole lineage". "See it table by table (Explore)" opens the same lineage table by table, "Trend" its figures over time.
  8. Remove the filter. A click on the pill, "Remove the filter", or Esc. In full screen, Esc closes the selection first, then the filter, then the full screen.

To check The studio's address holds table=full.name, so a copied link opens the same view again.

If it does not work1 case
"No lineage link for this table".
It may be fed outside Unity Catalog (external connector, JDBC, untracked files) or by another workspace, or nobody read or wrote it over the period. If the filter of links seen (in the map's calendar) or the link kinds hide links, "Show every link" brings them back.

Good to know While a table is followed, the "Catalog" and "Schema" lists stay inactive, because its lineage crosses catalogs. Your choice comes back when you remove the filter.

See the 3 screenshots
The map's field finds a catalog (first), a schema or a table.
The map's field finds a catalog (first), a schema or a table.
The lake catalog, gold schema: its station and the stations directly linked to it; the "Table" list keeps its tables only.
The lake catalog, gold schema: its station and the stations directly linked to it; the "Table" list keeps its tables only.
A table followed: its lineage only, its upstream and downstream listed on the right, nearest first.
A table followed: its lineage only, its upstream and downstream listed on the right, nearest first.

Data and lineage · 4 min

The dependencies of an asset

Contents

The "Dependencies of an asset" view (the former Explore) draws each table of the route, from source files to dashboards, with the card of the chosen table. Its figures follow Databricks: the direct neighbors first, then the whole chain, named as such.

Before you start

  • The map loaded.
  1. Open the table. Search at the top of the studio, the arrow of a table in a list, the "Dependencies" button of its sheet, or the chip in Databricks then "Open the map".
  2. Choose the direction and the depth. "Upstream", "Both" or "Downstream", then "Depth": 1, 2, 3 or "All", in one click.
  3. Choose the kinds of assets. On the "Show" line, under the filters, one chip per kind with its count: "Tables and views", "Jobs and pipelines", "Notebooks", "Dashboards and queries", "External paths". A click hides or shows the kind; "Show all" brings them all back.
  4. Read the figures. On the right, "N upstream · N downstream · direct links", like the Lineage tab of Catalog Explorer, then "Along the whole chain: …". A link that only exists through a view, or around a table deleted or unreadable, is not direct. YALineo marks such a bypass "bridged", with its reason.
  5. Read the card on the right. Description, costs (refresh, per read, with all upstream, and the job that pays, with "Why?" next to a missing cost), type, zone, line, owner, freshness. "Open the sheet" opens its full page ("The sheet of an asset" tutorial).
  6. Change the view. "Graph" or "List" (keyboard friendly), level "Tables" or "Columns" (read on demand), "Path" to show the path to another asset. The "Scope" bar shows the filters set, and "Export (CSV)" writes the links and their evidence.
  7. Re-center. Click another table of the drawing, then "Center here". "Filter on a table" keeps only the lineage of one table of the drawing. "Follow" adds it to your followed tables. A badge warns you on refresh if it falls behind.
  8. In the side panel. The panel has the "Card", "Route" and "Impact analysis" tabs. "Copy the pre-flight" gives text ready for a ticket, and "Analyze columns" goes down to the column.

To check The chosen table is circled at the center of the drawing; on the right, its direct neighbors then the whole chain.

If it does not work2 cases
"No downstream link in this view."
Click "Why?" next to it. YALineo says what the view hides and offers to check over 365 days ("Why no link, why this cost" tutorial).
"Large view: N assets."
"Depth 1" goes back to the direct neighbors, "See as a list" browses the relations without a drawing; "Hide this notice" closes it.

Good to know The "Coverage and reliability of this view" band, under the drawing, says how many links are shown out of those in scope, how many are old, and what was bridged. "Full screen" gives the drawing the whole window.

See the 2 screenshots
"Dependencies of an asset": the route of fact_inventory, its direct neighbors counted as Catalog Explorer counts them, the whole chain apart, and its card on the right.
"Dependencies of an asset": the route of fact_inventory, its direct neighbors counted as Catalog Explorer counts them, the whole chain apart, and its card on the right.
The route in the side panel.
The route in the side panel.

Data and lineage · 3 min

The sheet of an asset

Contents

Every asset has its sheet: table, view, job, pipeline, SQL warehouse, serverless notebook, dashboard or query. It shows only what YALineo read. When Databricks does not give a value, the sheet says so and never guesses it.

Before you start

  • The map loaded; for costs, the FinOps side.
  1. Open the sheet. From the search, the home page, the catalog, the map, the dependencies, impact analysis, the cost lists, the glossary or a finding. It takes its place in the side bar ("Sheet: name"). Back, forward and reload lead back to it.
  2. Read the header. Type, domain, layer, owner, freshness, and the actions: "Impact analysis", "Dependencies", "Open in Databricks", "Export the sheet" (CSV), "Copy the link", "Follow".
  3. Go through the tabs. "Overview", "Columns", "Lineage", "Usage and costs", "History".
  4. Read the reads. "Freshness and reads" says how many reads come from a named object (job, notebook, dashboard, query), how many serve to write another table, how many have no Databricks object tied to them, and why the "Popularity" of Catalog Explorer counts differently. "Read the reads over 30 and 90 days" reads them on demand.
  5. Read the costs. "Cost over the FinOps period": the refresh, the reads on SQL warehouses and the total, each amount with its nature, then "How it is split". "Why?" next to an amount gives its reason.
  6. A view has its own sheet. A view stores no data: no last write, no write pace, no delay. "Defined on" gives its direct sources, like Catalog Explorer.

To check The studio's address contains view=asset, so "Copy the link" opens the same sheet again.

Good to know The sheet shows the whole asset. The cost filters do not apply to it, but the workspace discount does. Without readable billing, it says "Cost unavailable" and why, never 0.

See the screenshot
The sheet of fact_inventory: its header and actions, its tabs, its description, its freshness and its reads, with where they come from.
The sheet of fact_inventory: its header and actions, its tabs, its description, its freshness and its reads, with where they come from.

Data and lineage · 3 min

Why no link, why this cost

Contents

When a table has no upstream or no downstream, or a cost is zero, missing or partial, YALineo says why, with the values read, and offers the action that changes the result. No reason creates a figure.

Before you start

  • The map loaded; for costs, the FinOps side.
  1. Spot "Why?". Next to a missing link (dependencies, sheet, map, findings) and next to a cost that is zero, missing or partial (sheet, panels, cost lists, the chip's card in Databricks). Under a table's lineage figures, when the map reads less than a year of links, YALineo also offers "Check over 365 days", because Catalog Explorer shows the last year by default.
  2. For a missing link, read what the view hides. YALineo gives, for this map and with its figures, the link window, the chosen catalogs and the hidden tables gone since.
  3. Click "Check over 365 days". One read-only query, for this table only, counts the links the map does not show. It sorts them by reason, with their number and the date of the latest: older than the link window, towards a catalog this workspace does not see or that you did not choose, towards a table gone since. YALineo reads nothing before your click.
  4. Read what the system tables do not trace. "What the system tables do not trace": external lineage, retention, renames, each with its Databricks page.
  5. For a cost, read the measured reason. Billing not readable and the missing permission, a view, no writer over the period, a shared job with the table's share, the other tables' and the rest (to the cent), reads not counted, list price with no discount.
  6. Act. Each case offers the button that changes the result. For a link: "Widen the link window", "Choose the catalogs", "Show the whole metastore", "Show both directions", "Show all", "Reload the map" or "Open in Databricks". For a cost: "Widen to N days", "Enter a discount", "Copy the SQL for an admin", "Remove the spend filters", or the workload that pays, opened in one click.

To check After "Check over 365 days", the "Checked on …" line names the query run (yalineo:why-links) and the system tables read.

Good to know A cost per read under a cent shows as "less than $0.01", and the "Why?" gives its exact value. The list of reasons and their rules: method, section 6.7.

See the 2 screenshots
"Why no downstream link?" for lake.reporting.orders_history: what the view hides, then the links checked over 365 days, sorted by reason.
"Why no downstream link?" for lake.reporting.orders_history: what the view hides, then the links checked over 365 days, sorted by reason.
"Why?" next to a refresh: the job that pays, the table's share and that of the other tables, to the cent.
"Why?" next to a refresh: the job that pays, the table's share and that of the other tables, to the cent.

Data and lineage · 5 min

Impact analysis: before changing a table or a job

Contents

Impact analysis shows what a change touches and how badly, who to tell with a message ready, and what to do in order.

Before you start

  • The map loaded, with activity (reads) to weigh severity.
  1. Open "Impact analysis". "Data and lineage" section (or the tab of the same name in the panel, or the button of a table's sheet). At the top, search the table, the view or the job, or browse the catalog, the schema and the table.
  2. Choose the scenario. "Change (schema, logic)", "Drop or rename the table", "Late refresh", or "Job stopped" for a job. A sentence says what the scenario breaks and when.
  3. Read what was read. The banner gives the object studied, the change considered, the scope, the period, the catalogs, the kinds of assets, the depth, the source, the freshness and the limits of the load, and recalls that the graph's filters do not apply.
  4. Read the figures that decide. Assets touched, critical, high, dashboards and queries, owners to notify, and what refreshing the touched tables costs each month. Then the split per downstream level and per catalog.
  5. Go through the list. Most severe first, with the reason for each severity (gold or platinum, number of reads, certified, sensitive data, directly downstream). Filter by severity, type or name.
  6. Tell the owners. Per owner, "Copy the message" prepares an e-mail or a Teams message with their touched assets; "Copy every message" takes them all.
  7. Follow "To do, in order". The steps of the scenario: tell, point direct readers elsewhere, keep a copy (DEEP CLONE), test downstream. "Copy the list".
  8. For a column. Pick it in "Column", or click "Analyze columns". YALineo then reads the column lineage on demand. Each asset gets one answer, "Relation observed", "Table dependency to check" or "Coverage too thin", and each change (rename, change the type, drop) has its plan.
  9. Keep a record. "Copy the report" and two CSV exports, the view shown or all the data available, each with its scope, for the ticket or the change request.

To check The figures at the top change when you change the scenario, because severity depends on what you plan to do.

Good to know Severity follows a published points rule. Nothing is guessed, and each level gives its reasons. Column impact reads column lineage on demand.

See the 2 screenshots
Impact analysis of a silver table, "Drop or rename the table" scenario: what was read, the figures that decide, then each touched asset with its reason.
Impact analysis of a silver table, "Drop or rename the table" scenario: what was read, the figures that decide, then each touched asset with its reason.
Impact analysis in the panel.
Impact analysis in the panel.

Data and lineage · 3 min

The health check

Contents

The health check runs eight checks on the loaded map, with no new query, and ranks what deserves your attention, from the most urgent to the most informative.

Before you start

  • The map loaded.
  1. Open "Health check". "To review" section. Its checks also show, asset by asset, in "Findings" ("To review: findings and your follow-up" tutorial).
  2. Read the checks. "Late but still read", "Reports wired to bronze or silver", "Backward flows", "Unlabeled sensitive data downstream", "No owner", "No description", "Not read in the period", "Off the network".
  3. Sort by severity. "Act", "Review", "For info". Each check says why it matters.
  4. See the tables. "Show tables" opens the data catalog filtered on that check.
  5. Add your rules. The architecture rules are your team's conventions, written in a shared JSON file ("Start from the example" to begin). For example: no dashboard reads a bronze or silver table, no data flows back from gold to bronze, every gold table has an owner, a description and tags, is written at least every 24 hours, follows a naming convention. "What are the rules for?" recalls it on the page. A rule not met becomes a finding, with the tables at fault, in the health check, in "To review" and in the pre-flight of pull requests.

To check The summary "N of M tables without findings" shows at the top of the check.

Good to know "Not read in the period" flags candidates for a review, never for automatic deletion, because a read may come from a tool outside Databricks.

See the screenshot
The health check: each check with its severity and the number of tables involved.
The health check: each check with its severity and the number of tables involved.

Data and lineage · 4 min

The data catalog: search, filter, read the sheet

Contents

The catalog shows every table of the map in one table, with one search across their names, columns, descriptions, owners, tags, domains and glossary terms. A click opens the table's sheet beside the list.

Before you start

  • The map loaded.
  1. Open "Catalog". "Data and lineage" section. On the left, "Browse": all catalogs, then each catalog and its schemas, with their number of tables.
  2. Search. Type a word in the "Search: name, column, description, owner, tag, domain or glossary term" field. Under each result, a line says where the word was found ("Column", "Description", "Term"…), with the word highlighted.
  3. Choose where to search. The "All", "Name", "Column", "Description", "Owner", "Tag", "Domain" and "Term" chips narrow the search to one place. Each gives its number of results.
  4. Filter. The "Zone", "Line", "Freshness", "Type", "Owner", "Description" and "Usage" lists combine with the search. When a sheet is open, they fold under "Filters".
  5. Sort and adjust. A click on a header sorts the column, a second one reverses it. Drag the right edge of a header to widen or narrow its column ("Sort and adjust the tables" tutorial).
  6. Open a table's sheet. A click on its row opens the sheet: its definition, the glossary terms tied to it, its columns with their type and description, its owner, tags, freshness, upstream and downstream, its cost (refresh, reads, and its schema's, exact or split), its documentation and "Open in Catalog Explorer".
  7. Read the columns. Until they are read, "Read the columns" loads them with one read-only query on system.information_schema.columns. They then stay with the map.
  8. Link a documentation page. Set a doc_url, documentation or wiki tag with an https address on the table, and the sheet shows the link. "See the lineage" opens the table in "Dependencies of an asset".

To check The number of tables above the table follows the search and the filters, and each chip gives its own count.

Good to know The health check opens this same catalog, filtered on the tables of one check. The terms come from the business glossary (next tutorial).

See the 2 screenshots
The data catalog: browse by catalog and schema, filter, sort with a click on a header.
The data catalog: browse by catalog and schema, filter, sort with a click on a header.
Searching "stock": the reason for each result, the word highlighted, and the sheet of the table open beside the list.
Searching "stock": the reason for each result, the word highlighted, and the sheet of the table open beside the list.

Data and lineage · 6 min

The business glossary: extract, import, validate, publish

Contents

The words of the business (stock on hand, average basket…), each with its definition and the tables and columns that carry it. YALineo proposes the terms from what Unity Catalog already holds, and you validate them. Nothing is written to Unity Catalog until the owner runs the script.

Before you start

  • The map loaded.
  • To propose column terms too: "Read the columns" (one read-only query on system.information_schema.columns).
  • To publish: the owner of a schema where the glossary table can be created.
  1. Open "Glossary". "Data and lineage" section. At the top: terms proposed or validated, those to validate, the tables linked. Below, "Fill the glossary" and its three ways to fill it. Once the glossary is filled, fold this section with a click on its title to leave all the room to the terms. It stays folded next time.
  2. "Extract from Unity Catalog", no AI. "Extract now" reads the map already loaded again: table comments (except those of the bronze zone), commented columns, column names found in several tables, metric views, business tags (business_term, domain…) and personal data (pii). Terms arrive as "Proposed", the most used first. "See the details" says what is read.
  3. "Import Excel or CSV". "Choose a file": the first sheet of an .xlsx workbook, or a CSV. YALineo keeps every column of the file under its own name, in your team's words (a code, a subdomain, a definition in English…). The columns YALineo recognizes by their header, in French or English, even glued (TermFR, DefinitionEN), also serve the links, the filters and the publication: term, definition, synonyms (several columns possible), domain, owner, table, columns. The page's language picks the term when the file has two, and an identifier column is never taken for the term. When two columns name the same field (a French and an English definition), the one half as full as the other never takes it. When they are about as full, the page's language picks. Change the mapping if needed ("Ignore" leaves a column out), tick or not "The first row names the columns", read the preview, then click "Import N terms". YALineo flags a file too heavy to be kept on the device before the import. The browser reads the file, which is sent nowhere, and a formula in a cell stays text.
  4. "Generate with the AI of Databricks", for tables without a comment. Click "Prepare the statement", then enter your business sector. YALineo writes an ai_query statement (one row per term, the answer in an imposed JSON format) with the databricks-meta-llama-3-3-70b-instruct model by default, which you can change. Before running, the estimate gives the number of calls and the input and output tokens. It also gives an amount if you enter the model's prices per million tokens (input and output), from the Databricks pricing page.
  5. Run it yourself. "Copy the statement", run it in a SQL editor of your workspace, export the result as CSV, then "Import the result". YALineo calls no AI. Databricks bills the call to your account.
  6. Validate in bulk. Tick terms, then "Validate", "Back to validate", "Change the domain" or "Reject". A rejected term is not proposed again. The pen of a term, or a double click on its row, opens each of its columns above the list, yours included, to change them.
  7. Browse the list. It fits the screen: its headers stay on top, the term and the actions on the sides, and its sideways scroll bar stays in view. The arrows under the list scroll the columns too. The search follows each letter typed, across the whole glossary, your columns included, and highlights the words found. A click on a header sorts the list (a second one, the other way; "First order" goes back to the original ranking), and you drag the edge of a header to widen its column. Under the list: "Terms 1 to N of N", the number of columns, "Terms per page" (50 to 1,000) and the pages.
  8. Arrange the columns your way. "Columns": rename each column in your team's words, YALineo's included (leave it empty to go back to the original name), show or hide them, add your own, delete your own with their values ("Undo" right after). You can rename and hide YALineo's columns, but not delete them, because they serve the links, the filters and the publication. The search reads every column, and a term's sheet and the exports carry them. A new import of the same file, or of the CSV YALineo exported, fills the same columns.
  9. Open a term. Its sheet gives its definition, synonyms, domain, owner and the tables and columns tied to it; it exports as inert Markdown, where a booby-trapped name never becomes a link or an image.
  10. Export. "CSV" or "Excel" (.xlsx), with your columns and the names you gave them. In the workbook, every text stays text, never a formula. The table published to Unity Catalog keeps YALineo's columns.
  11. Start again from a clean glossary, if needed. After a failed import or an extraction that does not suit you, click "Erase terms", at the end of the terms bar. Choose the whole glossary, the terms that came from an imported file, or those extracted from Unity Catalog. The panel says how many go and how many stay (a validated term, or one another source also gave, stays when you erase a source). "Export first as CSV" keeps a copy. Type the erase password, then click "Erase N terms". "Undo", in the message that follows, puts everything back. YALineo erases nothing in Unity Catalog, nor in the published table.
  12. Publish to Unity Catalog. Enter the glossary table (catalog.schema.table). YALineo writes the script: CREATE TABLE IF NOT EXISTS, a MERGE of the validated terms and, if you tick the box, a business_term tag on each linked column. The owner of the schema reviews and runs it. "Read the published glossary" then reads that table back, read only, and the whole team shares the same glossary.

To check The tiles at the top follow your validations. After the script runs, "Read the published glossary" gives the number of terms read from the table.

If it does not work4 cases
"Columns not read yet".
Click "Read the columns". It is one read-only query, with your permissions, and its result stays with the map.
"Glossary not saved: it would exceed 2 MB".
Remove rejected terms or links, or publish the glossary to Unity Catalog and read it from the table.
The file is refused.
The message says why: 10 MB and 5,000 rows at most, a readable .xlsx workbook, a CSV with no quotation mark left open.
"Wrong password: nothing was erased".
The erase password prevents an erasure by mistake, and YALineo shows it nowhere. Without it, remove terms by ticking them, then click "Reject".

Good to know YALineo keeps the glossary on this device, with the workspace settings (2 MB at most), until "Erase everything on this device". On the fictional example, it keeps it in the open window only. Its terms also show in the data catalog, where the "Term" chip searches them and a table's sheet lists them.

See the 3 screenshots
The glossary after extraction: proposed terms, ranked by use, with their definition, their source and the tables that carry them.
The glossary after extraction: proposed terms, ranked by use, with their definition, their source and the tables that carry them.
Importing an Excel workbook: each column kept under its own name, the ones YALineo recognizes by their header ("Business term", "Also known as", "Data steward"…) matched, editable, then the preview of the linked terms.
Importing an Excel workbook: each column kept under its own name, the ones YALineo recognizes by their header ("Business term", "Also known as", "Data steward"…) matched, editable, then the preview of the linked terms.
"Erase terms" after an import: the choice (the whole glossary, an imported file, the extraction of Unity Catalog), what goes and what stays, the export first and the erase password.
"Erase terms" after an import: the choice (the whole glossary, an imported file, the extraction of Unity Catalog), what goes and what stays, the export first and the erase password.

Data and lineage · 5 min

Descriptions and tags: propose, validate, apply

Contents

A workshop to document the tables that matter. YALineo proposes a description, so does Genie, you validate, and the owner applies. YALineo never writes to Unity Catalog.

Before you start

  • The map loaded.
  • To apply: the owner of each table, with the MODIFY and APPLY TAG rights.
  1. Open "Descriptions and tags". "Data and lineage" section. At the top: tables documented, to document, validated and applied. Tables come most useful first: most read, feeding reports, gold or platinum, certified.
  2. Pick a table. YALineo writes its proposal from what it observed: sources, jobs, readers, refresh rhythm. "Read the columns" adds the main columns and spots personal data by name.
  3. Compare with Genie. In Catalog Explorer, "Generate with AI" also proposes a description. Paste it in "Genie's proposal", then choose "Use", "Combine with YALineo's", or rewrite in "Description to validate".
  4. Set the tags. Tick the proposed tags (domain, layer, refresh rhythm, personal data), change their value or add your own. The domain is the source system in bronze and silver (oracle for oracle_sftp_accounting), the business domain elsewhere.
  5. Validate. "Validate and go to the next" keeps the validation on your device.
  6. Have it validated. "Sheet to validate" downloads the catalog sheet of the tables handled, one section per owner with a decision to tick (approved, to fix), to send for validation. "Sheet as CSV" gives it as a spreadsheet.
  7. Have it applied. "SQL script" gathers every validation by owner (COMMENT ON TABLE, SET TAGS). The owner reviews and runs it. On the next load, the table shows as "Applied".

To check The "Validated, SQL to run" tile grows with each validation; after the SQL runs and a new load, "Applied in Unity Catalog" grows.

Good to know Validations stay on your device until the SQL runs. "Erase everything on this device" erases them too.

See the screenshot
The workshop: YALineo's proposal, Genie's pasted beside it, the proposed tags and the description to validate.
The workshop: YALineo's proposal, Genie's pasted beside it, the proposed tags and the description to validate.

Data and lineage · 2 min

Welcome a new employee

Contents

"Start here" sums up the platform for someone who arrives: how it is organized, the typical route of the data, and the tables to know.

Before you start

  • The map loaded, with activity (reads).
  1. Open "Start here". On the home page of the studio, under "To review first" and "Data read". "How to read this page" says in three points what each part does and where to click.
  2. Read "The platform". The number of catalogs and tables, the lines, where data comes in, the layers it goes through, who reads it and the share of tables described.
  3. Follow "The typical route". The most telling path, from source to report. Click a station to open it.
  4. Keep "The N tables to know". YALineo picks the most read, those that feed reports and the junctions, favoring gold and platinum, across several lines, each with its reason. It never includes information_schema or the system catalog, unless there is nothing else. A click on a table's card opens its sheet.
  5. Go on. Under "Next", three buttons: "See the platform map", "Search the catalog", "Measure the impact of a change".
  6. Share it. "Copy the guide" gives a Markdown text to paste in a wiki or a welcome message.

To check The guide names your catalogs and your most read tables.

Good to know The guide describes what was observed. Ask the owners for the business meaning of each table. To learn Databricks itself, the "Learn" section offers a path per profile and the platform dictionary ("Learn" tutorial).

Data and lineage · 5 min

Export the maps, pages, a view, see what changed

Contents

You download the maps as shown, and pages and their tables as an HTML report, a PDF, CSV files or an Excel workbook. A view of the map or of the dependencies downloads for a document, a slide or a wiki. You can also read what changed in the lineage since a date.

Before you start

  • To compare two versions of the map: "Keep the map and its history on this device" in the settings (step 5), and at least two loads.
  1. Download a map. "Export" page, first "The maps": "Platform map" and "Cost map" as a PNG image or an HTML page; "An asset's dependencies" as PNG, SVG or Excalidraw, once the full name of an asset of the map is entered (catalog.schema.table, the list offers the names). Each map opens for a moment with its current filters, downloads as shown, then you come back to "Export".
  2. Choose the pages and elements. Lower, "Export pages and elements": tick the data catalog, the glossary, the cost overview, the jobs, the table costs, the SQL warehouses, AI and Genie, catalogs and schemas, teams, projections, your estimates or the findings. An element with no data (costs not read, empty glossary…) or outside your profile says why it cannot be ticked.
  3. Choose the format, then "Export". HTML: a report in one file. PDF: the same report opens in a tab with the print window, where you choose "Save as PDF". CSV: the file of each page, several in a .zip. Excel: one workbook, a sheet per element. Each element holds the rows of its page's CSV, with the FinOps filters of the bar.
  4. Open the view to export. The map, or a table in "Dependencies of an asset".
  5. Open "Export". Choose the format: Excalidraw (editable diagram), SVG (image for a document or a slide), PNG (picture of the view as shown, metro or graph, for an email or a chat), Mermaid (text for a README or a wiki), draw.io (diagrams.net and Confluence), or "JSON snapshot" (to import back into YALineo).
  6. Mask the names if needed. "Mask names" keeps the structure and replaces the names, to show an architecture outside the company. It applies to the view exported and to an asset's dependencies. The platform map and the cost map keep their names.
  7. See what changed. "How dependencies change": over 7, 30 or 90 days, or since a date, the new links, the links no longer observed, the tables created and the workloads changing rhythm, to export as CSV.
  8. Compare two versions. On the same page, "Compare with an earlier version of the map": assets and links that appeared, and those gone now.

To check The "File downloaded" message gives the name of the file.

Good to know Something missing is not something deleted, because the period, the permissions and the scope also play a part. Check before concluding.

See the screenshot
"Export", first "The maps": the platform map and the cost map as PNG or HTML, the dependencies of a named asset as PNG, SVG or Excalidraw.
"Export", first "The maps": the platform map and the cost map as PNG or HTML, the dependencies of a named asset as PNG, SVG or Excalidraw.

Costs · 4 min

Read the cost overview

Contents

The "Costs" section ties Databricks billing to the map: what each table, each job, each SQL warehouse and each AI product costs, and what runs with no read observed. It opens on its overview.

Before you start

  • The FinOps side ticked in the settings, and read access to system.billing.
  1. Open the "Costs" section. The "Overview" opens. Its badge gives the number of points to look at.
  2. Choose the period. "7 d", "30 d", "90 d" or "12 months", or two dates in the calendar, always set against the period before. "Filters" and "Export" are next to it.
  3. Read the six cards. Spend at list price, DBUs used, share paid for table refresh, tables with no read observed (per month), failed runs, projection of the next 30 days. Each card carries its nature (Observed, Attributed, Estimated or Potential) and its curve when the billing gives it. Its "i", top right, explains how the figure is made, with a link to the method. Escape closes it.
  4. Read "The essentials". A few sentences written from the period's figures: the change, what causes it, the last week, the tables with no read observed, the failures. Each name opens its sheet.
  5. Look first at what matters. "Look at first": the cost findings that call for a decision, with their amount. "All findings" opens the "To review" section.
  6. Scroll down. The spend day by day or week by week, by product, against the period before; a waterfall from the period before to this one, to the cent; the split by product, team and schema; the largest changes, with their curve.

To check The sentence under the title gives the exact period, from the first to the last day, and whether amounts are at list price or after discount.

If it does not work1 case
"Billing is not readable".
Read access to system.billing.usage and system.billing.list_prices is missing ("Grant access" tutorial). The lineage map still loads.

Good to know Amounts are at Databricks list price. You set your discount in the settings, and it applies without reloading. The cloud bill (machines, storage) is not included.

See the screenshot
The cost overview: the period, six cards with their nature, the essentials in sentences and what to look at first.
The cost overview: the period, six cards with their nature, the essentials in sentences and what to look at first.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Costs · 3 min

Choose the period and read the changes

Contents

You choose the period to the day, today included. Each card then compares its figures with the days before, of the same length, and its color shows whether it is a good sign.

Before you start

  • The FinOps side loaded.
  1. On the overview and the cost analysis. The period bar: "7 d", "30 d", "90 d" or "12 months", or two dates in the calendar.
  2. On the other cost pages, choose a length. "Period": "Today", or the last 7 to 365 complete days, up to yesterday.
  3. Or choose two dates. "From" and "to": all of September, a given week, or a single day (the same day twice). The number of days counted shows before you confirm.
  4. Click "Apply". YALineo reads only the costs again, with the period before, and the map does not move. A message confirms "Costs reloaded from … to …".
  5. Read the changes. Under each figure: an arrow, the change in percent and in amount, and "vs the N days before". Green: down. Red: up, in bold from 10%. Gray: a change under 2% (or under 1 in amount).
  6. Hover a change. The tooltip gives the dates of the previous period and its value.

To check The sentence under the page title repeats the chosen dates.

If it does not work3 cases
The date is refused.
The message says why: a missing date, a start after the end, a date after today (UTC date, the one billing uses), or more than a year back (system tables keep one year).
A change starts with "≈".
The period holds today, or its last day ended less than 12 hours ago, and billing may still arrive. The color gives the direction of the change, but it may still move, as the tooltip says.
"Not comparable" or "No comparison".
The period before was not read the same way, or would start more than 365 days ago.

Good to know A share (for example the share paid by tables) that changes is neither good nor bad. It stays gray, and it only shows that the spend is split differently.

See the screenshot
The overview's period bar and its six cards: each with its nature, set against the previous period, in color.
The overview's period bar and its six cards: each with its nature, set against the previous period, in color.

Costs · 3 min

Filter: a team, a catalog, a product

Contents

Filters answer budget questions: how much the BI team spends, what the finance catalog costs, what serverless workloads cost.

Before you start

  • The FinOps side loaded.
  1. Open a "Spend" list. "Product", "Type", "Tag" (for example team = bi or cost_center), "Run as", and "All", "Serverless" or "Classic". YALineo computes everything again from the chosen workloads, table costs included.
  2. Or a "Tables" list. "Catalog", "Schema", "Domain", "Object type": the tables to add up and list. A view is never refreshed, so its reads count for its sources. The cost of a table does not change.
  3. Tick one or more values. "Search" finds a value in a long list; "Clear selection" starts over.
  4. Read the "Selected tables" block. Their exact cost, refresh and reading, per month, the most expensive, and "Who pays".
  5. Remove a filter. Each active filter shows as a pill. One click removes it, and "Clear filters" removes them all.

To check The sentence under the title ends with "Filters applied".

Good to know The amounts of the teams add up exactly to the total, with no cost counted twice or lost. Filters apply to every cost page, "Catalogs and schemas" included, and to the cost map.

See the screenshot
Table costs, filtered on the lake catalog: the filter's pill, then the tables sorted by refresh.
Table costs, filtered on the lake catalog: the filter's pill, then the tables sorted by refresh.

Costs · 4 min

What each catalog and schema costs

Contents

For each catalog and each of its schemas, the page gives the refresh of their tables over the period, their reads and the workloads that pay. Each amount shows whether it is exact or holds a split share.

Before you start

  • The FinOps side loaded (read access to system.billing).
  • For reads: the query history (system.query.history). Without it, they stay "not measured".
  1. Open "Catalogs and schemas". "Costs" section, under "By workload type". At the top: the refresh of every table over the period and per month, set against the previous period; then the share exact at catalog level, the number of schemas, the reads on SQL warehouses and what lies outside tables.
  2. Read a row. For each catalog: the "Refresh" of its tables, "Per month", "Precision", "Reads" (estimated, hence the "≈"), "Tables", "Written by" (the jobs and pipelines that pay) and "Share of spend".
  3. Read the precision. "exact": all the spend goes here, because every workload that wrote its tables wrote only in this catalog (or schema), or because the table is billed in its own name. "split" followed by a percentage: that share comes from a workload that also writes elsewhere, or from a SQL warehouse shared hour by hour, and follows the table rules. The rest of the amount is exact.
  4. Unfold the schemas. A catalog's arrow opens its schemas, with the same columns. "Filter catalogs and schemas" finds a name.
  5. Go down to the tables. A click on the name of a catalog or a schema opens "Tables", filtered on it.
  6. Check the total. The "The spend of the period, to the cent" card adds each catalog, files and objects outside catalogs, reads, idle warehouses, compute writing no table and tables gone from the map. The sum is the Databricks spend of the period.
  7. Export. "Export as CSV" writes each catalog and each schema, to the cent, for the budget review.
  8. Find them in Databricks. In Catalog Explorer, the list of catalogs and a catalog's list of schemas show the same amount next to each name, with no "≈" when it is exact. The tooltip gives the exact and split shares, the workloads that pay most, the reads and the number of tables with a cost.

To check On the example platform (90 days), the catalogs come to $9,734 of refresh, all exact: lake $8,984 (gold $3,962, silver $2,872, bronze $2,150) and semantic $750. YALineo estimates the reads at ≈ $884.40, and the reconciliation ends on the total spend, $12,535.

If it does not work2 cases
Reads are "not measured".
YALineo does not read the query history (analyst mode, or no access to system.query.history). Ask for the full grant script, which adds it ("Grant access" tutorial).
A catalog has "no cost".
No write of its tables was paid over the period, for example tables that were only read. Its reads still count.

Good to know A catalog's amount is surer than its tables'. A workload that writes only to one catalog gives it all its spend, even when the split between its tables stays estimated. The rule is published in the calculation method, section 5.9 "Catalogs and schemas".

See the 2 screenshots
The "Catalogs and schemas" page on the example platform: each catalog and its schemas, "exact" or "split", reads marked "≈", then the reconciliation to the cent.
The "Catalogs and schemas" page on the example platform: each catalog and its schemas, "exact" or "split", reads marked "≈", then the reconciliation to the cent.
In Catalog Explorer, the lake catalog: the refresh cost of each schema next to its name, with no "≈" when it is exact.
In Catalog Explorer, the lake catalog: the refresh cost of each schema next to its name, with no "≈" when it is exact.

Costs · 2 min

Find the tables with no read observed

Contents

Some tables are refreshed every day with no read seen in the lineage. Their cost is an attributed cost, to check with their owner before calling it a saving.

Before you start

  • The FinOps side loaded.
  1. Open "Tables". "Costs" section, under "By workload type", or the overview's "Tables with no read observed" card.
  2. Tick "Only tables with no read observed". For each: "Refresh" over the period, "Per month", the cost's "Precision" and "Paid by" (the job or pipeline).
  3. Compare with those that are read. Untick, then click the "Per read" or "Per M rows" header. Tables expensive for what they return rise to the top.
  4. Export. Click "Export as CSV" for a budget review.

To check The counter next to the filter gives the number of tables kept.

Good to know Before stopping a refresh, check with the owner. Databricks lineage does not see a read that bypasses Databricks (an outside engine reading the storage files directly, for example).

See the screenshot
Only tables with no read observed: what their refresh costs per month, and who pays for it.
Only tables with no read observed: what their refresh costs per month, and who pays for it.

Costs · 3 min

Jobs and compute: who uses what

Contents

The page lists every workload billed over the period: jobs, pipelines, SQL warehouses, clusters, AI agents, serving endpoints, apps. For each, it gives its cost, runs, failures and the tables it pays for.

Before you start

  • The FinOps side loaded.
  1. Open "Jobs and compute". "Costs" section, under "By workload type". The most expensive workloads come first.
  2. Choose what to show. "All", "Jobs and pipelines", "Warehouses and clusters", "AI", "Rising", "With failures", "Not attributed".
  3. Read the columns. "Cost" and "Per month", "DBU", "Runs", "Per run", "Failures" (number and amount), "7 days" (the trend of the last week), "Attributed" (the share linked to tables) and "Tables paid".
  4. Open a workload. A click on its name opens its card; the curve icon opens its trend over time (cost, runs, failures, duration).
  5. Spot the master jobs. A job that only launches other jobs ("Run Job" tasks) costs 0 on its own, because each launched job is billed under its own name. The "Jobs launched" column shows beside it what the jobs it launches cost, and how many. Hover it to see the list, job by job, with their runs. That amount is already counted under each launched job, so YALineo never adds it to the totals. In Databricks, the badge of such a job shows it after "↳".
  6. Sort and adjust. A click on a header sorts the column, and you drag its edge to widen it (next tutorial).

To check The "serverless" pill follows the name of serverless workloads; paid failures are in amber; a master job shows 0 and, beside it, the jobs it launches.

If it does not work1 case
A name followed by "(name of its latest run)".
This job is not in system.lakeflow.jobs (the documentation cites a recently created job). YALineo takes the name of its latest run and says so. A renamed job shows its latest name; with no name at all, its id shows, with the reason.

Good to know Databricks bills jobs run on an interactive cluster to the cluster, not to the job, so they show in the cluster's row.

See the screenshot
Jobs and compute: each workload with its cost, runs, failures and its 7-day trend; a master job with the cost of the jobs it launches.
Jobs and compute: each workload with its cost, runs, failures and its 7-day trend; a master job with the cost of the jobs it launches.

Costs · 4 min

SQL warehouses, AI and Genie, teams and tags

Contents

Under "By workload type", the "Costs" section has one page per kind of spend: Jobs and compute, Tables, SQL warehouses, AI and Genie, Catalogs and schemas, Teams and tags. Each has its cards, its curves and its export.

Before you start

  • The FinOps side loaded. For the cost per query and the Genie Agents' questions: the query history (system.query.history).
  1. Open "SQL warehouses". The cost of each warehouse, its cost per query, the capacity paid with no query, its size and its auto stop.
  2. Open "AI and Genie". The AI cost, its workloads, the Genie Agents' questions, and what is not billed (calls and messages). Then each AI workload, with its curve.
  3. Read the product of each line. A Genie line carries the product the billing gives (field usage_metadata.genie.surface): Genie Code (formerly Databricks Assistant), Genie One (formerly Databricks One) or Genie Agents (formerly Genie spaces). A line whose billing does not give the product is called "Genie, type not indicated", with "Why?" next to it. YALineo does not guess the product, and the cost stays measured on that separate line.
  4. Read "By product". Each AI product, rounded to add up to the AI cost to the cent. YALineo shows Genie's free usage (GENIE_FREE_USAGE) separately, "of which N free". This free usage does not lower the average price of a DBU.
  5. Read the Genie Agents' questions. The SQL compute of each agent's questions, attributed hour by hour, is already in the cost of the SQL warehouses and is never added to the AI cost. The language model of all agents sits on the Genie Agents line. These are two distinct costs, never added together.
  6. Open "Teams and tags". Here, a team is the value of a billing tag. YALineo sorts the spend by the chosen "Tag key" (team, cost_center, project…), to the cent, workloads without that tag included. YALineo offers the first key that tells at least two teams apart, preferring team, equipe, owner_team or squad. A key carried with a single value by all the spend (for example Application = Databricks) tells no one apart. The page says so; choose another key.
  7. Export. Click "Export as CSV". The "AI and Genie" export adds the product shown, the billing fields that give it, and the free DBUs.

To check The "By product" lines add up to the "AI cost", to the cent.

Good to know YALineo does not use Genie. It reads what the billing and the query history say about it. The names and the rules are in the method, section 4.3.

See the screenshot
AI and Genie: each line named as its billing says (Genie Code, Genie Agents, Genie One and its free DBUs), the cost by product to the cent, and the Genie Agents' questions on SQL warehouses.
AI and Genie: each line named as its billing says (Genie Code, Genie Agents, Genie One and its free DBUs), the cost by product to the cent, and the Genie Agents' questions on SQL warehouses.

Costs · 3 min

Projections: end of month, 30, 90 or 365 days

Contents

The page computes what the same usage would cost, and how many DBUs it would use, if each workload kept its pace: for the platform, each product, each job, pipeline or warehouse, and each table. It is a projection, never a measure, and the page says so.

Before you start

  • The FinOps side loaded, over a period that ends before today (for example "Last 30 days"), to get the pace of the last 7 days and the spread between the two paces.
  1. Open "Projections". "Costs" section, under "Plan ahead", or a click on the overview's "Projection, next 30 days" card.
  2. Choose the horizon. "End of month", "30 days", "90 days" or "365 days". The dates and the number of days show under the title.
  3. Or choose dates to come. "Window ahead": two calendars that offer future days only (from the day after the measured period up to three years later), then "Apply". The period of the bar, "Measured over", stays the past that gives the pace.
  4. Read the cards. "Projected cost" and "Projected DBUs", with the spread between the two paces below; "Per day, pace used"; "Recent pace against the period": above zero, spending is speeding up. Hover a card to see the two paces.
  5. Check where the figure comes from. The "Where the projection comes from" block gives what was measured, the pace of the period, that of the last 7 days and the days projected. "Projected cost by product" shows where the spend will go.
  6. Go down to the detail. "Jobs and compute" lists each workload: measured, per day, projected, between two paces, projected DBUs and recent pace. YALineo marks a workload billed on less than one day a week (a monthly close, a job run on demand) as "intermittent", and that workload follows the pace of the whole period. "Tables" gives the share of each table, with its precision.
  7. Export. "Export as CSV" writes the total, each workload and each table, with the dates of the horizon, for the budget.

To check The projected cost equals the sum of the workloads, and the overview's card gives the same figure as the "30 days" horizon.

Good to know A filter (a team, a product) or the discount of the settings apply to the projections too. To project a team's budget, filter on its tag.

See the screenshot
Projections over 30 days: the projected cost and DBUs between two measured paces, where the figure comes from, then each product, workload and table.
Projections over 30 days: the projected cost and DBUs between two measured paces, where the figure comes from, then each product, workload and table.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Costs · 2 min

Sort and adjust the tables

Contents

The cost tables, jobs, the catalog and the other lists sort with one click and adjust with the mouse or the keyboard.

  1. Sort with a click. Click a header. A column of numbers puts the largest first, and a column of words goes from A to Z. A ▲ or ▼ arrow shows the sorted column and the direction.
  2. Reverse with a second click. The order goes from Z to A, or from smallest to largest. Empty values stay at the end either way.
  3. Widen a column. When you rest the mouse on the right edge of a header, a blue bar appears. Drag right to widen, left to narrow. The other columns keep their width, and the table scrolls if needed.
  4. With the keyboard. Tab to the edge of a header, then press the right or left arrow. The column grows or shrinks by 16 pixels each time.
  5. Go back to the original widths. Double-click a header's edge.
  6. In lists. In a station's card or a table's upstream and downstream, the "Sort" list picks the order (most read, by name…) and the ↓ ↑ arrow next to it reverses it.

To check The order and the widths stay while the studio is open, even when moving from one view to another.

Good to know The sort of table costs goes into the studio's address, so a copied link opens the same order again.

See the screenshot
A sorted header (the arrow), and a column's grip on hover, to widen or narrow it.
A sorted header (the arrow), and a column's grip on hover, to widen or narrow it.

Costs · 2 min

What a dashboard costs

Contents

A dashboard costs what its queries cost, plus the refresh of every table it depends on.

Before you start

  • The FinOps side loaded.
  1. Open the dashboard. With the studio's search, or from a table it reads.
  2. Read its card, "Costs" section. The section gives its queries and its upstream data, with the number of tables.
  3. See the big picture. Open "Cost map", under "Plan ahead". Each station shows what its tables cost. The cost filters apply, and "Change filters" goes back to them.
  4. Follow a table and its depth. Choose a table ("Table" filter, search or a click on a station), and the "Depth" list appears, with direct links, 2 levels, 3 levels or the whole lineage. The same list is on the platform map.
  5. Download the map. Click "Download", at the top right. "PNG picture" gives the map as drawn, and "Self-contained HTML page" gives a page that opens in any browser, with the map's date, the periods and the table followed. The "Export" page offers it too, "The maps" card.

To check A table's "With all upstream" amount is at least its own refresh.

Good to know A table shared by several dashboards counts in full for each, because the amount measures what each dashboard relies on, not its share of the bill.

See the screenshot
The cost map: each station shows what its tables cost.
The cost map: each station shows what its tables cost.

Costs · 5 min

Cost analysis: compose, save, explore

Contents

This page lets you compose your cost analysis, save it and open it again. Lower down, you find an explorer that crosses any measure with any axis, the questions of one table, job or pipeline, and the comparison of several assets. Each section folds to leave room for the one you are working on.

Before you start

  • The map loaded; for spend, the FinOps side.
  1. Open "Cost analysis". "Costs" section. At the top, the period, "Filters" and "Export", as on the overview.
  2. Fold what you do not use. Each section ("Overall cost analysis", "Start from a question", the examples, the answer, the detailed analysis, the comparison, the saved analyses) has its "Fold" button. Folded, it keeps its sentence on one line, and it stays folded next time. "Fold all" folds the page at once.
  3. Start from a question. "Start from a question" is folded at first, and "Show" unfolds it. It offers, for example: which products cost more than the period before, which workloads rose most, which team spends most, where failures gather… One click sets the analysis.
  4. Compose. Set "Measure" (cost, DBUs, runs, failures, cost of failures, table cost, table reads), "Split by" (product, kind of workload, compute, workload, team, run by, catalog, schema, table), "Grain" (day, week, month), "Chart" (stacked, lines, bars, gaps, table), then the product and the team.
  5. Read the answer. A sentence gives the total, the gap with the period before, the value in the lead and the largest rise. "The data" gives each value. When a choice is not possible (no daily grain for a measure read over the whole period), the page says why.
  6. Save it. Name the analysis, then click "Save". It joins "My analyses" in the side bar and opens again from its address. "Link" copies it, "CSV" exports it.
  7. Explore lower. The example questions, folded at first, sit in three columns, the whole platform's costs first. They set the explorer: "Data", "X axis", "Y axis", "Calculation", "Color (series)", "Chart", and "Add a filter". While a question is being read, a "Reading" card says what is read and for how many seconds. The answer takes its place, and the rest of the page stays usable.
  8. Choose the period. Use the two independent calendars of the Costs pages, one for the start and one for the end, or choose 7, 30, 90 or 180 days. The page refuses a period too old for the trend queries, and says why. An answer that reads the costs loaded with the map says so in one line.
  9. Dig into one asset. Choose a table, a job or a pipeline, and the detailed analysis asks its questions. For a table: who reads it, its last write, what its refresh and its reads cost, who writes it, what depends on it. For a job or a pipeline: runs and failures, the cost of failures, duration, cost per run, what it writes. Data missing from the map shows as "not read", never 0.
  10. Compare assets. "Compare assets": search and add up to six tables, jobs or pipelines. Choose the measure (cost, DBUs, runs, reads, writes) and the chart (lines, columns, table). Each asset keeps its color, whatever the order, and a table gives its totals, the period before and the gap.
  11. Save the whole page. "Save the analysis" keeps the question, the exploration, the period, the asset analyzed, the assets compared and the filters, never an amount, because YALineo reads the figures again when it opens. Open, rename or delete it in "Saved analyses". "Undo" brings a deletion back.
  12. Download. Each chart downloads as PNG or SVG. "Download the page (HTML)" gives one self-contained file, with no script, to attach to a report.
  13. Click a bar. Keep it, exclude it, analyze it over time, or open its impact analysis or its description workshop. Each view has its address; "Back to the question's view" resets it.

To check The values shown, "Other" included, add up to the total of the sentence, to the cent.

Good to know A saved analysis keeps its name, settings, filters and dates, never an amount, under the map's retention rule. You can keep thirty at most per workspace.

See the 3 screenshots
The cost analysis: ready-made questions, the measure, the breakdown, the grain and the chart, and the sentence that sums it up.
The cost analysis: ready-made questions, the measure, the breakdown, the grain and the chart, and the sentence that sums it up.
Lower on the same page: the example questions and the explorer, set to the spend per week and per kind of workload.
Lower on the same page: the example questions and the explorer, set to the spend per week and per kind of workload.
"Compare assets": three jobs on one chart, each in its color, with their totals, the period before and the gap.
"Compare assets": three jobs on one chart, each in its color, with their totals, the period before and the gap.

Simulations · 4 min

Learn with the worked cases

Contents

To get to grips with the Calculator, the quickest way is to start from ten business situations. Each is an estimate already filled in, with a guide that says where the figures come from and what to change to understand. The situations are fictional; the prices are those of the bundled public snapshot.

Before you start

  • Nothing but the extension.
  1. Open "My estimates". "Simulations" section. "Learn with a worked case" is open while you have fewer than three estimates. After that, a click on its title unfolds it.
  2. Pick a case. "Deliver web orders every night" (a job sized from 40 million rows), "Move five years of history" (2 TB copied once), "The dashboard of 200 store managers" (serverless SQL and active hours), "Till receipts in real time" (a stream on for 730 hours), "The monthly financial close" (a large cluster, few runs, Photon), "The data science team" (interactive cluster against a job), then four cases of services: "The customer service document assistant", "Supplier invoices", "The field technicians' app", "Sensors and the CRM, continuously".
  3. Open it. "Open this case" creates a new estimate, leaving yours untouched, on the cloud of your settings, and opens its first workload.
  4. Read the guide. At the top of the estimate: "The need" (who asks for what), "What to gather" (the figures to ask for before estimating), "Try this" (two or three changes and what they show), "Keep in mind" (the lesson of the case). A click on its title folds it.
  5. Try. Do what "Try this" says: double a volume then "Use this duration", put workers on spot, compare a reservation. The total updates as you type, and a workload's "i" shows the calculation step by step.
  6. Keep what helps you. The case is an estimate like any other. You can rename it, change the region and your discounts, duplicate it to compare, or delete it ("Undo" brings it back for a few seconds).

To check Each case of jobs, pipelines, warehouses and clusters opens fully priced on Azure and AWS, its total equal to the sum of its workloads to the cent. On GCP, you must enter the machines' hourly price, and the case says so, line by line.

If it does not work1 case
The "Open this case" button is grayed out.
YALineo keeps thirty estimates at most on the device. Delete one.

Good to know Durations taken from a volume follow a proportional rule, which describes a best case. As soon as you have a real measured run, put its duration instead.

See the 2 screenshots
The worked cases of the "My estimates" page: each describes a business need and opens as a complete estimate.
The worked cases of the "My estimates" page: each describes a business need and opens as a complete estimate.
An open case: its guide at the top (the need, what to gather, what to try, what to keep in mind), then its workloads and their total.
An open case: its guide at the top (the need, what to gather, what to try, what to keep in mind), then its workloads and their total.

Simulations · 6 min

Price a workload before you build it

Contents

The Calculator tells what a job, a pipeline, a SQL warehouse or a cluster would cost on Azure, AWS or GCP, before you build it: DBUs, machines, total per month and per year, and where each figure comes from. Nothing is sent, because the public prices, dated, are bundled in the extension.

Before you start

  • Nothing but the extension. The Calculator works without a workspace, with the bundled public prices.
  • To start from your real workloads, the FinOps side loaded (tutorial "Make a measured workload vary").
  1. Open the "Simulations" section. "My estimates" lists what you kept on this device.
  2. Choose the right tool. "Which tool for which question?", under the title, says what each costing tool does: the estimate (what do these exact workloads cost per month?), "Cost a project" (what will this project cost, from the source to the dashboard?), "Size and tune" (what size or setting for a given workload?) and "AI: Genie and agents". Each comes with what you describe, what you get and an example. "You are here" marks the open page, and a click on another name opens that tool. They price with the same engine and the same prices, and chain together.
  3. Read the page's course. "Learn more", next to the title of "My estimates" and to that of an estimate, opens the page's course in a reading panel, on the right: the pricing states, reopening or recalculating, comparing, sharing, paying for machines, where each price comes from, with its official sources. Escape or the cross closes it.
  4. Start from a worked case or the fictional example. "Learn with a worked case" offers ten business situations, each with its guide (tutorial "Learn with the worked cases"). "Fictional example" creates an estimate already filled, that of a fictional retailer: a nightly sales ingestion, the finance dashboards on serverless, a stock pipeline with its quality rules, the data scientists' cluster, a serverless reconciliation and the analysts' SQL Pro warehouse.
  5. Set the cloud, the region and your discounts. In "Settings": Azure, AWS or GCP (your workspace's cloud is marked), the region (prices depend on it), the Databricks plan, then your discounts on DBUs and on machines. Once folded, the block keeps a one-line summary.
  6. Add a workload. "Job", "Pipeline", "SQL warehouse" or "All-purpose cluster", "A measured workload", a service ("Add a service" tutorial), or a fictional example from the list. Each field has its "i": hover or tab onto it to know what to enter.
  7. Describe it. The instance type (type D4ds, m5d or n2, or "Browse" to filter by vCPUs and memory, cheapest first), the number of workers, Photon, how the workers and the driver are paid, then the pattern: runs a day and minutes, hours a day, or hours a month.
  8. Start from the volume, if you know it. "Size from the data volume": choose the ingestion type (batch, Auto Loader, streaming, pipeline), the volume per run in GB or million rows, and a reference run (a test or a comparable job: volume, minutes, workers). YALineo computes the duration and shows the formula, and "Use this duration" applies it. Databricks bills compute time, not rows, so the volume counts through the duration.
  9. Read the total. "Estimate total" stays beside the workloads: per month, per year, DBUs (billed by Databricks), machines (billed by the cloud) and each workload's share. It changes as you type, without losing the field you are typing in.
  10. Read the pricing state. "priced", "partly priced" or "missing assumptions", with what is missing. Only a fully priced estimate can be named the cheapest.
  11. Open it again as saved. An estimate keeps its prices, the date of its measures and its total. "As saved" gives the same total to the cent. "Recompute with the current data" shows the gap, and nothing changes without your say.
  12. Check the calculation. A workload's "i" opens "The calculation step by step" with its own numbers; "The four ways to pay for the machines" compares pay as you go, spot and reservations, and marks the cheapest.

To check The estimate total equals the sum of its workloads, to the cent. A workload without a price is marked "not priced" with what to enter, never counted as zero.

If it does not work3 cases
"Choose your workspace's region".
Without a region, nothing is priced. Choose it in "Settings".
"No … price for this instance type".
On GCP, Google publishes its machine prices only through an API with a key. Enter the price per hour beside it, and YALineo keeps it for that cloud. Elsewhere, choose another way to pay or another type.
An unknown instance type after switching cloud.
Take the equivalent offered: the cheapest one with at least as many vCPUs and as much memory.

Good to know When you switch cloud, the workloads stay. Each machine becomes its equivalent, and the region becomes the nearest one. Duplicate the estimate first to compare both clouds side by side.

See the 3 screenshots
"Which tool for which question?" on "My estimates": each costing tool with its question, what you describe, what you get and an example; "You are here" marks the open page.
"Which tool for which question?" on "My estimates": each costing tool with its question, what you describe, what you get and an example; "You are here" marks the open page.
An estimate: "As saved", its exports, its folded settings, six fictional workloads and their total always in sight.
An estimate: "As saved", its exports, its folded settings, six fictional workloads and their total always in sight.
An open workload: each field with its help, its calculation step by step and the four ways to pay its machines.
An open workload: each field with its help, its calculation step by step and the four ways to pay its machines.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Simulations · 4 min

Add a service: AI, models, data and platform

Contents

Beyond jobs, pipelines, warehouses and clusters, an estimate prices seventeen kinds of services, from official rates that are sourced and dated, and from your account's prices when you have read them.

Before you start

  • An open estimate ("Price a workload before you build it" tutorial).
  1. Click "Service". In "Workloads", next to "Job", "Pipeline", "SQL warehouse", "All-purpose cluster" and "A measured workload".
  2. Pick the kind. "AI and models": Model Serving (CPU, GPU), Foundation Model APIs, proprietary models (Anthropic, OpenAI, Google, xAI), AI Search, Agent Evaluation, Unity AI Gateway, AI Parse, AI Extract, AI Classify, AI Runtime, Shutterstock ImageAI. "Data and platform": Lakebase, Databricks Apps, Lakeflow Connect, Zerobus Ingest, default storage, platform options.
  3. Describe the usage. Each field says its unit and where its value comes from. Any quantity nobody can know in advance is left for you to enter. Until you do, the workload stays "missing assumptions", never counted as zero.
  4. Read the calculation step by step. YALineo prices each workload to the cent and gives each price with its origin (entered, your account's price, public price), the published ranges, the free shares and the discount rule.
  5. Or start from a case. Four guided cases of services: "The customer service document assistant", "Supplier invoices", "The field technicians' app", "Sensors and the CRM, continuously" ("Learn with the worked cases" tutorial).

To check The estimate's total is the sum of its priced workloads, to the cent; a workload to complete says so, with what is missing.

Good to know You enter what Databricks does not publish, or what two official pages give differently, and the page gives the reason and the pages checked. The services snapshot is dated 26 October 2026. See the method, sections 13.11 and 13.12.

See the screenshot
An estimate's "Service" menu: seventeen kinds in two groups, "AI and models" and "Data and platform".
An estimate's "Service" menu: seventeen kinds in two groups, "AI and models" and "Data and platform".

Simulations · 5 min

Cost a data project

Contents

Describe a project in plain terms (sources, transformations, people who read, questions, models). YALineo derives its Databricks workloads and their monthly cost with the Calculator's engine, giving a range and every rule written down.

Before you start

  • Nothing but the extension; for the comparison with your platform, the FinOps side loaded.
  1. Open "Cost a project". "Simulations" section, in the side bar; or "Which tool for which question?", on "My estimates" and "Size and tune"; or the "Cost a project" button of "Size and tune". The same band, under the title, recalls what each costing tool does.
  2. Start from a model. "BI warehouse for a domain", "Real-time e-commerce", "Analyst self-service and Genie", "Data science and a served model", or "Empty project". You can change each value afterwards. YALineo presents no figure of a model as a measure, and a value nobody can know in advance stays to be entered.
  3. Set each building block. For each block (ingestion, transformations, dashboards, questions, models…), a switch turns it on, and its fields say their unit (feeds to load, ingestion mode, runs per day, minutes, minutes at peak, machine type). Each block opens in the tool that sizes it, the same one as in "Size and tune".
  4. Read the cost. "Expected monthly cost" and its range, the pricing state, per year, the DBUs per month, the development share, then your platform measured over the period and what the project would add, on the same scope.
  5. Change the assumptions. "Prices" changes the cloud, the region or the plan; lower, how sensitive the cost is to the assumptions, and the same project on another cloud.
  6. Save the scenario. Until it is saved, it is a draft. Save it to find it again in "My estimates". "CSV", "Excel" (the cost by brick and the summary), "Word" and "PDF" (the summary) take it away in a file.
  7. Read the course. "Learn more", next to the title: from the need to the workloads, the rule of each building block, the range and what weighs most, the comparison with your platform, costing by phases, and its official sources, in a reading panel on the right. Escape closes it.

To check The expected monthly cost sits within its range, and the pricing state says what is still missing.

Good to know Each rule and each source is written in the method, sections 13.9 and 13.10. The "How it is calculated" card, closed at first, opens with a click on its title for the detail. "Reset" starts again from the chosen model.

See the screenshot
"Cost a data project": a starting model, the building blocks of the project, the expected monthly cost with its range, and the comparison with the measured platform.
"Cost a data project": a starting model, the building blocks of the project, the expected monthly cost with its range, and the comparison with the measured platform.

Simulations · 7 min

Size and tune: the fourteen calculations

Contents

Fourteen calculations of size, tuning, cost and buying, sorted by capability (data engineering, data science, analytics, dashboards, architecture) and by role. Each answers a question, starts from a concrete example, shows its steps and its sources, and never guesses a missing figure. This tutorial presents the page, and the next four present the calculations of each capability.

Before you start

  • Nothing but the extension. Prices come from the bundled public snapshot, or from those you enter.
  • Your profile can put a padlock on some calculations ("Pick your profile" tutorial).
  1. Open "Size and tune". "Simulations" section. At the top, "Cost a complete case" leads to the ten cases of "My estimates", and "Cost a project" to the page of that name. With a role, a line says how many of the fourteen calculations it includes. The others carry a padlock. Below, "Which tool for which question?" tells "Size and tune" apart from the other costing tools ("Price a workload before you build it" tutorial).
  2. Start from a concrete example. "Concrete examples": one fictional case already filled in per capability (the workers of a nightly load, the auto termination of the data scientists' cluster, the clusters of a warehouse at the morning peak, a dashboard refreshed every hour, the growth of a sales table). "Open the example" opens the calculation filled in.
  3. Go through "Calculations by capability". One line per capability, with its sentence and its calculations. Each calculation carries its kind ("Cost", "Size", "Setting" or "Buying"), the question it answers and "For:" the roles that use it. A click opens it without the example, ready for your figures.
  4. Read "Who uses what". One row per calculation, one column per role, a dot when that role uses it; the last row, "Calculations per role", counts them. YALineo builds the table from the calculations themselves, like the profile of use, so the two cannot disagree. A click on its title folds it ("Fold") or opens it ("Show"). Once folded, it stays so the next times.
  5. In a calculation, fill in "Your figures". The trail "Size and tune · capability" leads back to the page. "Fill with this example" puts in the example's figures, with your cloud's machines. Each field says its unit, its bounds and where its value comes from (typed, measured, your choice, or the Databricks default with its page). A required field left empty stays "To enter", never guessed.
  6. Unfold a field's help. Under each field, "Help" unfolds: "What it is", "Where to find the value" in Databricks, the "Usual values" and, when a system table holds it, a read-only query (one SELECT or WITH statement) to paste in the Databricks SQL editor with "Copy the query". "See the full course" opens the calculation's course.
  7. Read the course. "Learn more", next to the title of the page and of each calculation, opens the course in a reading panel, on the right, without hiding the fields on the left. The course gives: what the tool is for, the notions, the formula term by term, how to measure your own figures (with queries to copy), reading the answer, the pitfalls, a worked example, then the official sources, each with the date it was read. "In this course" leads to each part, and another course replaces the first. Escape or the cross closes the panel and gives the focus back to the button.
  8. Read "The answer". The result in large, a sentence that says it with your figures, then the amounts that lead to it. While a value is missing, "To enter before an answer" names it.
  9. Check the calculation. "The calculation, step by step": each step with its formula, your figures, its result and its numbered source, or "derived". "What the rules say": a rule that points to another calculation opens it in one click. "Prices used and their origin": the price base (cloud, region, plan, snapshot date, discounts, your account's prices read or not) and, for the five sizing calculations, each price used with its origin. "Sources": the official pages, each with the date it was read.
  10. Keep the result. "Save" keeps the calculation in "Saved calculations", at the bottom of the tool and at the top of the page, for the session. It goes when the studio's window closes or reloads. "Reopen" brings it back with its figures, and "Remove" takes it away. Each calculation, or the whole history, downloads as Excel, Word or PDF, made on the device. "Copy the calculation" copies everything as text (your figures, the answer, the steps, the rules, the prices, the sources), for a ticket or an e-mail.
  11. Add it to an estimate, when it is a workload. Four calculations give a workload the estimate knows how to price: workers to fit a window (a job), clusters at peak (a SQL warehouse, off peak and at peak), auto termination (an all-purpose cluster), endpoint size (Model Serving, where the services snapshot publishes that rate). "Add to the estimate" first shows "What will be added", priced as the estimate will price it. "Confirm and add" puts the workload in the open estimate of the same cloud, or else in a new "Sizings" estimate.
  12. Or apply it to a project. The five sizing calculations of "Cost a project" (workers to fit a window, clusters at peak, auto termination, endpoint, storage) offer "Apply to a project: …". The block opens in the current project with your values, and "Apply to the project" shows the recap there before any change.

To check After "Fill with this example", "Example filled in: change a figure to fit your case." replaces the button and the answer shows; "Clear" puts each field back to its starting value. The studio's address keeps the open calculation and your figures, so a reload brings them back.

If it does not work3 cases
"Calculation not included in your profile".
Your profile leaves this calculation out. Use "Change the profile in the settings", or choose "Full" to open everything.
The answer waits for "Calculator price for this machine or product".
The snapshot has no price for this machine in the region chosen (on GCP, machines are not published). Pick the region in "Prices used and their origin", or enter the price in "Prices and discounts". Nothing is assumed.
No "Add to the estimate" button.
The result is not a workload the estimate prices. Copy the calculation. For storage, billed by your cloud, the page says so and offers "Apply to a project: Storage".

Good to know The queries of the courses and the help only read. YALineo shows them, and you run them. Each source gives the date it was read: 24 and 25 October 2026 for the five sizing calculations, 27 October 2026 for the next eight, 30 October 2026 for the ingestion. The rules: method, sections 13.10, 13.14 and 13.15.

See the 4 screenshots
"Size and tune": one concrete example per capability, fictional and already filled in, then the way to the ten complete cases.
"Size and tune": one concrete example per capability, fictional and already filled in, then the way to the ten complete cases.
"What a training run costs", filled with its example: the trail back to the page, the capability, the kind and the roles, your figures (unit, bounds, origin) and the answer.
"What a training run costs", filled with its example: the trail back to the page, the capability, the kind and the roles, your figures (unit, bounds, origin) and the answer.
"Learn more" next to the calculation's title: its course in a reading panel, on the right, its contents, then each part with its sources.
"Learn more" next to the calculation's title: its course in a reading panel, on the right, its contents, then each part with its sources.
"Help" unfolded under "Hours per run": what it is, where to find the value, the usual values, and the read-only query to copy.
"Help" unfolded under "Hours per run": what it is, where to find the value, the usual values, and the read-only query to copy.

Simulations · 6 min

Data engineering: ingest, finish on time, tune a job, a stream, a Delta table

Contents

Six calculations for what you build: a data ingestion, the workers that fit a window, a job's timeout and retries, autoscaling, a stream, the files of a Delta table. They serve the data developer, the data architect and the FinOps lead; the Delta files one also serves the data steward or owner, who decides the retention.

Before you start

  • "Size and tune" open (previous tutorial).
  • Your measures, if you have them: a trial on a known volume, the durations of a job's last runs (system.lakeflow.job_run_timeline), a cluster's load (system.compute.node_timeline), a stream's rate, a table's size and number of files (DESCRIBE DETAIL). Each field's "Help" says where to read these values, with the query to copy.
  1. "Size a data ingestion". How many cores, workers, hours and DBUs to load your data, and what does it cost a month? Choose the way to ingest (Auto Loader on files, COPY INTO in batches, a Lakeflow Connect managed connector, a Kafka or Event Hubs stream) and the compute (classic jobs cluster, serverless jobs, classic or serverless declarative pipeline), then the volume a day, the files a day or the events at peak, the rhythm (a number of loads a day, one every N minutes, or continuous), the time given to a load, the throughput per core measured on a trial and the cluster start. On classic compute, the answer gives the cores and the workers. On serverless or with a connector, the DBUs come from a measured number of DBUs per GB. It prices the month at list price and after discounts, flags small files and counts the backfill of history. The example: 900 GB of till receipts a day, loaded 4 times, 45 minutes each.
  2. "Workers to fit a window". How many workers to finish before 6 am, or any other deadline? Enter the "Volume of one run", the trial ("Trial volume", "Trial duration", "Trial workers", "Trial machine"), the "Share that does not parallelize", the "Start time" and the "Available window". The answer gives the workers, the end time against the deadline and, with a price, the cost a month. The example: 300 GB of sales arrive at 2 am and must be ready by 6 am.
  3. "Timeout and retries of a job". Paste the "Durations of the last runs" (in minutes, separated by commas, 200 at most), then the "Margin above the base", the "Retries (max_retries)" and the "Timeout set today". The answer proposes a timeout and prices the worst case, retries included. In the example, a 45-minute timeout killed 4 runs out of 30. The tool proposes 63 minutes.
  4. "Minimum and maximum autoscaling workers". Describe the measured load in three levels: hours a day and busy cores at low load, at mean load and at peak. The answer gives the minimum and maximum workers, and compares their cost a month with that of a fixed size.
  5. "Size a streaming pipeline". Enter "Peak rate", "Mean rate", "Rate a core holds" (measured in a test), "Acceptable delay", stateful or not, and "Kafka topic partitions". The answer gives the workers, the cores and the trigger. When the delay allows it, it prices AvailableNow batches and compares them with the stream running continuously.
  6. "Files and partitions of a Delta table". Enter "Table size", "Files today", "Growth a day", "Partitions today" and "VACUUM retention". The answer gives the target file size, whether to partition or move to liquid clustering, and how often to run OPTIMIZE and VACUUM. With "Unity Catalog managed table" ticked, a rule offers predictive optimization.
  7. Move on to the estimate. "Size a data ingestion" puts its lines in an estimate with "Add to the estimate", priced as the estimate will. "Workers to fit a window" gives a workload too. "Add to the estimate" makes it a job (these workers, this machine, the runs a day and their duration), and "Apply to a project" sends it to the Ingestion or Transformations block. For the four settings, keep the result with "Copy the calculation".

To check With the example of "Workers to fit a window", the answer gives the end time against the 6 am deadline. If you move the "Available window" from 4 to 2 hours, more workers are needed.

If it does not work1 case
"No number of workers is enough".
The share that does not parallelize alone exceeds the window. Widen the "Available window", lower the "Safety margin", or measure the "Share that does not parallelize" again on a trial.

Good to know With the Data developer profile, "For you" asks "How many workers to finish this job on time?", which opens this calculation filled with its example. The duration taken from a trial follows a proportional rule and Amdahl's law. It gives a candidate configuration, to confirm with a real run. Method, sections 13.10 and 13.14.

See the screenshot
"Size a data ingestion", filled with its fictional example: 900 GB of till receipts a day, loaded 4 times a day with Auto Loader on a jobs cluster.
"Size a data ingestion", filled with its fictional example: 900 GB of till receipts a day, loaded 4 times a day with Auto Loader on a jobs cluster.

Simulations · 4 min

Data science: shared cluster, training run, model endpoint

Contents

Three calculations for the data scientists' work: the auto termination of the notebooks' shared cluster, what a training run or an experiment campaign costs, the size of a model endpoint. They serve the data scientist and the FinOps lead, and depending on the calculation the data developer or the data architect.

Before you start

  • "Size and tune" open ("Size and tune: the fourteen calculations" tutorial).
  • If you have them: the hours of use of the shared cluster, the duration of a training trial, a model's expected traffic (requests per second, execution time).
  1. Start from "For you". With the Data scientist profile, the studio's home page asks three questions that open these calculations, filled with their example: "Which auto termination for the notebooks' shared cluster?", "What will this training run cost?", "What endpoint size for the expected traffic?".
  2. "Cluster auto termination". Fill in the "Cluster machine" and its "Workers", the "Hours of use a day", the "Ends of use a day", the "Current auto termination" and the "Proposed auto termination". The answer prices the idle time saved, or added, a month, and says how many hours a day the cluster stays on. The example moves from 120 to 30 minutes.
  3. "What a training run costs". Fill in the "Compute" (job or all-purpose), the "Machine", GPU or CPU, the "Hours per run", the "Runs a month" and the "Runs at the same time". The answer gives the cost a month, that of one run and the campaign's duration. If you tick "The model will be served", a rule opens "Concurrency and endpoint size" in one click.
  4. "Concurrency and endpoint size". Fill in "Requests per second at peak", "Model execution time", the off-peak traffic and its hours, the "Processor" (CPU or GPU) and "Scale to zero without traffic". The answer gives the endpoint size, the concurrent requests provisioned at peak and the DBUs a day. With the price of a Model Serving DBU, it also gives the cost a month.
  5. Move on to the estimate. "Add to the estimate" turns the auto termination into an all-purpose cluster on for the hours computed, and the endpoint into a Model Serving workload, where the services snapshot publishes that rate on this cloud (otherwise, the page says so). Keep the training run cost with "Copy the calculation".

To check With the example of "What a training run costs", the answer details one run, the month, the DBUs, the machines and the campaign's duration; the DBUs and the machines add up to the month.

If it does not work1 case
The endpoint has no cost a month.
The price of a Model Serving DBU is missing. Enter it in "Model Serving DBU price (optional)". It never counts as zero.

Good to know "Workers" at 0 gives a single node, one machine, GPU included. A scheduled training run costs less on jobs compute than on an all-purpose cluster. On spot, save checkpoints. Method, sections 13.10 and 13.14.

Simulations · 4 min

Analytics and dashboards: warehouse at peak, what a dashboard costs

Contents

Two calculations for the analysts' queries and the dashboards: how many SQL warehouse clusters at the busiest time, and what a dashboard costs a month, per refresh and per viewer. They serve the BI analyst, the data steward or owner, who estimates for the analysts of their domain, and the FinOps lead. The warehouse size also serves the data architect.

Before you start

  • "Size and tune" open ("Size and tune: the fourteen calculations" tutorial).
  • The number of people active at peak and the average duration of a query; for a dashboard, its scheduled refreshes and its viewers.
  1. Start from "For you". With the BI analyst profile, the home page asks "How many clusters to hold the busiest time?" and "What does a dashboard cost per viewer and per refresh?". Each opens its calculation, filled with its example.
  2. "Clusters and size at peak". Fill in the warehouse "Type" (Serverless, Pro or Classic) and "Size", the "Active people at peak", the "Queries per person per hour", the "5-minute peak ÷ hourly average", the "Average query duration", then the "Open hours a day", "Of which peak hours" and the "Auto stop". The answer gives the number of clusters and the queries at once. With a price, it also gives the cost a month.
  3. "What a dashboard costs". Fill in the warehouse, the "Scheduled refreshes a day" and their queries, the "Viewers" and their "Openings per viewer a day", the "Share served by the cache", the "Warehouse starts for it alone, a day" and the "Data updates a day". The answer gives the cost a month, the warehouse hours, the cost per refresh and per viewer.
  4. Understand the example. The example is a dashboard refreshed every hour, even at night, while its data changes once a day. The night refreshes wake the warehouse up and cost more than all the views. Lower the "Scheduled refreshes a day" and watch the answer change.
  5. Go from one calculation to the other. A rule of the dashboard cost opens "Clusters and size at peak" in one click. That one goes into the estimate as two SQL warehouse workloads, off peak at the minimum of clusters and at peak at the number found, and "Apply to a project: Dashboards (BI)" takes it to a project's block.

To check With the dashboard example, a rule counts the refreshes that cannot show any new data. In the recap of "Clusters and size at peak", "Together: … a month" adds up the two workloads.

If it does not work1 case
The answer only gives the warehouse hours.
Without a price, it stops at the hours. Pick the workspace region. For a Pro or Classic warehouse on a cloud whose snapshot does not publish the machines, enter "Machines of a cluster an hour (pro and classic)".

Good to know A dashboard keeps its results 24 hours, and a result served by the cache does not start the warehouse. You find the durations to enter in system.query.history (query_source.dashboard_id). Method, sections 13.10 and 13.14.

Simulations · 4 min

Architecture: storage, Photon, spot

Contents

Three calculations that hold for the whole platform: a table's growth and its storage, the speed-up from which Photon costs less, and what spot saves once its evictions are counted. They serve the data architect, the FinOps lead and the data developer. The storage one also serves the data steward or owner, who sets how many days of history to keep.

Before you start

  • "Size and tune" open ("Size and tune: the fourteen calculations" tutorial).
  • For Photon, one run measured with and one without Photon, on the same data; for spot, the share of your runs hit by an eviction.
  1. "Growth, retention and storage". Fill in "Size today", "Growth a day", "Data rewritten a day" (a MERGE, for example), "Deleted files retention" and the "Horizon". The answer gives the size at the end of the horizon and the average over the period. No snapshot gives the storage price. Enter your cloud's in "Storage price (optional)" to price the month.
  2. "Photon: the speed-up it needs". Fill in the "Compute" (job, all-purpose or classic pipeline), the "Machine", the "Workers", the "Duration of a run without Photon", the "Runs a month", then, if you have it, the "Speed-up measured with Photon". The answer gives the break-even (Photon costs less if it goes at least N times faster) and, with your speed-up, the gap a month.
  3. "Spot workers or on demand". Fill in the machine and the workers, the "Duration of a run", the "Runs a month", then the "Runs hit by an eviction" and the "Share of the work redone after an eviction". The answer says what spot saves or costs a month, the extra work expected against what spot tolerates, and the mean lengthening of a run.
  4. Read the rules. Photon changes the DBU rate, supports neither UDFs nor stateful streaming, and is already on in serverless and SQL warehouses. On spot, the driver stays on demand, and a rule gives your cloud's eviction notice (30 seconds on Azure, two minutes on AWS, 0 or 120 seconds on Google Cloud). Each rule points to its source.
  5. Keep the result. These three calculations add no workload to an estimate. Storage, billed by your cloud, goes into a project with "Apply to a project: Storage". For Photon and spot, keep the result with "Copy the calculation", then set them in the estimate's workload (Photon, how the workers are paid).

To check With the example of "Growth, retention and storage", the answer gives the size in 12 months. If you enter your cloud's storage price, it adds the average cost a month.

If it does not work1 case
"The snapshot has no Photon rate for this product and machine".
Nothing is assumed. Choose another machine or another compute.

Good to know With the Data architect profile, "For you" asks "Photon: from when does it pay off?", which opens this calculation filled with its example. No cloud publishes an eviction rate. Measure it on your own runs. Method, sections 13.10 and 13.14.

Simulations · 6 min

AI: Genie and agents, build and price

Contents

A Simulations page to build a Genie Agent (formerly a Genie space) or an AI agent, step by step, and estimate what it will cost each month, in DBU and in dollars, at list price and then after your contract discounts. The prices are the Calculator's, and YALineo reads nothing in your workspace.

Before you start

  • Nothing but the extension. For a sound figure: the expected usage and, for Genie, the language model's DBUs per question, measured on a trial.
  1. Open "AI: Genie and agents". "Simulations" section, or from "Which tool for which question?". "Learn more", next to the title, opens the page's course when there is one.
  2. Choose what you build. "Genie Agent": a space where your users ask questions in everyday language on Unity Catalog tables. "AI agent": an agent that searches your documents, calls a model and tools, deployed on Databricks and evaluated with MLflow.
  3. Read "What is a DBU?". The unit Databricks bills, the formula of a bill (DBU × a DBU's price for the product × (1 − discount)) and the case of classic compute, where the cloud's machines add up, with its sources. The card folds with a click.
  4. Set the prices. "The prices": the cloud, the region (without it, nothing is priced), the plan, then your discounts on DBUs and on machines.
  5. Describe your usage. "Your usage", by groups. For Genie: the usage (users, questions a day, SQL statements per question, duration of a statement, peak), the SQL warehouse, Genie's language model. For an agent: the usage, the agent's model (with, for a proprietary model, the tokens written to the cache and their duration), search in your documents (AI Search), serving the agent, Unity AI Gateway (with, when a feature is checked, the GB of payload logged), evaluation and monitoring (MLflow). Each field says its unit and where its value comes from, and its "Help" unfolds below. A choice that makes a field useless hides it. "Back to the starting values" starts again from the example.
  6. Read the month. "What your … will cost each month": the DBUs per month (with the free DBUs), the list price, after discounts, and the cost per question or per request. Then each part, with its DBUs, its two amounts, its calculation and its source. A one-time spend, like the first embedding, shows outside the month. "How it is calculated, step by step" unfolds each step with your figures.
  7. Complete what is missing. While a value is missing, "To complete" names it, and a click leads to its field. The total is then marked partial, without the parts still waiting for a value. Lower, the assumptions to replace with your figures, those of the calculation, what stays outside it, and the method.
  8. Follow "Build step by step". On the right, the steps of the build (from scoping the need to following usage and costs), each with what it decides, its official page and its checklist to tick. The studio keeps the count of ticks while it is open.
  9. Keep it, send it. "Save and send": name the estimate, then "Add to an estimate" puts it in "My estimates". To send it: "Report (HTML)", a self-contained page to attach; "Estimate file", which your recipient's YALineo imports; "Copy the summary"; "Prepare a mail", which opens your mail app with the summary.

To check The month's total is the sum of the parts, each rounded to the cent once. A part that waits for a value is not counted, and the total says so.

If it does not work1 case
No amount shows.
Choose the region in "The prices". For Genie, also enter the language model's DBUs per question, measured on a trial. "To complete" names them.

Good to know Nothing leaves YALineo. The report and the file download on this device, and the mail leaves from your mail app if you decide so. YALineo calls neither Genie nor any model. It prices the usage you describe.

See the 2 screenshots
"AI: Genie and agents": what you build (Genie Agent or AI agent), "What is a DBU?", the prices (cloud, region, discounts) and "Build step by step" on the right.
"AI: Genie and agents": what you build (Genie Agent or AI agent), "What is a DBU?", the prices (cloud, region, discounts) and "Build step by step" on the right.
The month of a fictional Genie Agent: DBUs, list price, after discounts and per question, then each part with its calculation and its source.
The month of a fictional Genie Agent: DBUs, list price, after discounts and per question, then each part with its calculation and its source.

Simulations · 3 min

Compare scenarios and share them

Contents

Fictional case: the retailer wonders whether to move the nightly ingestion to spot or to reserve its machines for three years. You make one estimate per assumption, then compare them side by side.

  1. Duplicate the estimate. "Duplicate", at the top of the estimate, creates a copy to change without touching the original.
  2. Change one assumption. For example, open "Daily sales ingestion" and set paying for the workers to "Reserved 3 years", or switch cloud in the settings.
  3. Tick the estimates. On the "My estimates" page, tick two to four estimates. With "side by side", you get their total per month and per year, DBUs, machines, and the gap to the cheapest.
  4. Filter by cloud. "All", "Azure", "AWS" or "GCP" above the list, with the number of estimates of each, and the search by name.
  5. Share. "Export as CSV" (one line per workload, then the total), "Export as Excel" (summary, workloads with the calculation step by step, prices used, method notes), "Word" or "PDF" (the total, each workload, its whole calculation step by step, the state of the estimate), or "Copy as Markdown" (with your notes), for a ticket, an e-mail or a wiki.
  6. Or pass on the estimate itself. "Export the estimate file" gives a file another person opens with "Import an estimate". The estimate arrives with its workloads, services included, through no server at all.

To check In the CSV, the lines of the priced workloads ("priced" set to "yes") add up to the total line, to the cent.

If it does not work1 case
An estimate says "partial total" and has no gap.
Some of its workloads have no price (for example GCP machines, to enter). Its total leaves them out, so it is not ranked against the others. Fill in the prices its workloads ask for.

Good to know Write your assumptions in the estimate's notes. They go with the Markdown copy, and the reader knows where the figures come from.

See the screenshot
Two estimates side by side: total per month and per year, DBUs, machines, and the gap to the cheapest.
Two estimates side by side: total per month and per year, DBUs, machines, and the gap to the cheapest.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Simulations · 4 min

Make a measured workload vary

Contents

Start from a real job of your workspace, measured by the FinOps side, and see what it would cost with twice as many runs, half as much data again, or machines on spot.

Before you start

  • The FinOps side loaded (tutorial "Read the cost overview").
  • To price the machines: read access to system.compute.node_timeline (tutorial "Grant access"). Without it, you enter the cost of the machines per month in the workload.
  1. Open "Measured workloads". The page brings each workload of the FinOps side to a month: its DBUs, their cost, its machines' hours and what they would cost pay as you go, on spot or reserved 3 years. The columns sort.
  2. Add one. Click "Add" on its line. It joins the open estimate (or a new one), where it opens, highlighted.
  3. Move the levers. "Runs (multiply)" at 2: twice as many runs; "Data volume (multiply)" at 1.5: half as much data again. DBUs, their cost and machine hours follow.
  4. Change how the machines are paid. Workers on spot, driver pay as you go, with the share of hours rerun pay as you go if you know it; or reserved 1 or 3 years, as a new commitment or on an existing reservation. The result shows today, then the scenario.

To check With the levers at 1, the DBU cost is the FinOps one brought to an average month (365/12 days), minus your discount. With the machine hours, DBUs and hours cover the same days, and the page shows the coverage.

If it does not work3 cases
The machines say "not read".
YALineo did not read the system.compute.node_timeline table. Ask for the SELECT privilege (tutorial "Grant access"), then refresh.
The workload added is "not priced: cost of the machines unknown".
YALineo did not read its machine hours (missing access, or a Classic or Pro SQL warehouse, absent from the compute tables). Enter "Cost of the machines per month" in the workload, from the cloud bill, and the workload joins the total. It is never counted as zero.
The workload says "measured on Azure: its cost does not hold on AWS".
The estimate moved to another cloud. Go back to your workspace's cloud, or describe the workload as a new one on the other cloud.

Good to know The rule is proportional, and the page says so. A much larger volume may need another cluster size. To change the size, describe a new workload instead. "Learn more", next to the title, opens the page's course.

See the screenshot
The measured workloads: DBUs, cost, machine hours and what they would cost depending on how they are paid, with "Add" on each line.
The measured workloads: DBUs, cost, machine hours and what they would cost depending on how they are paid, with "Add" on each line.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Simulations · 3 min

Prices, discounts and the price snapshot

Contents

The Calculator prices at the public prices of a dated snapshot bundled in the extension: Microsoft for Azure, AWS and Databricks for AWS, Databricks for GCP. Your contracts come on top: discounts and negotiated prices.

  1. Open "Prices and discounts". Your default settings: cloud, region, plan, paying for machines and discounts. Each new estimate starts from them, then keeps its own.
  2. Read the snapshot. Its date, its content and its public sources. YALineo never downloads it. The snapshot changes with YALineo's versions, and an alert warns every week when public prices move.
  3. Enter your price of a DBU. In "Price of a DBU", the "Your price" column replaces the public price for that cloud, and the page shows the average price measured in your workspace as a reference.
  4. Turn the assistant on, if you want it. The assistant is off by default. Click "Turn the assistant on", then enter the name of a serving endpoint of your workspace. An estimate's page then offers "Describe a workload". The model proposes lines. YALineo checks each one and refuses, with its reason, a rate missing or not offered on this cloud. Nothing is added without your click. Your description goes through an ai_query statement to that endpoint; if it serves an external model, Databricks passes it on to its provider.
  5. Enter a machine price. "Price of a machine" lists every type, filtered by name, vCPUs and memory, with one box per way to pay. On GCP, this is where you enter machine prices.

To check An entered price shows "entered" in the step-by-step calculation of every workload that uses it.

Good to know The machine discount also applies to spot and reservations, as on a bill that discounts every meter. Leave 0 if your contract does not cover them. "Learn more", next to the title, opens the page's course.

See the screenshot
Prices and discounts: your default settings, the dated snapshot with its sources, the public price of a DBU and yours.
Prices and discounts: your default settings, the dated snapshot with its sources, the public price of a DBU and yours.
The steps on video, on the example platform (fictional data, no sound), recorded on an earlier version of the studio.

Simulations · 3 min

SKU prices, token prices, your account's prices

Contents

Two pages of the "Simulations" section for the prices themselves: the price of each SKU of your account, and the monthly cost of models served by Databricks or proprietary ones for your token volumes.

Before you start

  • Nothing but the extension for public prices; for your account's prices, read access to system.billing.list_prices.
  1. Read your account's prices. "Read my account's prices" reads, on request only, the list prices of your account (system.billing.list_prices). They then come before the public prices and stay in the page's memory. Nothing is sent before your click.
  2. Open "SKU prices". Each SKU with its unit, its list price, its effective price, its date and its origin. Search a SKU, filter by unit or by family, enter a quantity to get quantity × price, to the cent, then "Export as CSV". Without access to the billing, these are the public prices, and the page says so.
  3. Open "Token prices". Choose the cloud, the region and the kind of rate (per token, priority, batch, provisioned throughput), enter your tokens per month in millions (input, output, cache reads and writes), then tick the models.
  4. Read the table. For each model ticked: the price per million tokens, the DBUs per month, the price of a DBU with its origin, per month and per year, from cheapest to dearest. The page flags a price not published for this cloud, and never counts it as zero.
  5. Add it to an estimate. "Add to the estimate" puts the model in the open estimate of the same cloud, or in a new "Compared models" estimate.

To check The cheapest model carries "the cheapest", and each line says its state ("priced" or what is missing).

Good to know The services snapshot is dated 26 October 2026, drawn from the pricing pages of Databricks and Microsoft Learn. Method, sections 13.11 to 13.13. Each page has its course, behind "Learn more", next to the title.

See the screenshot
"Token prices": two models for 100 million input tokens and 20 million output tokens a month, sorted, each price with its origin.
"Token prices": two models for 100 million input tokens and 20 million output tokens a month, sorted, each price with its origin.

Going further · 3 min

Grant access (for admins)

Contents

System tables are often admin-only. A metastore admin grants read access to a group, once.

Before you start

  • Being a metastore admin, or the owner of the system catalog.
  1. Create or pick a group. For example "yalineo-users", in the Databricks account console.
  2. Run the SQL below in the SQL editor. Or the full script shipped with YALineo, which adds job names, job runs and query history (for cost by compute time).
  3. Tell the users. They reload their map, with the full lineage, then the FinOps side. "Check lineage access", in the settings, confirms it.
GRANT USE CATALOG ON CATALOG system TO `yalineo-users`;
-- Lineage
GRANT USE SCHEMA ON SCHEMA system.access TO `yalineo-users`;
GRANT SELECT ON TABLE system.access.table_lineage TO `yalineo-users`;
GRANT SELECT ON TABLE system.access.column_lineage TO `yalineo-users`;
-- FinOps
GRANT USE SCHEMA ON SCHEMA system.billing TO `yalineo-users`;
GRANT SELECT ON TABLE system.billing.usage TO `yalineo-users`;
GRANT SELECT ON TABLE system.billing.list_prices TO `yalineo-users`;
-- Calculateur / Calculator (facultatif / optional)
GRANT USE SCHEMA ON SCHEMA system.compute TO `yalineo-users`;
GRANT SELECT ON TABLE system.compute.node_timeline TO `yalineo-users`;

To check A user of the group sees "Access to system.access.table_lineage confirmed." in their settings.

Good to know Everything is read only. Each block is independent. Open only what you want (lineage alone, or costs too).

Going further · 2 min

Without system tables: analyst mode

Contents

Without access to the system tables, YALineo still draws the map, with the Unity Catalog lineage API, the one behind the Lineage tab of Catalog Explorer.

Before you start

  • BROWSE on the catalogs to draw.
  1. Settings, "Lineage source": "Analyst mode". "Check lineage access" confirms the lineage API answers.
  2. Choose the "Analyst-mode map". "The whole map on loading" (one request per table, four at a time, recommended) or "Only opened and followed tables" for a very large workspace.
  3. Save and load. Tables you cannot see stay hidden.

To check The map is drawn with the tables you can see; opening a table loads the lineage around it.

Good to know Analyst mode gives no activity (reads, delays) and no costs. For those, ask for the grants above.

Going further · 3 min

Large platforms: a lighter load

Contents

On a platform with hundreds of thousands of links, choose what YALineo reads before loading. The map then stays complete and fast.

Before you start

  • The connection tested and a SQL warehouse chosen.
  1. Settings, step 3: "Estimate the volume". One read-only query counts, catalog by catalog, the links and tables your settings would read, without loading or saving anything.
  2. Read the verdict. Light; heavy (a few minutes, tens of MB, but complete); or beyond the limits (250,000 rows per query, beyond which the rest would be cut).
  3. Follow "To lighten it". The heaviest catalogs and their share, the link window that fits (90 or 30 days, with the estimated number of links), this workspace's catalogs only, then analyst mode. Each tip has an "Adjust" button that opens the right setting.
  4. After a load. If a result reached its limit or its time-out, "Incomplete load" shows in the status bar of the panel and the studio, with the same tips.

To check The verdict turns "light" or "heavy" after your settings.

Good to know A larger or serverless SQL warehouse reads system tables faster, useful when the 15-minute time-out is reached. Filters reduce what is read.

Going further · 2 min

Keep or erase your data

Contents

YALineo computes and keeps the map and the costs in your browser. You choose for how long, and you erase everything in one click.

  1. Choose retention. Settings, step 5: "Keep nothing after the browser closes" (default) or "Keep the map and its history on this device" (to compare versions and follow the map over time).
  2. Erase right away. "Erase now", at the bottom of every studio screen, erases the map from this device.
  3. Erase everything. Settings, "Erase everything on this device": maps, glossary, description validations, followed tables.
  4. Forget a workspace. Settings, "Configured workspaces", "Forget" (click twice to confirm) erases that workspace's settings, token and maps.
  5. Pause. The switch in the chip's card stops YALineo on Databricks pages without erasing anything.

To check The strip at the bottom of the studio says what is kept: "Nothing is kept after the browser closes", or "kept on this device until you erase it".

Good to know No YALineo server receives your metadata, so erasing on your device erases everything. For your security teams, "Security and evidence" (in the privacy strip and the health check) opens the site's page on what the extension is allowed to do, what it never does and how it is tested. In the studio, "Diagnostics" checks at once the read-only guard and that the network is limited to Databricks ("Troubleshooting" tutorial).

Going further · 3 min

Pre-flight in pull requests

Contents

In your continuous integration, the pre-flight gives the tables a pull request touches, what depends on them, who to warn and the rules broken.

Before you start

  • A read-only Databricks service principal, and access to system.access.
  1. Add three variables to your CI. DATABRICKS_HOST, DATABRICKS_TOKEN (the service principal's token) and DATABRICKS_WAREHOUSE_ID.
  2. Run the tool on the changed files. Example below: it writes a Markdown comment, to post in the pull request.
  3. Block if needed. With the "--fail-on-high" option, the check fails if a touched table breaks an "Act" rule.
git diff --name-only origin/main...HEAD \
  | node cli/yalineo-preflight.mjs --files-from - --out preflight.md

To check The preflight.md file lists the touched tables, their downstream and their owners.

Good to know The pre-flight runs in your CI, with your secrets. No YALineo server is involved.

Going further · 4 min

Troubleshooting

Contents

This tutorial starts from the Diagnostics badge, then gives the most frequent cases and what to do.

  1. Look at the Diagnostics badge. Studio, "Diagnostics": at the top, a green badge "YALineo is secure and healthy" shows when every check passes. YALineo verifies them at once on your device: the read-only guard refuses a write and lets a read through; the extension's network permissions only go to the Databricks domains; the last reading of the map worked; no error in the last 24 hours; the map is less than 7 days old. Orange, "Worth a look": security is verified, and the health point in cause is named. Red, "Security check failed": stop reading the map, and write to support with the copied diagnostic.
  2. "The workspace refuses the session". Under the message, the "Reload the Databricks tab and try again" button reloads the workspace tab then tries again, in one click. YALineo never reloads without that click, because a reload drops what the page had not saved. The message says if the tab went to a sign-in page on another site. In that case, sign in then try again; otherwise, switch to a personal token.
  3. "your account cannot read system.access.table_lineage" or "lineage tables cannot be found". Ask for the grants ("Grant access" tutorial) or switch to analyst mode. On Databricks Free Edition, the system schema may be disabled.
  4. "Billing is not readable". Read access to system.billing is missing. The lineage map still loads.
  5. The chip does not show. YALineo adds it to the open tabs, at install and at every update. Check in the settings that it is not "Hidden".
  6. "Incomplete load". Follow the tips shown ("Large platforms" tutorial).
  7. A table is missing from the map. It is out of scope (chosen catalogs), you are not allowed to see it, or it had no lineage event over the history loaded.
  8. temp/ paths followed by a number. These are temporary working folders written or read by a job, outside Unity Catalog. The map gathers them in the "External sources" station. Untick "Loads" and "Exports" to hide them.
  9. A notebook shown by its number. YALineo finds names in your folders (/Users/you, /Shared, /Repos/you). A notebook kept elsewhere, or in a folder you cannot list, keeps its number.
  10. For everything else. Studio, "Diagnostics": the duration of each step and the last errors; the names YALineo recognizes are masked there, with no guarantee that all are. "Write to support" prepares an e-mail with this diagnosis.

Good to know In the support diagnosis, YALineo masks the names it recognizes, with no guarantee that all are (see the privacy policy). You read it before sending.

See the screenshot
"Diagnostics": the badge, orange here because an error of the last 24 hours is recorded, and each check with its result.
"Diagnostics": the badge, orange here because an error of the last 24 hours is recorded, and each check with its result.

Going further · 2 min

Under the hood

Contents

YALineo is a browser extension, with no server. Everything is computed in your browser.

  1. Three screens, one extension. The chip, the panel and the studio are pages of the extension, in your browser.
  2. Reads to your workspace only. YALineo reads with your session (from the Databricks tab) or your token. The reads go through a closed list of API paths and are only SELECT statements, which YALineo checks before each send.
  3. Computed on your machine. YALineo computes the map and the costs on your machine and keeps them in the extension's storage. They are gone when the browser closes (a map too big for session memory, at the browser's next start), unless you choose to keep them.

Good to know No YALineo server receives your metadata, because there is none.

Your browser: YALineo extension (Chip, Panel, Studio, Local storage). SELECT, with your permissions. Your Databricks workspace: API (allowed paths), SQL warehouse, System tables, Lineage API. YALineo server: does not exist.
YALineo server: does not exist

YALineo's words

Line
A data domain (sales, stock, finance…), drawn in one color. YALineo reads it from the domain tag, the catalog, the schema or the table prefix, as set in "Map lines detected by".
Section
One of the six parts of the studio's side bar: Home, Data and lineage, Costs, Simulations, To review, Learn. A click opens its main page, the chevron unfolds it.
Station
A schema, with its number of tables. The "External sources", "Jobs", "Dashboards", "Notebooks" and "SQL queries" stations gather what is not a table; on the map by schemas, the paths a table writes form the "External destinations" station.
Band
A zone of the map: source, bronze, silver, gold, platinum, consumption (medallion layers), or a catalog, as you choose.
Upstream, downstream
A table's upstream is what feeds it. Its downstream is what it feeds: other tables, dashboards, notebooks, queries.
Direct link
A table read or written directly by another, with no step in between. This is what the Lineage tab of Catalog Explorer counts. A link that only exists through a view, or around a table deleted or unreadable, is not direct. "The whole lineage" adds the next steps, one after the other, and YALineo gives the figures of the whole chain separately.
Bridged link
Two observed links put end to end around a table deleted since, or unreadable with your permissions. YALineo marks it "bridged", with its reason, and gives its figures separately, never adding them to what was observed.
Sheet
The page of an asset (table, view, job, pipeline, SQL warehouse, notebook, dashboard, query): Overview, Columns, Lineage, Usage and costs, History.
Finding
A situation to check on an asset (freshness, failures, costs, usage, sensitive data, documentation, architecture), with the possible effect, the evidence and an action. Your follow-up: "To review", "Checked by me" or "Ignored".
External path
A file or folder in cloud storage, outside Unity Catalog, read or written by a job: a drop of source files, an export, or a temporary working folder (often named temp/ followed by a number). The map shows its last two segments; its card gives the full path.
Business term
A word of the business (stock on hand, average basket) with its definition, synonyms, domain and owner, tied to the tables and columns that carry it. It stays "Proposed" until someone validates it.
Refresh
What writing a table costs over the period: the share of the jobs, pipelines and notebooks that write it.
Nature of an amount
Observed: read as is from the billing. Attributed: a share of an observed spend, given by a written rule. Estimated: computed from an assumption. Potential: a possible saving, to confirm before acting.
Cost precision
From the surest: "billed" (the bill names the table, a materialized view for example); "single table" (the only table written by its job's runs); "compute time" and "even split" (a run that writes several tables, shared by compute time or evenly); "warehouse" (a share of a SQL warehouse, hour by hour); "period" (no run linked: shared over the period).
Exact, split
The precision of a catalog's or a schema's cost. Exact: all the spend goes to it, to the cent, because every workload that wrote its tables wrote only there, or because a table is billed in its own name. Split: a share comes from a workload that also writes elsewhere, or from a shared SQL warehouse, and follows the table rules.
"Why?"
The button next to a missing link, or to a cost that is zero, missing or partial. It gives the measured reason and the action that changes the result; it adds no figure.
Previous period
The days just before the chosen period, of the same length. Every FinOps card compares its figures with those days.
List price
Databricks' public price, before your discount. You set your discount in the settings.
DBU
Databricks Unit: the unit of compute Databricks bills, multiplied by each product's list price.
Genie
Three products, named as the billing names them (field usage_metadata.genie.surface): Genie Code (formerly Databricks Assistant), Genie One (formerly Databricks One) and Genie Agents (formerly Genie spaces). YALineo shows free usage (GENIE_FREE_USAGE) separately.
Projection
What the same usage would cost over the coming days if it kept its pace: the pace of the last 7 days, with the spread up to the pace of the whole period (two measured paces, not a probability). It is a derived figure, never a measure.
Estimate
A set of workloads on one cloud, region and Databricks plan, with your discounts: what the Calculator prices, per month and per year. It is an estimate, never a measure.
Paying for machines
Pay as you go (full price), spot (cheaper, but the cloud can take the machine back; the driver stays pay as you go), reserved 1 or 3 years. A reservation counts as a new commitment (730 hours a month per machine, used or not) or as an existing reservation (the hours used); for spot, you can enter a share of hours rerun pay as you go.
Price snapshot
The public prices of the clouds and of Databricks on a date, bundled in the extension and never downloaded.
Service
In an estimate, a workload other than a job, a pipeline, a warehouse or a cluster: Model Serving, AI Search, models served per token, Lakebase, Databricks Apps and the other kinds of the "Service" menu, each priced with its sourced rates.
Capability
The family of a "Size and tune" calculation: data engineering, data science, analytics, dashboards or architecture. Each calculation also has a kind: cost, size, setting or buying.
Profile
The profile of use picked in the settings, at the "Your profile" step: "Full" (the default, everything included) or one of the seven roles. A role puts its pages and calculations first; a page outside the profile keeps its place in the menu with a padlock and does not open. It helps you focus. Anyone can change it, and YALineo still reads the data with your own permissions.
Course
What "Learn more" opens, next to the title of a Simulations page or a calculation: what the tool is for, the notions, the formula term by term, how to measure your own figures, read-only queries to copy, the pitfalls and the official sources, each with the date it was read.
Path
In the "Learn" section, the series of modules of a profile, each ending with a quiz; "The whole course" gathers them all. YALineo keeps your progress on this device.
Shortcut
A page, a calculation or an action the palette (Ctrl K) shows first. You choose and order them in the settings, "Ctrl K shortcuts" card, or with a right click.