Databricks StrawberryKnow which overnight job failed before you sit down
At 09:00 the failed task is named, with its error and timestamp, and you already know whether the dashboard everyone opens is serving Friday’s numbers. Reads cost nothing.
The pipeline runs at 02:00 and somebody checks it at 09:00. Usually it is fine. The interesting mornings are the ones where task three failed at 02:14, the retry also failed, and the dashboard everyone opens at standup is showing Friday’s numbers with no sign that it is.
A companion has that 09:00 pass written up before anybody sits down: the failed task, the error, the timestamp, and whether anything downstream is stale. It works through an account you connect, and the actions it uses matter because in Databricks the difference between reading and doing is measured in dollars per hour.
| The workspace UI | A Strawberry companion | |
|---|---|---|
| Seeing which overnight runs failed | Open the Jobs page and scan | Summarised with error and timestamp |
| Knowing whether the dashboard is stale | Open the dashboard separately | Checked in the same pass |
| Re-running a failed job | Two clicks | Only if enabled, and it asks first |
| Naming the owner and opening the ticket | You do both | Same pass, from run and commit history |
| Writing the note that tells the team | You write it | Drafted from the run history |
Why does the read and write split matter here?
Because compute is the product.
Checking whether a job succeeded costs nothing. Starting that job spins up a cluster, and a cluster that starts because an agent was being helpful runs until something stops it. On generous instance types that is a mistake you notice on an invoice.
So read the Available tools list on the Databricks entry under `strawberry://settings/integrations`, which names every tool the connection exposes and marks the read-only ones, then use the Permissions block to keep anything that triggers a run at Ask every time. If nobody should be starting jobs from a chat window, switch those actions off.
- Run status, job history and failure detail: safe reads, genuinely useful every morning.
- Triggering a job or starting a cluster: Ask every time, and only for people who own the budget.
- A service principal scoped to the specific jobs, rather than a workspace admin token.
- Cluster policies and auto-termination configured in Databricks, because that is where they bind.
What is the morning pass worth?
The difference between finding out at 09:00 and finding out when a stakeholder asks.
A companion reads overnight run status, names the tasks that failed with their error and timestamp, then checks whether the downstream dashboard is stale, which is the fact people actually care about.
That write-up is operations work with a data flavour, and on a cadence it stops a silent pipeline failure turning into a bad decision. If the audience is engineering, weekly engineering metrics is the neighbouring shape; if it is the business, weekly reporting is.
What the connection covers
Run visibility and the write-up that follows: status, history, failure detail, staleness downstream, and a note somebody can act on. Notebooks and code stay in the workspace, where the schema, the lineage and the reviews already live.
Unity Catalog governance, cluster policies, permissions and cost controls also stay where they are. Those settings make a workspace safe for anyone to use, agents included, and this connection benefits from every one of them.
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents