Common issues

Most problems in Databasin resolve with one of about a dozen fixes. This is the cheat sheet. If yours isn't here, jump to Getting support.

"I can't log in"

"Access Denied" or "User Not Invited"

You're authenticated with Azure AD, but no Databasin org matches your email. Either:

• Your org admin hasn't invited you yet. Ping them.
• You're signing in with the wrong account (personal vs. work). Try a different account.

"Account Not Found"

You're authenticated but not part of any org. On SaaS, you can self-sign up from the landing page; on self-install, ask the admin.

Nothing happens when I click Sign In

Usually a popup blocker. Check your browser, allow popups for the Databasin domain, and retry.

"My pipeline is failing"

A run that ends in Failed — you'll see it next to Running and Successful in the project's Command Center — almost always names the reason in its logs. The usual culprits:

"Authentication failed"

The connector's credentials changed upstream. Open the connection, re-enter creds, and hit Test connection.

"Cannot find table"

The source schema changed — column renamed, table dropped. Open the pipeline, re-pick the objects, and save.

"Target is not writable"

The target connector's user lost write permission. Fix at the source, re-test the connection.

"Run duration suddenly doubled"

Three likely causes:

• Source has more data (new backfill, big import).
• Source is slower (load on source side, network).
• Target is slower (target cluster is contended, resize or wait).
• A Delta pipeline lost its watermark and re-pulled from scratch — effectively a Historical-style backfill. Check the pipeline's ingestion mode and its change marker.

Pipeline run logs show where time is going.

"A dependent route suddenly takes hours longer"

Expected, if the connector was recently upgraded. Routes that fan out one call per parent record used to stop early against vendors that cap or misreport their totals — so the run was fast because it was under-reading. Cap recovery makes the fanout complete, and complete costs time. See Native connectors in pipelines.

"The run log shows a warning about duplicate parent ids"

That's the dependent-fetch integrity guard doing its job: it spotted a parent listing shifting under it mid-page, re-fetched, and carried on. The run recovered. If it happens on every run, mention it in a support ticket.

"A connector fix shipped but my pipeline didn't change"

Some route settings are copied onto a pipeline's tables when the pipeline is created, so an existing pipeline can keep the old behavior after the connector is upgraded. Open the affected table in the pipeline and re-save it. Re-saving the connector alone doesn't always push the change down.

"My SQL query is slow"

The workspace was asleep

Expected — a sleeping workspace never starts itself. Click Start workspace in the editor toolbar, give it about a minute, and queries run at normal speed from there.

Predicate pushdown isn't happening

Check EXPLAIN output. Often solved by pushing filters into CTEs near the scan, or explicitly casting types so the optimizer can push down.

Wide SELECT on a big table

Always project only columns you need. SELECT * on a 200-column Iceberg table is almost never what you want.

"My automation task won't save"

Several task types refuse to save when a required connection or field is missing, rather than letting the run fail later on a schedule. The error names the field. The usual ones:

• File Drop / Unzip File — needs both a source and a target connection.
• Automation — needs a child automation selected.
• OCR — needs the document, LLM, and lakehouse connections, plus at least one declared field.

See The task types.

"My notebook lost work"

This class of bug was fixed — typed text now reaches storage, autosave is on, and the Object Explorer no longer overwrites the cell you're editing. If you're on an older session, reload the page first. Version history (Save to server, then Version history) is the safety net for anything important. See Notebooks.

"My scheduled notebook says it succeeded but produced nothing"

Scheduled notebook runs used to report unfinished and cancelled runs as successes. That's fixed. Open the notebook's run panel and check Recent runs for the per-cell detail — it'll show where the run actually stopped.

"Databasin One isn't finding my data"

"Query budget reached"

You hit the 3-query budget — three execute_sql calls — for this conversation. Lighter tools, like sampling a table, don't count against it. Start a new conversation and the budget resets.

"Not seeing the tables I expect"

Check the data context picker in the left sidebar. Only tables in scope are visible to the agent. Add the right warehouse or schema.

"It's using the wrong table"

Provide the table name explicitly: "Use the orders_daily table, not orders." The agent will respect that.

"My chart looks wrong"

Numbers aren't formatting

Ask for the formatting explicitly: "Format the y-axis as currency." Databasin One will rebuild the chart with updated config.

Wrong chart type

Click Chart this to see 2–3 options, or ask for a specific type: "Make this a stacked bar chart."

Dashboard tile looks different from chat

Shouldn't happen anymore — the Canvas uses the exact same SimpleChart config. If it does happen, open support with a link to the dashboard.

"I got a 'Credits Exhausted' overlay"

The org used up its credit balance, so Databasin shows a full-screen, non-dismissible Credits Exhausted overlay with a fixed notice that you have 24 hours to upgrade before the account is disabled. Options:

• If you're an admin, click Upgrade Now to set up metered billing through Stripe.
• If you're not an admin, ping your org admin — only they can upgrade. Sign Out is always available to everyone.

See Credits and billing.

"A dashboard I published is empty"

Usually one of:

• The viewer doesn't have access to the underlying data — published dashboards re-run SQL with viewer permissions.
• The connection the SQL was saved against has been deleted. Duplicate, re-run, re-save.
• The SQL returns zero rows today (time-bounded query, nothing in the window).

"Something is clearly broken and I need help"

Click the support icon (the headset) in the top bar and open a New Support Ticket. Pick a type (Support / Bug Report / Enhancement), write a clear subject, and include as much detail as you can in the description. See Getting support for what makes a strong description.

Where to go next