Question 1

What causes the 'Function[like] pattern argument must be a literal' error when using LIKE with UPPER() in Apache Druid SQL?

Accepted Answer

This error occurs because Druid's SQL optimizer (based on Apache Calcite) requires the pattern argument of LIKE to be a string literal (constant) during query planning. When you apply functions like UPPER() directly to the pattern, the planner cannot reduce it to a literal at compile time, causing the error. A workaround is to pre-evaluate the function by using a Common Table Expression (CTE) or subquery, forcing the function result to be computed before the LIKE filter is applied. Example: WITH t AS (SELECT 'a string' AS col), t2 AS (SELECT col, UPPER(col) AS upper_col FROM t) SELECT * FROM t2 WHERE upper_col LIKE UPPER('a%'). Alternatively, ensure the pattern is a literal constant by moving the UPPER() call outside if possible. This is a known limitation in Druid versions up to at least 30.0.0, and may be addressed in future releases.

Question 2

How to fix the flaky SeekableStreamSupervisorStateTest.testSupervisorStopTaskGroupEarly failure with TaskQueue.shutdown expected 1 but actual 0?

Accepted Answer

The test is racy: it assumes `getNoticesQueueSize() == 0` means the handoff notice has been fully handled, but the notice thread calls `poll()` before `handle()`. The queue can therefore be empty before `handoffEarly` is set, causing the second `runInternal()` call to miss `TaskQueue.shutdown("id1", ...)`. Fix the test by overriding `emitNoticeProcessTime` in `TestSeekableStreamSupervisor` to count down a `CountDownLatch` when `handoff_task_group_notice` is processed, then replace the `while (getNoticesQueueSize() > 0) sleep(...)` wait with `latch.await(timeout, TimeUnit.SECONDS)`. This ensures the test waits for notice completion, not just queue removal. The fix was implemented in PR #20453.

Question 3

How to fix the flaky CompactionTaskTest.testCompactionWithTimestampDimension failure where Broker sys.segments returns an empty result?

Accepted Answer

The failure is caused by a race condition: the test waits for one Broker `segment/metadataCache/sync/time` event and then immediately queries `sys.segments`, but that sync can occur before the compaction result becomes visible. Update to the fix in PR #20320, which changes `verifyNumVisibleSegmentsIs` to poll `sys.segments` until the expected segment count is returned or a bounded timeout is reached. If applying the fix manually, replace the single metadata-cache event wait and one-time assertion with a retry loop around the Broker `sys.segments` count query.

Question 4

How to fix IngestionDockerTest.test_runIndexTask_andKillData timing out while waiting for segment/schemaCache/dataSource/removed?

Accepted Answer

This flaky test fails because it waits 240 seconds for a single Broker `segment/schemaCache/dataSource/removed` metric event; a missed or delayed background refresh causes a timeout. The fix is merged in PR #20291: emit `segment/schemaCache/dataSource/removed` from the Broker's last-segment callback instead of relying on background refresh. Update to a master build from 2026-09-26 or later, or backport/cherry-pick #20291. If the failure recurs, replace the single metric-event wait with polling of the Broker schema or `sys.segments`.

Question 5

Why does the web-console MSQ e2e test time out waiting for 'div.result-table-pane'?

Accepted Answer

The test fails because the workbench displays the capacity warning dialog when a compaction task from `auto-compaction.spec.ts` is still occupying one of two task slots. `WorkbenchOverview.runQuery` waits for query results without dismissing the dialog, causing the 240-second `waitForSelector('div.result-table-pane')` timeout. Fix it in `web-console/e2e-tests/component/workbench/overview.ts`: when the `CapacityAlert` appears during `runQuery`, click the `Run it anyway` button before waiting for the result table. As an alternative, make `auto-compaction.spec.ts` remove its compaction config and wait for running compaction tasks to finish before exiting. This was resolved in PR #20443.

Apache Druid

Core Features

What It Can't Do

Use Cases

Tags

Getting Started

Similar Projects