Fix absoluteTimeoutFiresEvenIfCallsKeepArriving flake: stop debouncing before the key is freed

[devhawk]

Closes #596.

DebouncerTest.absoluteTimeoutFiresEvenIfCallsKeepArriving flaked with expected: <1> but was: <2> on the callCount assertion. The debouncer behaves as documented; the test raced against it.

Cause

The test kept calling debounce while callCount == 0.

When the debounce deadline passes, the delayed-workflow sweep moves the row to ENQUEUED and clears its deduplication ID, which frees the key. The workflow only runs later, when the queue poller dequeues it, so callCount stays 0 in between. A call in that gap started a second workflow, as the Debouncer Javadoc says it should.

Passing runs create that second workflow too. They pass only because the assertion runs before it does. This run's getResult took about a second to return, and the second workflow ran in that time.

Fix

The test now stops at the first call whose own 800 ms period would end past the deadline:

• Every call lands while the row is DELAYED. That call starts about 650 ms before the deadline, and the row holds its key until the first sweep after the deadline, so no call can start a second workflow.
• The deadline decides when it runs. The call's uncapped delay is always past the deadline, so it is always capped. The test asserts the row's stored delay_until_epoch_ms equals debounce_deadline_epoch_ms. This checks the cap directly, with no assertion on timing.
• Every call bounces the first workflow. Each call asserts it got that workflow's ID back. A call that missed the key now fails there with a clear message, not as callCount == 2.
• One run with the last argument. The workflow runs once, with the last call's argument ("result:" + lastArg). The old test checked only startsWith("result:").

The test is the same on release/v1.2, so this should be cherry-picked there.

Testing

• DebouncerTest (52 tests) passes on Postgres.
• absoluteTimeoutFiresEvenIfCallsKeepArriving passed 5 times in a row with --rerun.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

The loop can still call debounce after the deadline because its cutoff is checked only afterward.

Review effort: Balanced
Findings: 1

Open (1)

• Check deadline before debounce to prevent post-sweep key reuse · New

What changed in this PR

Updates the debouncer timeout test to avoid creating a second workflow after key release.

Changes:

• Bounds debounce calls by the stored deadline.
• Verifies workflow identity, final arguments, and deadline clamping.

File	Description
DebouncerTest.java	Strengthens absolute-timeout behavior assertions.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.