[adapters] Increase wait_for_records timeout in tests.#6614
Conversation
suspend_barrier5 times out waiting for 4000 input records while suspending in CI. The log seems to indicate that the connector is making progress, just slowly under load. We increase the timeout from 10 to 100s and improve progress logging. If the issue occurs again, this means there's a real bug somewhere. Signed-off-by: Leonid Ryzhyk <ryzhyk@gmail.com>
mythical-fred
left a comment
There was a problem hiding this comment.
Test-only, targeted, well-commented. Splits the wait budget into two named constants (OUTPUT_TIMEOUT_MS=10s / SUSPEND_OUTPUT_TIMEOUT_MS=100s), threads a new wait_for_records_within helper into the two suspend paths that legitimately need the longer budget, and adds a tprintln! macro to timestamp progress lines so future flake diagnosis is easier. The 100s budget matches the existing suspend-completion recv_timeout in the same file — good symmetry. LGTM.
Nit (non-blocking): the docstring on SUSPEND_OUTPUT_TIMEOUT_MS explains why the longer budget is needed better than the two inline // After the first iteration a suspend is pending... comments — those two site comments now duplicate what's on the constant. Could just say // Use SUSPEND_OUTPUT_TIMEOUT_MS: suspend pending, see constant doc. No action needed.
suspend_barrier5 times out waiting for 4000 input records while suspending in CI. The log seems to indicate that the connector is making progress, just slowly under load. We increase the timeout from 10 to 100s and improve progress logging. If the issue occurs again, this means there's a real bug somewhere.