[alexluong]

Run the logmq tests on a fake clock. Based on #1162.

The logmq tests waited real time for batch delays and emit timeouts: fixed 200 ms sleeps for a batch to flush, and require.Eventually polls for sends, evaluations and acks to land. Now they run inside a synctest bubble, with Redis from testutil.NewSynctestRedis where they need it, so those waits cost nothing and the checks no longer depend on CPU load.

The package now runs in under 1 s.

1. batchprocessor_test.go: the tests run in a bubble, except TestMessageHandler_NacksAfterShutdown, which has no waits. The 200 ms sleeps that waited for a flush become synctest.Wait().
2. The characterization harness, the tests using it and the suppression window tests run in a bubble with an in-memory Redis. require.Eventually polls become synctest.Wait() plus a direct assert.
   • waitTerminal steps the fake clock 1 ms at a time until every message is acked or nacked, so batch delays and emit timeouts elapse without real waiting.
   • waitBlocked returns once a send is parked on the blocking sink, with no poll.

Assertions are kept; some are stricter. A flush asserted after synctest.Wait() has to happen without the clock moving, where the 200 ms sleep allowed up to 200 ms. TestRedisSuppressionWindow_InFlightCallerSkipsWithoutWaiting asserts zero elapsed time instead of under 100 ms.

The package was stress-tested with repeated runs, plain, with -race, and with -cpu=1,2 and -shuffle=on.

🤖 Generated with Claude Code

[leggetter]

Nit: the old Eventually here had the message "the slow delivery should reach the sink". require.GreaterOrEqual(t, h.sink.inflightSends(), 1, "...") would keep it and print the actual count on failure. Same for :88 and characterization_postprocess_test.go:39, which also lost their messages, and the other require.True(... >= n) checks would get the count too.

[alexluong]

Thanks, updated. All six now use GreaterOrEqual, with the old messages restored where they had one.