test(#834): rate-limit writers in TestMoveWithVarcharPK

[morgo]

Summary

Fixes #834 — TestMoveWithVarcharPK timing out at 10 minutes.

The test runs 4 writer goroutines in an uncapped tight loop, each doing INSERT + UPDATE + (sometimes) DELETE per iteration with no delay. On slow CI runners this generates binlog faster than the move's reader can drain it. Inside Flush() (pkg/repl/client.go:1000) the loop never sees GetDeltaLen() < binlogTrivialThreshold because the source position keeps outrunning the buffered position. move.Run() never returns → writers never stop → wg.Wait() hangs → the 10-minute Go test timeout fires.

The MySQL 8.4 GA failure earlier today on main showed the same symptom on the other subtest (queue-full-time vs the force-enable-buffered-map one in #834), with the reader ~460 binlog files behind the live source — same root cause.

Fix

Add time.Sleep(10 * time.Millisecond) between writer iterations. Caps total throughput at ~400 ops/sec (~1200 statements/sec). The test exists to verify FIFO-queue replay correctness for VARCHAR PKs, not raw throughput, so pacing doesn't undermine its purpose.

Test plan

• CI passes (both subtests on all MySQL versions, especially 8.4 GA which already flaked once today)
• Spot-check locally that the test still meaningfully exercises concurrent writes (writeCount > 0 in the t.Logf output)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Reduces flakiness/timeouts in TestMoveWithVarcharPK by pacing the concurrent writer goroutines so they can’t continuously outproduce the move’s binlog reader on slow CI runners.

Changes:

• Adds an explanatory comment describing the root cause of the timeout (issue #834) and why pacing is acceptable for this test.
• Inserts a time.Sleep(10 * time.Millisecond) in the writer loop to cap write throughput and allow Flush() to converge.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.