Commit b660184
committed
test(poa): use PING readiness probe for redis test server startup
`wait_for_redis_ready` was checking only TCP connectability, which
succeeds as soon as redis-server's listener binds — before the
process is ready to handle commands. On a cold start this is fast
enough to mask, but the restart-and-lose-data test
(`unreconciled_blocks__when_redis_node_restarts_and_loses_data__drops_block_below_quorum`)
hit the gap consistently enough to flake locally and in CI:
adapter_b's first command on the restarted node would fail because
redis was still warming up.
Same class of half-ready issue as the 2026-04-22 mainnet hang —
TCP accept doesn't imply protocol readiness. Replace the TCP probe
with a real PING/PONG round-trip so we only return once redis is
actually serving commands.
Stress: previously flaky test now passes 20/20 in isolation and
the full leader_lock suite passes 5/5.1 parent 1891e64 commit b660184
1 file changed
Lines changed: 12 additions & 3 deletions
Lines changed: 12 additions & 3 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
1650 | 1650 | | |
1651 | 1651 | | |
1652 | 1652 | | |
1653 | | - | |
1654 | 1653 | | |
1655 | 1654 | | |
1656 | 1655 | | |
| |||
3406 | 3405 | | |
3407 | 3406 | | |
3408 | 3407 | | |
| 3408 | + | |
| 3409 | + | |
| 3410 | + | |
| 3411 | + | |
| 3412 | + | |
3409 | 3413 | | |
3410 | | - | |
| 3414 | + | |
3411 | 3415 | | |
3412 | 3416 | | |
| 3417 | + | |
3413 | 3418 | | |
3414 | | - | |
| 3419 | + | |
| 3420 | + | |
| 3421 | + | |
| 3422 | + | |
| 3423 | + | |
3415 | 3424 | | |
3416 | 3425 | | |
3417 | 3426 | | |
| |||
0 commit comments