github-actions[bot] commented on code in PR #68434:
URL: https://github.com/apache/doris/pull/68434#discussion_r4089466300
##########
regression-test/suites/fault_injection_p0/cloud/test_tablet_state_change_in_publish_phase.groovy:
##########
@@ -90,6 +90,9 @@ suite("test_tablet_state_change_in_publish_phase", "docker") {
// block FE's task report handler to avoid alter task re-sended to
BE before we enable debug points for SC
GetDebugPoint().enableDebugPointForAllFEs("ReportHandler.block")
cluster.startBackends(beIndex)
+ // Heartbeat/HTTP readiness does not imply the peer BE's brpc
connection has recovered.
+ // Wait longer than brpc's default 3s health-check interval before
starting load 1.
Review Comment:
The added delay still does not wait for the resource that caused the
failure: `startBackends()` checks FE membership and the restarted BE's HTTP
port, while the surviving BE's cached brpc channel recovers asynchronously. In
the vendored brpc 1.4.0 path, the first check can be delayed by circuit-breaker
isolation (up to 30s after repeated failures) and then waits without a deadline
for socket references to drain; the 3s value governs only subsequent retries.
Load 1 can therefore remain the first real peer RPC and hit the same
`EHOSTDOWN`. Please keep `ReportHandler.block` enabled and use a bounded
`awaitUntil` to GET the actual live-BE-to-restarted-BE path via
`/api/check_rpc_channel/{ip}/{port}/{payload_size}`, validating its success
response, before installing the two BE debug points and unblocking reporting.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
