bitflicker64 commented on issue #3124: URL: https://github.com/apache/hugegraph/issues/3124#issuecomment-5130588848
The full Store-replacement lifecycle has now been run against the audited `master` baseline plus #3126 and #3128. I added the exact evidence and corrected root-cause conclusion to the issue description. The important result is that the finite DNS TTL from #3126 is necessary but not sufficient: PD/Raft recovered, pod-local DNS resolved the replacement IP, and the new Store was reachable, while the existing Server still used the removed IP through `AbstractGrpcClient`'s retained channel and stub pools. A shorter connect timeout only made the same failure return sooner (`PASS=73 FAIL=1`). With address-change detection, atomic channel replacement, stub rebinding, and the required concurrency guards, the same deployment recovered without a Server restart: focused client tests `6/0/0`, Store replacement `PASS=90 FAIL=0`, final cluster `PASS=61 FAIL=0`, and dataset `PASS=30 FAIL=0`. I will keep #3126's DNS policy and #3128's channel-indexing fix scoped as they are and follow this issue with a separate Store-client PR for channel invalidation. The validated integration patch is https://github.com/bitflicker64/hugegraph-helm-stack/commit/c71c77a0f63846fece72579f9372888592a10ea9. -- 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]
