[ https://issues.apache.org/jira/browse/IMPALA-7283?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Tim Armstrong updated IMPALA-7283: ---------------------------------- Description: [~stiga-huang] suggested here https://gerrit.cloudera.org/#/c/10744/11/be/src/service/client-request-state.cc@628 {quote} What about sleep several seconds before the next retry like this? for (int i = 0; i < 3; ++i, sleep(3)) Usually it will increase success rate if there're network issues or the target server is stuck temporarily. {quote} This seems worth considering but I don't have the knowledge to really evaluate it This would likely involve: * For KRPC, using the backoff logic in IMPALA-8143 for all RPCs, maybe doing exponential backoff sometimes * For thrift, adding backoff logic like this to the thrift version of DoRpcWithRetry. was: [~stiga-huang] suggested here https://gerrit.cloudera.org/#/c/10744/11/be/src/service/client-request-state.cc@628 {quote} What about sleep several seconds before the next retry like this? for (int i = 0; i < 3; ++i, sleep(3)) Usually it will increase success rate if there're network issues or the target server is stuck temporarily. {quote} This seems worth considering but I don't have the knowledge to really evaluate it cc [~kwho] [~sailesh] > Consider backing off before retrying RPC > ---------------------------------------- > > Key: IMPALA-7283 > URL: https://issues.apache.org/jira/browse/IMPALA-7283 > Project: IMPALA > Issue Type: Improvement > Components: Distributed Exec > Reporter: Tim Armstrong > Priority: Major > > [~stiga-huang] suggested here > https://gerrit.cloudera.org/#/c/10744/11/be/src/service/client-request-state.cc@628 > {quote} > What about sleep several seconds before the next retry like this? > for (int i = 0; i < 3; ++i, sleep(3)) > Usually it will increase success rate if there're network issues or the > target server is stuck temporarily. > {quote} > This seems worth considering but I don't have the knowledge to really > evaluate it > This would likely involve: > * For KRPC, using the backoff logic in IMPALA-8143 for all RPCs, maybe doing > exponential backoff sometimes > * For thrift, adding backoff logic like this to the thrift version of > DoRpcWithRetry. -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org For additional commands, e-mail: issues-all-h...@impala.apache.org