[ https://issues.apache.org/jira/browse/GOBBLIN-1162?focusedWorklogId=438944&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-438944 ]
ASF GitHub Bot logged work on GOBBLIN-1162: ------------------------------------------- Author: ASF GitHub Bot Created on: 29/May/20 21:30 Start Date: 29/May/20 21:30 Worklog Time Spent: 10m Work Description: sv2000 commented on a change in pull request #3002: URL: https://github.com/apache/incubator-gobblin/pull/3002#discussion_r432744369 ########## File path: gobblin-yarn/src/main/java/org/apache/gobblin/yarn/GobblinYarnTaskRunner.java ########## @@ -209,6 +210,12 @@ public static void main(String[] args) throws Exception { } catch (ParseException pe) { printUsage(options); System.exit(1); + } catch (ContainerHealthCheckException e) { + // Ideally, we should not be catching this exception, as this is indicative of a non-recoverable exception. However, + // simply propagating the exception may prevent the container exit due to the presence of non-daemon threads present + // in the application. Hence, we catch this exception to invoke System.exit() which in turn ensures that all non-daemon threads are killed. + LOGGER.error("Exception encountered: {}", e); + System.exit(1); Review comment: The System.exit(1) will cause the JVM container running the GobblinYarnTaskRunner to terminate. The AMRM client running inside YarnService will then get a callback with ContainerExitStatus == ABORTED. For this case, the hostAffinity check will return to false, implying that the replacement container can be allocated on a different physical host. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org Issue Time Tracking ------------------- Worklog Id: (was: 438944) Time Spent: 50m (was: 40m) > Provide an option to allow slow containers to commit suicide when unhealthy > --------------------------------------------------------------------------- > > Key: GOBBLIN-1162 > URL: https://issues.apache.org/jira/browse/GOBBLIN-1162 > Project: Apache Gobblin > Issue Type: Improvement > Components: gobblin-cluster > Affects Versions: 0.15.0 > Reporter: Sudarshan Vasudevan > Assignee: Hung Tran > Priority: Major > Fix For: 0.15.0 > > Time Spent: 50m > Remaining Estimate: 0h > > In execution environments such as Gobblin-on-Yarn, where Gobblin workers can > be re-assigned when the worker dies or is killed, it is useful to add a mode > where each Gobblin task running inside a Gobblin worker can perform > application-level health checks and report results of the health checks back > to the worker hosting the tasks. The worker on receiving a health check > failure event can be configured to exit the JVM. In the case of > Gobblin-on-Yarn mode, this will result in the worker getting re-assigned to a > different node. The proposed behavior is similar in flavor to speculative > execution modes provided in other execution frameworks such as MapReduce. > This change also provides an example of such an application-level health > check that arises in the case of Kafka ingestion. > -- This message was sent by Atlassian Jira (v8.3.4#803005)