[
https://issues.apache.org/jira/browse/GOBBLIN-1099?focusedWorklogId=411517&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-411517
]
ASF GitHub Bot logged work on GOBBLIN-1099:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 27/Mar/20 23:41
Start Date: 27/Mar/20 23:41
Worklog Time Spent: 10m
Work Description: autumnust commented on pull request #2940:
GOBBLIN-1099: Handle orphaned Yarn containers in Gobblin-on-Yarn clus…
URL: https://github.com/apache/incubator-gobblin/pull/2940#discussion_r399583366
##########
File path: gobblin-yarn/src/main/java/org/apache/gobblin/yarn/YarnService.java
##########
@@ -635,10 +644,26 @@ protected void handleContainerCompletion(ContainerStatus
containerStatus) {
containerStatus.getContainerId(), containerStatus.getDiagnostics()));
}
- if
(this.releasedContainerCache.getIfPresent(containerStatus.getContainerId()) !=
null) {
- LOGGER.info("Container release requested, so not spawning a replacement
for containerId {}",
- containerStatus.getContainerId());
- return;
+ if (containerStatus.getExitStatus() == ContainerExitStatus.ABORTED) {
+ if
(this.releasedContainerCache.getIfPresent(containerStatus.getContainerId()) !=
null) {
+ LOGGER.info("Container release requested, so not spawning a
replacement for containerId {}", containerStatus.getContainerId());
+ return;
+ } else {
+ LOGGER.info("Container {} aborted due to lost NM",
containerStatus.getContainerId());
+ // Container release was not requested. Likely, the container was
running on a node on which the NM died.
+ // In this case, RM assumes that the containers are "lost", even though
the container process may still be
+ // running on the node. We need to ensure that the Helix instances
running on the orphaned containers
+ // are fenced off from the Helix cluster to avoid double publishing
and state being committed by the
+ // instances.
+ if (!UNKNOWN_HELIX_INSTANCE.equals(completedInstanceName)) {
+ String clusterName = this.helixManager.getClusterName();
+ //Disable the orphaned instance.
+ if (HelixUtils.isInstanceLive(helixManager, completedInstanceName)) {
+ LOGGER.info("Disabling the Helix instance {}",
completedInstanceName);
+
this.helixManager.getClusterManagmentTool().enableInstance(clusterName,
completedInstanceName, false);
Review comment:
enableInstance, This is a very good API :)
----------------------------------------------------------------
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: 411517)
Time Spent: 1h 10m (was: 1h)
> Handle orphaned Yarn containers in Gobblin-on-Yarn clusters
> -----------------------------------------------------------
>
> Key: GOBBLIN-1099
> URL: https://issues.apache.org/jira/browse/GOBBLIN-1099
> Project: Apache Gobblin
> Issue Type: Improvement
> Components: gobblin-yarn
> Affects Versions: 0.15.0
> Reporter: Sudarshan Vasudevan
> Assignee: Abhishek Tiwari
> Priority: Major
> Fix For: 0.15.0
>
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> A Yarn application may leave behind orphaned containers, which can happen due
> to lost node managers. The orphaned containers however can continue to run
> (potentially forever) as participants in the Helix cluster. This can cause
> the following problems for a Gobblin-on-Yarn application:
> # Double publish of data and commit of state
> # Task failures and partition starvation during application restarts, as
> Helix may assign tasks to the orphaned containers which have a stale state
> and configuration
> # Container failures on application restarts due to Helix instance name
> collisions with orphaned containers
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
