[ 
https://issues.apache.org/jira/browse/FLINK-5193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15708684#comment-15708684
 ] 

ASF GitHub Bot commented on FLINK-5193:
---------------------------------------

Github user uce commented on a diff in the pull request:

    https://github.com/apache/flink/pull/2910#discussion_r90235653
  
    --- Diff: 
flink-runtime/src/main/java/org/apache/flink/runtime/jobmanager/ZooKeeperSubmittedJobGraphStore.java
 ---
    @@ -275,6 +242,25 @@ public void removeJobGraph(JobID jobId) throws 
Exception {
                }
        }
     
    +   @Override
    +   public Collection<JobID> getJobIds() throws Exception {
    +           Collection<String> paths;
    +
    +           try {
    +                   paths = jobGraphsInZooKeeper.getAllPaths();
    +           } catch (Exception e) {
    +                   throw new Exception("Failed to retrieve entry paths 
from ZooKeeperStateHandleStore.", e);
    +           }
    +
    +           List<JobID> jobIds = new ArrayList<>(paths.size());
    +
    +           for (String path : paths) {
    +                   jobIds.add(jobIdfromPath(path));
    --- End diff --
    
    If we have a malformatted sub node, this will skip recovery of all jobs 
again, right? Wondering if we should wrap this line in a `try-catch`. It's 
quite unlikely though as someone would need to put the node there manually ;) 
Feel free to not address this.


> Recovering all jobs fails completely if a single recovery fails
> ---------------------------------------------------------------
>
>                 Key: FLINK-5193
>                 URL: https://issues.apache.org/jira/browse/FLINK-5193
>             Project: Flink
>          Issue Type: Bug
>          Components: JobManager
>    Affects Versions: 1.2.0, 1.1.3
>            Reporter: Till Rohrmann
>            Assignee: Till Rohrmann
>             Fix For: 1.2.0, 1.1.4
>
>
> In HA case where the {{JobManager}} tries to recover all submitted job 
> graphs, e.g. when regaining leadership, it can happen that none of the 
> submitted jobs are recovered if a single recovery fails. Instead of failing 
> the complete recovery procedure, the {{JobManager}} should still try to 
> recover the remaining (non-failing) jobs and print a proper error message for 
> the failed recoveries.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to