Xuan Gong commented on YARN-2468:
bq. If LogContext is not specified, we're running into the traditional log
handling case, right? We will still have a combined log file identified by the
node id? Or node id will always be the directory, and there exists only one
file under it?
node id will always be the directory, and there exists only one file under it
bq. Let's say if work-preserving NM restarting happens, NM is going to forget
all the uploaded logs files, and redo everything, right?
If NM restarts happens, it will upload all logs which are previous uploaded,
but not deleted.
I think that we can solve this problem in separate ticket, because this ticket
is the first step to solve Log handling for LRS.
bq. LogContext doesn't need to be in ApplicatonSubmissionContext, because
ApplicatonSubmissionContext contains ContainerLaunchContext. LogContext is
container related stuff, such that ContainerLaunchContext should be the best
place. Concurrently, we can have one context for all containers. Maybe in the
future we can think of setting different LogContext for each individual
bq. In getFilteredLogFiles, the logic is that if the log file matches the
include pattern, it will be added first, and if then if it matches the exclude
pattern, it will be removed. Shall we do the sanity check to make sure we can
not include and exclude the same pattern, otherwise, the semantics is a bit
Add more explanation in javaDoc.
Uploaded a new patch to address all comments.
> Log handling for LRS
> Key: YARN-2468
> URL: https://issues.apache.org/jira/browse/YARN-2468
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: log-aggregation, nodemanager, resourcemanager
> Reporter: Xuan Gong
> Assignee: Xuan Gong
> Attachments: YARN-2468.1.patch, YARN-2468.2.patch, YARN-2468.3.patch,
> YARN-2468.3.rebase.2.patch, YARN-2468.3.rebase.patch, YARN-2468.4.1.patch,
> YARN-2468.4.patch, YARN-2468.5.patch
> Currently, when application is finished, NM will start to do the log
> aggregation. But for Long running service applications, this is not ideal.
> The problems we have are:
> 1) LRS applications are expected to run for a long time (weeks, months).
> 2) Currently, all the container logs (from one NM) will be written into a
> single file. The files could become larger and larger.
This message was sent by Atlassian JIRA