hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Steve Loughran (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-7029) FileOutputCommitter is slow on filesystems lacking recursive delete
Date Fri, 19 Jan 2018 18:27:00 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-7029?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16332693#comment-16332693

Steve Loughran commented on MAPREDUCE-7029:

bq. I don't think FileOutputCommiter actually requires atomicity.

more: "FileOutputCommitter doesn't deliver the atomicity which callers sometimes expect".

In particular, spark OutputCommitCoordinator assumes that after a timeout of a task commit
it's OK to have another task commit. But if you are doing merge in task commit, which v2 does,
that assumption isn't valid. So not really atomic then, more "a failure/partition during task
commit must not interfere with the final result.

The S3A committers have do have this feature

(BTW, congratulations on delving into the hadoop commit code; its one of those bits where
stepping through on a debugger is needed to make sense of the code)

> FileOutputCommitter is slow on filesystems lacking recursive delete
> -------------------------------------------------------------------
>                 Key: MAPREDUCE-7029
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-7029
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>    Affects Versions: 2.8.2
>         Environment: - Google Cloud Storage (with the GCS connector: https://github.com/GoogleCloudPlatform/bigdata-interop/tree/master/gcs)
for HCFS compatibility.
> - FileOutputCommitter algorithm v2.
> - Running on Google Compute Engine with Java 8, Debian 8, Hadoop 2.8.2, Spark 2.2.0.
>            Reporter: Karthik Palaniappan
>            Assignee: Karthik Palaniappan
>            Priority: Minor
>             Fix For: 3.1.0, 2.10.0
>         Attachments: MAPREDUCE-7029-branch-2.004.patch, MAPREDUCE-7029-branch-2.005.patch,
MAPREDUCE-7029-branch-2.005.patch, MAPREDUCE-7029.001.patch, MAPREDUCE-7029.002.patch, MAPREDUCE-7029.003.patch,
MAPREDUCE-7029.004.patch, MAPREDUCE-7029.005.patch
> I ran a Spark job that outputs thousands of parquet files (aka there are thousands of
reducers), and it hung for several minutes in the driver after all tasks were complete. Here
is a very simple repro of the job (to be run in a spark-shell):
> {code:scala}
> spark.range(1L << 20).repartition(1 << 14).write.save("gs://some/path")
> {code}
> Spark actually calls into Mapreduce's FileOuputCommitter. Job committing (specifically
cleanupJob()) recursively deletes the job temporary directory, which is something like "gs://some/path/_temporary".
If I understand correctly, on HDFS, this would be O(1), but on GCS (and every HCFS I know),
this requires a full file tree walk. Deleting tens of thousands of objects in GCS takes several
> I propose that commitTask() recursively deletes its the task attempt temp directory (something
like "gs://some/path/_temporary/attempt1/task1"). On HDFS, this is O(1) per task, so this
is very little overhead per task. On GCS (and other HCFSs), this adds parallelism for deleting
the job temp directory.
> With the attached patch, the repro above went from taking ~10 minutes to taking ~5 minutes,
and task time did not significantly change.
> Side note: I found this issue with Spark, but I assume it applies to a Mapreduce job
with thousands of reducers as well.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: mapreduce-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: mapreduce-issues-help@hadoop.apache.org

View raw message