[
https://issues.apache.org/jira/browse/HADOOP-18521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17634407#comment-17634407
]
ASF GitHub Bot commented on HADOOP-18521:
-----------------------------------------
snvijaya commented on PR #5133:
URL: https://github.com/apache/hadoop/pull/5133#issuecomment-1315456701
> Having spent the last couple of weeks staring at this code, I think what I
would like from the module is
>
> * per file system instance
> * stats reporting to that fs (gauges on queues; actual fetch/hit/miss
numbers could come via abfs input stream)
> * demand creation of the buffer pool. Maybe when the last active stream is
closed it could even free them all, no that at the reference counting problem
to the mix.
> * when an abfs input stream is created for vector/random IO, prefetch
being automatically disabled.
> * anything we can do to improve testing, especially simulation of heavy
load from multiple tasks.
Thanks for your inputs @steveloughran . Agree that all these apply for the
long term transformation on the prefetch handling. I am going to take a day
and see what of these I can get into this change as well, but want to
prioritize a hotfix and remove the buggy bit quickly.
> ABFS ReadBufferManager buffer sharing across concurrent HTTP requests
> ---------------------------------------------------------------------
>
> Key: HADOOP-18521
> URL: https://issues.apache.org/jira/browse/HADOOP-18521
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs/azure
> Affects Versions: 3.3.2, 3.3.3, 3.3.4
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Priority: Critical
> Labels: pull-request-available
>
> AbfsInputStream.close() can trigger the return of buffers used for active
> prefetch GET requests into the ReadBufferManager free buffer pool.
> A subsequent prefetch by a different stream in the same process may acquire
> this same buffer. This can lead to risk of corruption of its own prefetched
> data, data which may then be returned to that other thread.
> On releases without the fix for this (3.3.2+), the bug can be avoided by
> disabling all prefetching
> {code}
> fs.azure.readaheadqueue.depth = 0
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
