[
https://issues.apache.org/jira/browse/HDFS-12613?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lei (Eddy) Xu updated HDFS-12613:
---------------------------------
Attachment: HDFS-12613.02.patch
Thanks for the reviews, [~drankye] and [~Sammi]
bq. note "NativeRSRawDecoder" should be "NativeRSRawEncoder".
Done
bq. Apart from add synchronized on release, performEncodeImpl and
performDecodeImpl can also have the synchronized keyword
Done.
bq. If it's already null, we don't need to call the native code through JNI.
Good point. If we go this route, we should {{throw IOException}} in java, to
notify the client that the coder is closed. I prefer does it in the JNI as the
logic of setting it to NULL is in JNI. But I am fine either way. What do you
think.
> Native EC coder should implement release() as idempotent function.
> ------------------------------------------------------------------
>
> Key: HDFS-12613
> URL: https://issues.apache.org/jira/browse/HDFS-12613
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: erasure-coding
> Affects Versions: 3.0.0-beta1
> Reporter: Lei (Eddy) Xu
> Assignee: Lei (Eddy) Xu
> Attachments: HDFS-12613.00.patch, HDFS-12613.01.patch,
> HDFS-12613.02.patch
>
>
> Recently, we found native EC coder crashes JVM because
> {{NativeRSDecoder#release()}} being called multiple times (HDFS-12612 and
> HDFS-12606).
> We should strength the implement the native code to make {{release()}}
> idempotent as well.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
