hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ramkrishna.s.vasudevan (Jira)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-16998) WASB : NativeAzureFsOutputStream#close() throwing java.lang.IllegalArgumentException instead of IOE which causes HBase RS to get aborted
Date Tue, 21 Apr 2020 13:07:00 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-16998?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17088658#comment-17088658
] 

ramkrishna.s.vasudevan commented on HADOOP-16998:
-------------------------------------------------

+1 on patch's approach. (non binding). We need test anyway. 

> WASB : NativeAzureFsOutputStream#close() throwing java.lang.IllegalArgumentException
instead of IOE which causes HBase RS to get aborted
> ----------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-16998
>                 URL: https://issues.apache.org/jira/browse/HADOOP-16998
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: fs/azure
>            Reporter: Anoop Sam John
>            Assignee: Anoop Sam John
>            Priority: Major
>         Attachments: HADOOP-16998.patch
>
>
> During HFile create, at the end when called close() on the OutputStream, there is some
pending data to get flushed. When this flush happens, an Exception is thrown back from Storage.
The Azure-storage SDK layer will throw back IOE. (Even if it is a StorageException thrown
from the Storage, the SDK converts it to IOE.) But at HBase, we end up getting IllegalArgumentException
which causes the RS to get aborted. If we get back IOE, the flush will get retried instead
of aborting RS.
> The reason is this
> NativeAzureFsOutputStream uses Azure-storage SDK's BlobOutputStreamInternal. But the
BlobOutputStreamInternal is wrapped within a SyncableDataOutputStream which is a FilterOutputStream.
During the close op, NativeAzureFsOutputStream calls close on SyncableDataOutputStream and
it uses below method from FilterOutputStream
> {code}
> public void close() throws IOException {
>   try (OutputStream ostream = out) {
>               flush();
>   }
> }
> {code}
> Here the flush call caused an IOE to be thrown to here. The finally will issue close
call on ostream (Which is an instance of BlobOutputStreamInternal)
> When BlobOutputStreamInternal#close() is been called, if there was any exception already
occured on that Stream, it will throw back the same Exception
> {code}
> public synchronized void close() throws IOException {
>   try {
>               // if the user has already closed the stream, this will throw a STREAM_CLOSED
exception
>               // if an exception was thrown by any thread in the threadExecutor, realize
it now
>               this.checkStreamState();
>               ...
> }
> private void checkStreamState() throws IOException {
>   if (this.lastError != null) {
>               throw this.lastError;
>   }
> }
> {code}
> So here both try and finally block getting Exceptions and Java uses Throwable#addSuppressed()

> Within this method if both Exceptions are same objects, it throws back IllegalArgumentException
> {code}
> public final synchronized void addSuppressed(Throwable exception) {
>               if (exception == this)
>                              throw new IllegalArgumentException(SELF_SUPPRESSION_MESSAGE,
exception);
>               ....
> }
> {code}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

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


Mime
View raw message