avro-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Doug Cutting (JIRA)" <j...@apache.org>
Subject [jira] Commented: (AVRO-524) DataFileWriter.appendTo leads to intermittent IOException during write()
Date Wed, 28 Apr 2010 23:03:48 GMT

    [ https://issues.apache.org/jira/browse/AVRO-524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12861987#action_12861987

Doug Cutting commented on AVRO-524:

Do we even still need RandomAccessFile and a FileHandle at all anymore?  These  were only
needed before because it's the only way I could find to get a single file descriptor that's
open for both read and write.  If we're willing to close and re-open the file then the header
can be read with a FileInputStream, then the appends can be made to a FileOutputStream.

Note this is a minor, probably insignificant, change in semantics.  If the file were to be
renamed between the time its header reader is opened and the appender is opened then it would
now fail, where before it would succeed.  This seems very unlikely and not worth protecting
against, but, for the record, it was fears of issues like this that led me to use RandomAccessFile,
so that the file was only opened once, that read and write permission, file existence, etc.
would all be only checked once.  In general, re-opening files is risky, but, in this specific
case, it probably isn't.

Finally, should we perhaps close the reader in a finally clause?

> DataFileWriter.appendTo leads to intermittent IOException during write()
> ------------------------------------------------------------------------
>                 Key: AVRO-524
>                 URL: https://issues.apache.org/jira/browse/AVRO-524
>             Project: Avro
>          Issue Type: Bug
>          Components: java
>            Reporter: Thiruvalluvan M. G.
>            Assignee: Thiruvalluvan M. G.
>             Fix For: 1.4.0
>         Attachments: AVRO-524.patch
> To append to a data file, we first open the file as RandomAccessFile in read-write mode,
read some information such sync, seek to the end of the file and then use its FileDescriptor
to create a FileOutputStream. Sharing a FileDescriptor this way could lead to problem if one
of its containers is garbage-collected while the other is still in use. Please see: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6322678.
The bug got fixed in Java 7 (b06). In our case, RandomAccessFile sometimes gets garbage-collected
leading to write errors. If the Java unit-tests are run multiple times, this occurs about
25% of the time on my Windows machine and about 50% on my Ubuntu Linux box.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message