[
https://issues.apache.org/jira/browse/CASSANDRA-13692?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16722172#comment-16722172
]
Dimitar Dimitrov commented on CASSANDRA-13692:
----------------------------------------------
First, apologies for making this look like a 19th century trans-continental
chess correspondence game...
I think you're absolutely right - after
[{{849a438690aa97a361227781108cc90355dcbcd9}}|https://github.com/apache/cassandra/commit/849a438690aa97a361227781108cc90355dcbcd9],
we return solely candidates from some subset of
{{Directories.dataDirectories}}, all of which are initialized in the
{{clinit}}, so currently it doesn't look like there's a case in which the error
handling could be hit.
I agree with your suggestion (although I'm a tiny bit sad that the
oh-so-laborious test would go as well).
I'll dust off the ancient branches that I had for this and update here with the
corresponding patches soon.
> CompactionAwareWriter_getWriteDirectory throws incompatible exceptions
> ----------------------------------------------------------------------
>
> Key: CASSANDRA-13692
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13692
> Project: Cassandra
> Issue Type: Bug
> Components: Compaction
> Reporter: Hao Zhong
> Assignee: Dimitar Dimitrov
> Priority: Major
> Labels: lhf
> Attachments: c13692-2.2-dtest-results.PNG,
> c13692-2.2-testall-results.PNG, c13692-3.0-dtest-results-updated.PNG,
> c13692-3.0-dtest-results.PNG, c13692-3.0-testall-results.PNG,
> c13692-3.11-dtest-results-updated.PNG, c13692-3.11-dtest-results.PNG,
> c13692-3.11-testall-results.PNG, c13692-dtest-results-updated.PNG,
> c13692-dtest-results.PNG, c13692-testall-results-updated.PNG,
> c13692-testall-results.PNG
>
>
> The CompactionAwareWriter_getWriteDirectory throws RuntimeException:
> {code}
> public Directories.DataDirectory getWriteDirectory(Iterable<SSTableReader>
> sstables, long estimatedWriteSize)
> {
> File directory = null;
> for (SSTableReader sstable : sstables)
> {
> if (directory == null)
> directory = sstable.descriptor.directory;
> if (!directory.equals(sstable.descriptor.directory))
> {
> logger.trace("All sstables not from the same disk - putting
> results in {}", directory);
> break;
> }
> }
> Directories.DataDirectory d =
> getDirectories().getDataDirectoryForFile(directory);
> if (d != null)
> {
> long availableSpace = d.getAvailableSpace();
> if (availableSpace < estimatedWriteSize)
> throw new RuntimeException(String.format("Not enough space to
> write %s to %s (%s available)",
>
> FBUtilities.prettyPrintMemory(estimatedWriteSize),
> d.location,
>
> FBUtilities.prettyPrintMemory(availableSpace)));
> logger.trace("putting compaction results in {}", directory);
> return d;
> }
> d = getDirectories().getWriteableLocation(estimatedWriteSize);
> if (d == null)
> throw new RuntimeException(String.format("Not enough disk space
> to store %s",
>
> FBUtilities.prettyPrintMemory(estimatedWriteSize)));
> return d;
> }
> {code}
> However, the thrown exception does not trigger the failure policy.
> CASSANDRA-11448 fixed a similar problem. The buggy code is:
> {code}
> protected Directories.DataDirectory getWriteDirectory(long writeSize)
> {
> Directories.DataDirectory directory =
> getDirectories().getWriteableLocation(writeSize);
> if (directory == null)
> throw new RuntimeException("Insufficient disk space to write " +
> writeSize + " bytes");
> return directory;
> }
> {code}
> The fixed code is:
> {code}
> protected Directories.DataDirectory getWriteDirectory(long writeSize)
> {
> Directories.DataDirectory directory =
> getDirectories().getWriteableLocation(writeSize);
> if (directory == null)
> throw new FSWriteError(new IOException("Insufficient disk space
> to write " + writeSize + " bytes"), "");
> return directory;
> }
> {code}
> The fixed code throws FSWE and triggers the failure policy.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
