[
https://issues.apache.org/jira/browse/CASSANDRA-4847?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13495960#comment-13495960
]
Kirk True commented on CASSANDRA-4847:
--------------------------------------
I'd like to know if dying on startup with a bad disk regardless of
disk_failure_policy setting is desired behavior. I'd lean toward it being 'no'
from a pragmatic standpoint, but would like to know if there's a precedent
against _simply_ black-listing the bad disk and moving on. I'd like to fix it
from a robustness POV.
> Bad disk causes death of node despite disk_failure_policy
> ---------------------------------------------------------
>
> Key: CASSANDRA-4847
> URL: https://issues.apache.org/jira/browse/CASSANDRA-4847
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Affects Versions: 1.2.0 beta 1
> Reporter: Kirk True
> Assignee: Kirk True
>
> Steps:
> # Create a bad disk via device mapper
> # Specify good disk and bad disk is data directory
> # Set {{disk_failure_policy}} to {{best_effort}} in cassandra.yaml
> # Start node
> Expected:
> Attempts to create system directories to fail (as expected) on bad disk, and
> have it added to blacklisted directories.
> Actual:
> Node start up aborts due to uncaught error:
> {noformat}
> FSWriteError in /mnt/bad_disk/system_traces/sessions
> at
> org.apache.cassandra.io.util.FileUtils.createDirectory(FileUtils.java:258)
> at org.apache.cassandra.db.Directories.<init>(Directories.java:104)
> at org.apache.cassandra.db.Directories.create(Directories.java:90)
> at
> org.apache.cassandra.db.ColumnFamilyStore.scrubDataDirectories(ColumnFamilyStore.java:404)
> at
> org.apache.cassandra.service.CassandraDaemon.setup(CassandraDaemon.java:227)
> at
> org.apache.cassandra.service.CassandraDaemon.activate(CassandraDaemon.java:393)
> at
> org.apache.cassandra.service.CassandraDaemon.main(CassandraDaemon.java:436)
> Caused by: java.io.IOException: Failed to mkdirs
> /mnt/bad_disk/system_traces/sessions
> ... 7 more
> {noformat}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
