[
https://issues.apache.org/jira/browse/CASSANDRA-10957?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
T Jake Luciani updated CASSANDRA-10957:
---------------------------------------
Description:
In JVMStabilityInspector we only mark ourselves unstable when we get some
special messages in file not found exceptions.
{code}
// Check for file handle exhaustion
if (t instanceof FileNotFoundException || t instanceof SocketException)
if (t.getMessage().contains("Too many open files"))
isUnstable = true;
{code}
It seems like the OS might also have the same issue of too many open files but
will instead return "No such file or directory".
It might make more sense, when we check this exception type, to try to read a
known-to-exist file to verify the disk is readable.
This would mean creating a hidden file on startup on each data disk? other
ideas?
was:
In JVMStabilityInspector we only mark ourselves unstable when we get some
special messages in file not found exceptions.
{code}
// Check for file handle exhaustion
if (t instanceof FileNotFoundException || t instanceof SocketException)
if (t.getMessage().contains("Too many open files"))
isUnstable = true;
{code}
It seems like the OS might also have the same issue of too many open files but
will instead return "No such file or directory".
It might make more sense when we check this exception type we try to read a
known to exist file to verify if the disk is readable vs relying on the current
check.
This would mean creating a hidden file on startup on each data disk? other
ideas?
> Verify disk is readable on FileNotFound Exceptions
> --------------------------------------------------
>
> Key: CASSANDRA-10957
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10957
> Project: Cassandra
> Issue Type: Improvement
> Reporter: T Jake Luciani
>
> In JVMStabilityInspector we only mark ourselves unstable when we get some
> special messages in file not found exceptions.
> {code}
> // Check for file handle exhaustion
> if (t instanceof FileNotFoundException || t instanceof
> SocketException)
> if (t.getMessage().contains("Too many open files"))
> isUnstable = true;
> {code}
> It seems like the OS might also have the same issue of too many open files
> but will instead return "No such file or directory".
> It might make more sense, when we check this exception type, to try to read a
> known-to-exist file to verify the disk is readable.
> This would mean creating a hidden file on startup on each data disk? other
> ideas?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
