[Gluster-devel] removing the statedump options file
rabhat at redhat.com
Tue Feb 19 07:27:39 UTC 2013
On 02/19/2013 12:47 PM, Anand Avati wrote:
> On Mon, Feb 18, 2013 at 11:03 PM, Raghavendra Bhat <rabhat at redhat.com
> <mailto:rabhat at redhat.com>> wrote:
> As of now when statedump command is issued via cli (gluster volume
> statedump <volname> [options]) depending upon what options is
> given via cli a temporary file (glusterdump.<pid>.options) file is
> created by glusterd and glusterfsd process read that file to
> decide what information should be dumped. But the problem is
> glusterd after issuing the SIGUSR1 signal to the glusterfsd
> processes, sleeps for 1 second and then unlinks the options file.
> To fix that a patch was sent (
> http://review.gluster.org/#change,2585), where
> * the glusterfsd process after dumping the information to the
> statedump file, unlinked the options by (instead of glusterd doing it)
> Another approach suggested to me was this:
> * Have a separate thread in glusterd which keeps on polling for
> the file glusterdump.<pid>.options.over (i.e some renamed file).
> glusterfsd after dumping the information, renames the options file
> and thats when glusterd realizes the statedump is taken and
> unlinks the file. (The new thread is spawned whenever statedump is
> issued and is finished after unlinking the renamed options file).
> If the purpose of the other thread is to just keep polling for
> rename() to complete and eventually delete, why not just let the file
> be deleted by glusterfsd? If glusterd needs to know when glusterfsd
> "finished its work", then instead of polling for rename, it could just
> poll for unlink() of the options file, no?
I think the 1st method where glusterfsd unlinks the file is ok. Because
glusterd says statedump is successful or unsuccessful right after
issuing the SIGUSR1 signal to the brick processes. Here the whole
purpose of the new thread would be just to unlink the file. It just
exits after unlinking the file. If glusterfsd itself unlinks the file,
then there is no need for the new thread at all, as all it would be
doing is get spawned, wait till the file gets unlinked by glusterfsd and
then exit (without doing anything as informing the user about
success/failure of statedump would have been done right after issuing
SIGUSR1 signal to the brick process itself).
> Please let me know the inputs.
> Raghavendra Bhat
> Gluster-devel mailing list
> Gluster-devel at nongnu.org <mailto:Gluster-devel at nongnu.org>
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Gluster-devel