[Gluster-devel] Release 3.12: Glusto run status
amukherj at redhat.com
Tue Aug 29 13:31:26 UTC 2017
On Tue, Aug 29, 2017 at 4:13 AM, Shyam Ranganathan <srangana at redhat.com>
> Nigel, Shwetha,
> The latest Glusto run [a] that was started by Nigel, post fixing the prior
> timeout issue, failed (much later though) again.
> I took a look at the logs and my analysis is here [b]
> @atin, @kaushal, @ppai can you take a look and see if the analysis is
I took a look at the logs and here is my theory:
glusterd starts the rebalance daemon through runner framework with nowait
mode which essentially means that even though glusterd reports back a
success back to CLI for rebalance start, one of the node might take some
more additional time to start the rebalance process and establish rpc
connection. In this case we hit a race where while one of the node was
still trying to start the rebalance process a rebalance status command was
triggered which eventually failed on the node as rpc connection wasn't
successful and originator glusterd's commit op failed with ""Received
commit RJT from uuid: 6f9524e6-9f9e-44aa-b2f4-393404adfd9d" failure.
Technically to avoid all these spurious time out issues we try to check the
status in a loop till a certain timeout. Isn't that the case in glusto? If
my analysis is correct, you shouldn't be seeing this failure on the 2nd
attempt as its a race.
> In short glusterd has got an error when checking for rebalance stats from
> one of the nodes as:
> "Received commit RJT from uuid: 6f9524e6-9f9e-44aa-b2f4-393404adfd9d"
> and the rebalance deamon on the node with that UUID is not really ready to
> serve requests when this was called, hence I am assuming this is causing
> the error. But need a once over by one of you folks.
> @Shwetha, can we add a further timeout between rebalance start and
> checking the status, just so that we avoid this timing issue on these nodes.
> [a] glusto run: https://ci.centos.org/view/Gluster/job/gluster_glusto/377/
> [b] analysis of the failure: https://paste.fedoraproject.or
> On 08/25/2017 04:29 PM, Shyam Ranganathan wrote:
>> Nigel was kind enough to kick off a glusto run on 3.12 head a couple of
>> days back. The status can be seen here .
>> The run failed, but managed to get past what Glusto does on master (see
>> ). Not that this is a consolation, but just stating the fact.
>> The run  failed at,
>> 17:05:57 functional/bvt/test_cvt.py::TestGlusterHealSanity_dispersed_
>> glusterfs::test_self_heal_when_io_in_progress FAILED
>> The test case failed due to,
>> 17:10:28 E AssertionError: ('Volume %s : All process are not
>> online', 'testvol_dispersed')
>> The test case can be seen here , and the reason for failure is that
>> Glusto did not wait long enough for the down brick to come up (it waited
>> for 10 seconds, but the brick came up after 12 seconds or within the same
>> second as the test for it being up. The log snippets pointing to this
>> problem are here . In short there was no real bug or issue that caused
>> the failure as yet.
>> Glusto as a gating factor for this release was desirable, but having got
>> this far on 3.12 does help.
>> @nigel, we could try post increasing the timeout between bringing the
>> brick up to checking if it is up, and try another run, let me know if that
>> works, and what is needed from me to get this going.
>>  Glusto 3.12 run: https://ci.centos.org/view/Glu
>>  Glusto on master: https://ci.centos.org/view/Glu
>>  Failed test case: https://ci.centos.org/view/Glu
>>  Log analysis pointing to the failed check:
>> "Releases are made better together"
>> Gluster-devel mailing list
>> Gluster-devel at gluster.org
> Gluster-devel mailing list
> Gluster-devel at gluster.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Gluster-devel