The ZK ensemble appears to be OK. It is the Solr-related stuff that is borked. There are 110 items in /overseer/collection-queue-work/, which doesn’t seem healthy.
If it is really hosed, I’ll shut down all the nodes, clean out the files in Zookeeper and start over. wunder Walter Underwood wun...@wunderwood.org http://observer.wunderwood.org/ (my blog) > On May 22, 2019, at 8:53 AM, Erick Erickson <erickerick...@gmail.com> wrote: > > Good luck, this kind of assumes that your ZK ensemble is healthy of course... > >> On May 22, 2019, at 8:23 AM, Walter Underwood <wun...@wunderwood.org> wrote: >> >> Thanks, we’ll try that. Bouncing one Solr node doesn’t fix it, because we >> did a rolling restart yesterday. >> >> wunder >> Walter Underwood >> wun...@wunderwood.org >> http://observer.wunderwood.org/ (my blog) >> >>> On May 22, 2019, at 8:21 AM, Erick Erickson <erickerick...@gmail.com> wrote: >>> >>> Walter: >>> >>> I have no idea what the root cause is here, this really shouldn’t happen. >>> But the Overseer role (and I’m assuming you’re talking Solr’s Overseer) is >>> assigned similarly to a shard leader, the same election process happens. >>> All the election nodes are ephemeral ZK nodes. >>> >>> Solr’s Overseer is _not_ fixed to a particular Solr node, although you can >>> assign a preferred role of Overseer in those (rare) cases where there are >>> so many state changes for ZooKeeper that it’s advisable for them to run on >>> a dedicated machine. >>> >>> Overseer assignment is automatic. This should work; >>> 1> shut everything down, Solr and Zookeeper >>> 2> start your ZooKeepers and let them all get in sync with each other >>> 3> start your Solr nodes. It might take 3 minutes or more to bring up the >>> first Solr node, there’s up to a 180 second delay if leaders are not >>> findable easily. >>> >>> That should cause Solr to elect an overseer, probably the first Solr node >>> to come up. >>> >>> It _might_ work to bounce just one Solr node, seeing the Overseer election >>> queue empty it may elect itself. That said, the overseer election queue >>> won’t contain the rest of the Solr nodes like it should, so if that works >>> you should probably bounce the rest of the Solr servers one by one to >>> restore the proper election queue process. >>> >>> Not a fix for the root cause of course, but should get things operating >>> again. I’ll add that I haven’t seen this happen in the field to my >>> recollection, if at all. >>> >>> Best, >>> Erick >>> >>>> On May 21, 2019, at 9:04 PM, Will Martin <wmar...@urgent.ly> wrote: >>>> >>>> Worked with Fusion and Zookeeper at GSA for 18 months: admin role. >>>> >>>> Before blowing it away, you could try: >>>> >>>> - id a candidate node, with a snapshot you just might think is old enough >>>> to be robust. >>>> - clean data for zk nodes otherwise. >>>> - bring up the chosen node and wait for it to settle[wish i could remember >>>> why i called what i saw that] >>>> - bring up other nodes 1 at a time. let each one fully sync to follower of >>>> the new leader. >>>> - they should each in turn request the snapshot from the lead. then you >>>> have >>>> >>>> : align your collections with the ensemble. and for the life of me i can't >>>> remember there being anything particularly tricky about that with fusion , >>>> which means I can't remember what I did... or have it doc'd at home. ;-) >>>> >>>> >>>> Will Martin >>>> DEVOPS ENGINEER >>>> 540.454.9565 >>>> >>>> 8609 WESTWOOD CENTER DR, SUITE 475 >>>> VIENNA, VA 22182 >>>> geturgently.com >>>> >>>> >>>> On Tue, May 21, 2019 at 11:40 PM Walter Underwood <wun...@wunderwood.org> >>>> wrote: >>>> >>>>> Yes, please. I have the logs from each of the Zookeepers. >>>>> >>>>> We are running 3.4.12. >>>>> >>>>> wunder >>>>> Walter Underwood >>>>> wun...@wunderwood.org >>>>> http://observer.wunderwood.org/ (my blog) >>>>> >>>>>> On May 21, 2019, at 6:49 PM, Will Martin <wmar...@urgent.ly> wrote: >>>>>> >>>>>> Walter. Can I cross-post to zk-dev? >>>>>> >>>>>> >>>>>> >>>>>> Will Martin >>>>>> DEVOPS ENGINEER >>>>>> 540.454.9565 >>>>>> >>>>>> <urgently-email-logo> >>>>>> >>>>>> 8609 WESTWOOD CENTER DR, SUITE 475 >>>>>> VIENNA, VA 22182 >>>>>> geturgently.com <http://geturgently.com/> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>> On May 21, 2019, at 9:26 PM, Will Martin <wmar...@urgent.ly <mailto: >>>>> wmar...@urgent.ly>> wrote: >>>>>>> >>>>>>> +1 >>>>>>> >>>>>>> Will Martin >>>>>>> DEVOPS ENGINEER >>>>>>> 540.454.9565 >>>>>>> >>>>>>> 8609 WESTWOOD CENTER DR, SUITE 475 >>>>>>> VIENNA, VA 22182 >>>>>>> geturgently.com <http://geturgently.com/> >>>>>>> >>>>>>> >>>>>>> On Tue, May 21, 2019 at 7:39 PM Walter Underwood <wun...@wunderwood.org >>>>> <mailto:wun...@wunderwood.org>> wrote: >>>>>>> ADDROLE times out after 180 seconds. This seems to be an unrecoverable >>>>> state for the cluster, so that is a pretty serious bug. >>>>>>> >>>>>>> wunder >>>>>>> Walter Underwood >>>>>>> wun...@wunderwood.org <mailto:wun...@wunderwood.org> >>>>>>> http://observer.wunderwood.org/ <http://observer.wunderwood.org/> (my >>>>> blog) >>>>>>> >>>>>>>> On May 21, 2019, at 4:10 PM, Walter Underwood <wun...@wunderwood.org >>>>> <mailto:wun...@wunderwood.org>> wrote: >>>>>>>> >>>>>>>> We have a 6.6.2 cluster in prod that appears to have no overseer. In >>>>> /overseer_elect on ZK, there is an election folder, but no leader >>>>> document. >>>>> An OVERSEERSTATUS request fails with a timeout. >>>>>>>> >>>>>>>> I’m going to try ADDROLE, but I’d be delighted to hear any other >>>>> ideas. We’ve diverted all the traffic to the backing cluster, so we can >>>>> blow this one away and rebuild. >>>>>>>> >>>>>>>> Looking at the Zookeeper logs, I see a few instances of network >>>>> failures across all three nodes. >>>>>>>> >>>>>>>> wunder >>>>>>>> Walter Underwood >>>>>>>> wun...@wunderwood.org <mailto:wun...@wunderwood.org> >>>>>>>> http://observer.wunderwood.org/ <http://observer.wunderwood.org/> >>>>> (my blog) >>>>>>>> >>>>>>> >>>>>> >>>>> >>>>> >>> >> >