I had this exact problem but couldn't wait for them to reestablish connection and when node 2 came back up, everything recovered. It was with 3.4.8 - I posted logs in a previous message.
On Sat, Oct 8, 2016 at 7:55 AM, Flavio Junqueira <[email protected]> wrote: > Hi Anand, > > I don't understand whether 1 and 3 were able or even trying to connect to > each other. They should be able to elect a leader between them and make > progress. You might want to upload logs and let us know. > > -Flavio > > > On 08 Oct 2016, at 02:11, Anand Parthasarathy <[email protected]> > wrote: > > > > Hi, > > > > We are currently using zookeeper 3.4.6 version and use a 3 node solution > in > > our system. We see that occasionally, when a node is powered off (in this > > instance, it was actually a leader node), the remaining two nodes do not > > form a quorum for a really long time. Looking at the logs, it appears the > > sequence is as follows: > > - Node 2 is the zookeeper leader > > - Node 2 is powered off > > - Node 1 and Node 3 recognize and start the election > > - Node 3 times out after initLimit * tickTime with "Timeout while waiting > > for quorum" for Round N > > - Node 1 times out after initLimit * tickTime with "Exception while > trying > > to follow leader" for Round N+1 at the same time. > > - And the process continues where N is sequentially incrementing. > > - This happens for a long time. > > - In one instance, we used tickTime=5000 and initLimit=20 and it took > > around 3.5 hours to converge. > > - In a given round, Node 1 will try connecting to Node 2, gets connection > > refused waits for notification timeout which increases by 2 every > iteration > > until it hits the initLimit. Connection Refused is because the node 2 > comes > > up after reboot, but zookeeper process is not started (due to a different > > failure). > > > > It looks similar to ZOOKEEPER-2164 but there it is a connection timeout > > where Node 2 is not reachable. > > > > Could you pls. share if you have seen this issue and if so, what is the > > workaround that can be employed in 3.4.6. > > > > Thanks, > > Anand. > >
