On 10/6/2015 3:38 AM, Adrian Liew wrote:
> Thanks for the reply. Looks like this has been resolved by manually starting
> the Zookeeper services on each server promptly so that the tickTime value
> does not timeout too quickly to heartbeat other peers. Hence, I increased the
> tickTime value to about 5 minutes to give some time for a node hosting
> Zookeeper to restart and autostart its service. This case seems fixed but I
> will double check again once more to be sure. I am using nssm
> (non-sucking-service-manager) to autostart Zookeeper. I will need to retest
> this once again using nssm to make sure zookeeper services are up and running.
That sounds like a very bad idea. A typical tickTime is two *seconds*.
Zookeeper is designed around certain things happening very quickly.
I don't think you can increase that to five *minutes* (multiplying it by
150) without the strong possibility of something going very wrong and
processes hanging for minutes at a time waiting for a timeout that
should happen very quickly.
I am reasonably certain that tickTime is used for zookeeper operation in
several ways, so I believe that this much of an increase will cause
fundamental problems with zookeeper's normal operation. I admit that I
have not looked at the code, so I could be wrong ... but based on the
following information from the Zookeeper docs, I don't think I am wrong:
tickTime
the length of a single tick, which is the basic time unit used by
ZooKeeper, as measured in milliseconds. It is used to regulate
heartbeats, and timeouts. For example, the minimum session timeout will
be two ticks.
Thanks,
Shawn
