[Spread-users] very big lag when one spread daemon stopped

[Csaba Endre Simon]

Hi John,

> Can you explain the timings you took?  Do those numbers (e.g. - 1, 11802,
> 11599, etc.) represent the # of ms since the last message delivery?

Yes, it is # of ms since the last message was received.

> Are you sure that the daemon you kill isn't restarting/recrashing?

Yes. The daemon is started in foreground, so a CTRL-C will stop the daemon.

> You can turn on the MEMBERSHIP debug flag to see if the daemon is flailing
> for some reason in reestablishing the membership.

I will do it and I will come back with the log if I find something unusual.

> The timeouts that control membership are set at the top of membership.c
> In particular you might want to lower the token timeout (how quickly
> broken rings are declared)

Thanks, I will try it.

I posted the link to the C and the Java programs written by Datoh, so
anybody can reproduce it. Also I posted the configuration file. Just
start two spread instance on two different computer in foreground.
Start the C or Java client and server program on one node and after a
while kill the spread daemon with ctrl-c on the second node.

Regards,
Csaba