Aleksey Chudov wrote:
> Do you think the size of UDP buffer is enough?
If you get the send_async errors, the buffer probably is not enough. I
think the default value is the one that affects the buffer size.
Note that you have to restart the sync daemons for changed buffer
sizes to take effect. If you are not running anything else on your
LVS, you can be geneorous with the buffer sizes. Add a zero to the end
and see if you can get the problem to go away.
We have been running our production LVSs with 0.1s timeouts for some
time and it seems to push the problem far enough not to appear any
more. But of course, a dynamic wait would be more sexy. Or
alternatively a tunable wait. I don't think 100ms (or even 10ms) wait
would cause any problems.
Please read the documentation before posting - it's available at:
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users