Please also add a note that one should specify IP addresses in ringX_addr directives, not a domain name. Else corosync does not work properly in UDPu mode, and at the same time it does not say anything significant in its log files. I've spent 4 hours recently trying to figure this out.
On Monday, January 5, 2015, Jan Pokorný <[email protected]> wrote: > (if you let me, some more in-line) > > On 05/01/15 16:20 +0000, Christine Caulfield wrote: > > Looks good to me, thanks. I've fixed a few typos and pointed out a > spurious > > capital inline below > > > > On 05/01/15 14:39, Steven Dake wrote: > >> Add a troubleshooting guide. I'm sure other folks have some good stuff > >> to put in here. These are just the ones I know about :) > >> > >> Signed-off-by: Steven Dake <[email protected] <javascript:;>> > >> --- > >> man/corosync.conf.5 | 39 +++++++++++++++++++++++++++++++++++++++ > >> 1 file changed, 39 insertions(+) > >> > >> diff --git a/man/corosync.conf.5 b/man/corosync.conf.5 > >> index 8e774c1..16d84ca 100644 > >> --- a/man/corosync.conf.5 > >> +++ b/man/corosync.conf.5 > >> @@ -678,6 +678,45 @@ Native means one of shm or socket, depending on > what is supported by OS. On syst > >> with support for both, SHM is selected. SHM is generally faster, but > need to allocate > >> ring buffer file in /dev/shm. > >> > >> +.SH "TROUBLESHOOTING" > >> +.TP > >> +Ocassionally Corosync will not work with the default network. Here > are some > ^^^ Occasionally > > >> +common tips that people have used to find a working Corosync. > >> + > >> +.TP > >> +Disable the firewall. The firwall could block Corosync packets from > reaching > > ^^firewall > >> +the network. > >> + > >> +.TP > >> +Force IGMP v2. Some modern switches do not support the kernel IGMP v3 > >> + protocol. As a result, They will not properly register the cluster. > To do > ^^^ they > > >> +this, simply run the command > >> + > >> +.BR sysctl -w net.ipv4.conf.all.force_igmp_version=2 > >> + > >> +.TP > >> +If on a routed network, set a larger ttl. The TTL tells the routers > how long > >> +to let the packet multicast before dropping it permanently. The > Default ttl > > ^^^ default > > (inconsistent casing of ttl/TTL) > > >> +is set to 1, which means the packet will drop after its first hop. > This will > >> +not work well on a routed network. > >> + > >> +.TP > >> +I use a VLAN and Corosync doesn't work. If your using a VLAN, VLAN's > shave the > > ^^^ you're VLANs > > > >> +packet size available for Corosync to use in some cases. Corosync does > not > >> +automatically adjust to this change. Set netmtu appropriately when > using a > >> +VLAN. > >> + > >> +.TP > >> +If all else fails, use UDPU. The authors implemented UDPU to solve > the various > >> +problems with multicast that plague modern switch implementations. > The UDPU > >> +protocol was initially believed to be much slower but the reality after > >> +implementation is that it doesn't make much difference. > >> + > >> +Even with UDPU you would be hard pressed to find a faster group > messaging > >> +system than Corosync. The only downside of UDPU is it results in much > more > >> +packet copying across the network. > >> + > >> + > >> .SH "FILES" > >> .TP > >> /etc/corosync/corosync.conf > > -- > Jan >
_______________________________________________ discuss mailing list [email protected] http://lists.corosync.org/mailman/listinfo/discuss
