Re:GitHub: Recent Load Balancer Problems (RCA)
by seiji on Tuesday October 04, 2011 @12:51PM
(#3071767)
I rarely see an install of Heartbeat/Pacemaker/CRM preventing more downtime than they cause. If you add in DRBD on top, you get an entire suite of false-HA infrastructure.
[ Parent
| on HN ]
Re:GitHub: Recent Load Balancer Problems (RCA)
by stock_toaster on Tuesday October 04, 2011 @02:46PM
(#3072249)
Just curious, but what have you seen working well?
At a previous gig we used heartbeat with haproxy, and it worked pretty well. We would drop connections on cutover, but it was considered 'acceptable' for our purposes at the time. I wanted to try whackamole with haproxy, but we never got around to it.
Re:GitHub: Recent Load Balancer Problems (RCA)
by seiji on Tuesday October 04, 2011 @02:56PM
(#3072313)
The only IP failover I trust is carp (http://www.openbsd.org/faq/pf/carp.html) on OpenBSD/FreeBSD. Once set up properly with syncing, you lose no state on a failover (all connection and NAT state is gossiped between cluster nodes sharing an IP address).
The only downside is most services aren't well tested under OpenBSD/FreeBSD these days so you may end up hitting a few edge cases in software designed and tested only under Linux.
Re:GitHub: Recent Load Balancer Problems (RCA)
by ww520 on Tuesday October 04, 2011 @04:52PM
(#3072871)
For most apps and servers, migrating the IP state is not enough. The app server's connection state cannot be easily migrated. E.g. The MySQL connection of a failed MySQL server will be gone. Migrating the IP connection to a new MySQL server won't do anything.
Carp is good for firewall server redundancy because IP state is all that firewall is maintaining.