|
hi
I already posted this on the keepalived mailing list but since no one
responded I thought I'd give it a shot here.
Question 1 is about the option "preempt_delay" in keepalived.conf:
During my tests I just unplugged the power of a switch that connects the
MASTER director with the realservers. BACKUP takes over and as this one
is connected to another switch, my service is available - good!
When I plug the power back into the switch, MASTER falls back
immediately. But as the switch needs about 2 minutes to boot up, this is
how long none of my services are avaiable.
So from reading keealived.conf's man page, I assumed that "preempt_delay
150" would tell BACKUP to wait 2 1/2 minutes until it hands over
services to MASTER again.
Did I get this wrong? Because when I changed my conf and restarted
keepalived, the same test did the exact same thing - services fell back
to MASTER as soon as the link was up again.
I made sure initial state of every vrrp_instance is BACKUP (as the
manpage said) - only the server to be the first master has a higher prio
(master 100, backup 10).
I found out that if I stop and start keepalived on MASTER, the delay
works just fine. This error only seems to occur on link status change.
And another thing to add:
I read about the "portfast" feature on some cisco Switches that was
mentioned in a Thread about Fault to Master Transition early 2005 on
this list.
Unfortunately, my switch does not support anything like this.
So has anyone tested that patch Dimitri posted back then?
http://sourceforge.net/mailarchive/forum.php?thread_id=6610566&forum_id=3211
The patch seems to be incomplete in the archive according to the some
bracket-counts at the end of the first post in the thread, so I cannot
test it myself.
question 2 is about keepalived handling and synching multiple synch groups:
On my director there are 2 synch groups - synch1 and synch2. When both
are online on MASTER, there is a synch Thread from MASTER(synch master)
to BACKUP(synch slave) and every connection is synched - good!
But when synch1 is on MASTER, and (e.g. due to a link failure) synch2 is
on BACKUP, there is a synch Thread from BACKUP(synch master) to
MASTER(synch slave). So connections only get synched for synch2 :(
Is there any way to deal with this yet or is there any work on this
currently?
So in the end there were 3 questions ... :)
Thanks for your help and best regards,
Dominik
|
