On Wed, 16 Jul 2008, "David Coulson" <david@xxxxxxxxxxxxxxxx> wrote:
> The stickyness works fine on the local server, however when it syncs
> over to the backup, the timeout is wrong.
> On master:
> TCP 29:53 NONE 22.214.171.124:0 126.96.36.199:80 188.8.131.52:80
> On slave:
> TCP 02:57 NONE 184.108.40.206:0 220.127.116.11:80 18.104.22.168:80
Hrm... this is due to the synchronised connections being given a 3
minute timeout on the the backup so that backup director doesn't end
up stuffed full of closed connections. Incidentally, a patch has just
been accepted to make the synchronised connections go away on the
backup when they go away on the master director (ie. when they reach
CLOSE, CLOSE_WAIT or TIME_WAIT). But that won't solve this.
> Obviously there are instances when the backup connections timeout but
> they are in place on the master, so in the event the master fails, the
> slave doesn't have the connections in it's table.
With a non-open but persistent connection (say one which has been
closed for 5 minutes but still has 25 minutes of persistence left)
what do you see on the backup director? Anything?
If I recall correctly, they'll reach 0 seconds remaining and then
spring back to 3 minutes, but I could be wrong.