I hope someone here can shed some light on this issue.
After rethinking this and having read some of the earlier threads I suspect
that a hot-config-reload is too complex to fit in with how Pound works, and
we probably don't need it anyway. I like the suggestion of using the ha_port
to not only connect to but try to read a byte, big-endian 16/32 bit int or
whatever to change the priority of the said backend.
Off course, returning a priority of 0 should prevent new sessions to be
assigned to the backend while still honouring the existing sessions to the
backend.
I don't think that this will introduce a secrity hole nor will it conflict
with existing Pound logics. We only need an extra parameter on the backend
definition to tell Pound to treat the ha_port this way as opposed to how it
works now.
I would like to stress that there are lots of web-applications where
session-migration is impractical or even impossible and as such can not
cold-turkey kill backends while active sessions exist. For such applications
Pound would be an incredible asset even if load balancing is not needed.
Having a two-backend setup with a 1/0 priority setting and being able to
modify the priorities to allow for gradual server swaps (and thus, software
updates) makes all the difference.
Although loadbalancing is important to us the version swapping is also. I
regularly have to wait up until three in the morning now to wait for the
server to go idle :-( To make things worse, if we have overlooked a
functional bug in a version we have put online and someone notifies during
the day we have to choose to either restart backends and kill active
sessions or have the affected users wait until the next day.
Pound could really help out here if such a facility would be present.
Regards,
Silvio Biernman
(at) -----Original Message-----
(at) From: Silvio Bierman [mailto:sbierman(at)jambo-software.com]
(at) Sent: 02 December, 2005 13:40
(at) To: pound(at)apsis.ch
(at) Subject: [Pound Mailing List] Status on starving a backend?
(at)
(at)
(at) Hello all,
(at)
(at) Since a few days we are using Pound as loadbalancer in front of three
(at) backends running a session oriented web-application. I am using IP based
(at) session tracking (the COOKIE version for me appears unreliable, I
(at) regularly
(at) get dropped sessions, but IP is not a problem). It works like a
(at) charm and is
(at) a great relief from Apache...
(at)
(at) I browsed the archives and saw there has already been some discussion on
(at) this topic. I would like to know the status on this since it is crucial to
(at) me to be able to take the backends out of service one by one, update the
(at) software, restart them and put them back in the pool.
(at)
(at) Obviously, taking a backend out of service does not mean killing
(at) it directly
(at) but just not assigning new sessions to it so it will go idle by
(at) itself. Part
(at) of our system is an internet-survey-based market-research tool so
(at) there are
(at) many users almost around the clock that stay around for a few
(at) minutes up to
(at) about 30-45 minutes).
(at)
(at) Changing the config and doing "force-reload" seems to simply restart the
(at) server and drop all current sessions which is unacceptable.
(at)
(at) Is there already a way to do this? If not, will support for such a thing
(at) become available?
(at)
(at) Regards,
(at)
(at) Silvio Bierman
(at)
(at)
(at) --
(at) To unsubscribe send an email with subject 'unsubscribe' to
pound(at)apsis.ch.
(at) Please contact roseg(at)apsis.ch for questions.
(at) http://www.apsis.ch/pound/pound_list/archive/2005/2005-12/1133527229000
(at) -----Original Message-----
(at) From: Silvio Bierman [mailto:sbierman(at)jambo-software.com]
(at) Sent: 02 December, 2005 13:40
(at) To: pound(at)apsis.ch
(at) Subject: [Pound Mailing List] Status on starving a backend?
(at)
(at)
(at) Hello all,
(at)
(at) Since a few days we are using Pound as loadbalancer in front of three
(at) backends running a session oriented web-application. I am using IP based
(at) session tracking (the COOKIE version for me appears unreliable, I
(at) regularly
(at) get dropped sessions, but IP is not a problem). It works like a
(at) charm and is
(at) a great relief from Apache...
(at)
(at) I browsed the archives and saw there has already been some discussion on
(at) this topic. I would like to know the status on this since it is crucial to
(at) me to be able to take the backends out of service one by one, update the
(at) software, restart them and put them back in the pool.
(at)
(at) Obviously, taking a backend out of service does not mean killing
(at) it directly
(at) but just not assigning new sessions to it so it will go idle by
(at) itself. Part
(at) of our system is an internet-survey-based market-research tool so
(at) there are
(at) many users almost around the clock that stay around for a few
(at) minutes up to
(at) about 30-45 minutes).
(at)
(at) Changing the config and doing "force-reload" seems to simply restart the
(at) server and drop all current sessions which is unacceptable.
(at)
(at) Is there already a way to do this? If not, will support for such a thing
(at) become available?
(at)
(at) Regards,
(at)
(at) Silvio Bierman
(at)
(at)
(at) --
(at) To unsubscribe send an email with subject 'unsubscribe' to
pound(at)apsis.ch.
(at) Please contact roseg(at)apsis.ch for questions.
(at) http://www.apsis.ch/pound/pound_list/archive/2005/2005-12/1133527229000
|
