[Koha-devel] Plack and memcached (was: Sessions terminated at random under Plack)

[Tomas Cohen Arazi]

Ok, yesterday I took some time to fill (and provide patches for):
- Bug 17261
<https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=17261> - Add
memcached configuration info to about.pl
- Bug 17262
<https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=17262> - Plack
on packages is not having memcached set properly

Bug 17261 highlighted something we suspected: the Plack processes don't get
the right configuration variables through ENV for setting the memcached
backend (on packages setup). It was pretty obvious. That's why it is
falling back to CGI::Session even when you set sessionStorage=memcached,
and results in weird permission issues trying to write the session files
somewhere we don't have control of.

Bug 17262 does what is needed to get the Plack processes have the right
MEMCACHED_SERVERS and MEMCACHED_NAMESPACE variables set. It does so by
patching the koha-plack and koha-functions.sh scripts in an obvious way.

The problem is that Plack explodes with 17262 alone. It looks like circular
dependencies issues. But...

This morning I passed QA on:
- Bug 17189
<https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=17189> - Add the
ability to define several memcached namespaces
- Bug 11921
<https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=11921> -
memcached configuration should be moved back to koha-conf.xml

They both change the way memcached configuration is handled and better
reorganize the code around it (basically moving the configuration back to
koha-conf.xml, and removing memcached handling from C4::Context, solving
the circ dep issue).

As Chris mentioned on 17262, moving the memcached configuration back to
koha-conf.xml is the right thing to do.

That's why 17189 and 11921 should be pushed to master ASAP, and I belive
they should be backported down to 3.22 if the rmaints agree and with proper
testing. People supporting 3.22 and 16.05 deployments should get involved
on testing in the stable branches to make sure no regression is introduced
(is it really possible if it is not working at all?).

The only problem is that current setups won't have the koha-conf.xml
entries (memcached_servers and memcached_namespace)[1]. But this is handled
nicely by the current logic, that allows ENV to overload koha-conf.xml
values. So it will work, YAY! (already tested it)

So, IMHO, they could be backported to the stable branches, along with
17262, which will make it effective use of memcached. It shouldn't be
pushed to master as it would be a step backwards and is not needed.

My two cents.

Regards

[1] This might deserve a warning entry in about page IMHO.

El mar., 30 ago. 2016 a las 15:40, Tomas Cohen Arazi (<tomascohen at gmail.com>)
escribió:

Magnus, there's been a lot of movement on the caching layer, and some of
that work has been backported to the stable releases. Also, the
RestrictSessionByIP setting was getting in the middle (
https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=17050).


There's also a problem with memcached initialization in C4::Context that
makes sessionStorage=memcache fail to persist sessions.


My suggestion would be to set sessionStorage to 'mysql' and disable the
RestricSessionByIP syspref.


Regards





El mar., 30 ago. 2016 a las 8:24, Magnus Enger (<magnus at enger.priv.no>)
escribió:

Dear Community,

I am seeing a strange problem, and I'm not sure where to start digging.

I have a (big) server with 30 odd Koha instances. One of these
instances has been running under Plack for quite some time now,
without any problems.

Two new instances have a problem where librarians get kicked out of
the intranet more or less frequently, with a message like "Your
session has expired, please log in again".

We had this problem on 3.22.x.
After upgrading to 16.05.02 it went away.
After upgrading to 16.05.03 last night it came back again.

It is not consistent. Earlier today it looked like you could log in,
click on a link, get thrown out, login again and then things would
work. Later users got kicked out every single time they clicked on a
link in the intranet.

All sites are running off the official Debian packages, on the same server.
Memcached is installed, enabled and used for storing sessions.
Switching SessionStorage to the DB does not stop the problem.
Stopping and disabling Plack for these two instances makes the problem go
away.

I have not found anything interesting in Plack or Apache logs.

Anyone got a hunch what might be causing this? Or where to start digging?

Best regards,
Magnus
_______________________________________________
Koha-devel mailing list
Koha-devel at lists.koha-community.org
http://lists.koha-community.org/cgi-bin/mailman/listinfo/koha-devel
website : http://www.koha-community.org/
git : http://git.koha-community.org/
bugs : http://bugs.koha-community.org/

-- 

Tomás Cohen Arazi
Theke Solutions (https://theke.io <http://theke.io/>)
✆ +54 9351 3513384
GPG: B2F3C15F
-- 
Tomás Cohen Arazi
Theke Solutions (https://theke.io <http://theke.io/>)
✆ +54 9351 3513384
GPG: B2F3C15F
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.koha-community.org/pipermail/koha-devel/attachments/20160907/a09b68bb/attachment.html>