persist the syncml state in multiserver environments

I'm not sure if this is a bug therefore I write this feature

Thu, 25 Sep 2008 19:15:45 +0000

I'm not sure if this is a bug therefore I write this feature request: In a setup which uses multiple loadbalanced servers with a shared session backend (memcached) it can happen that the loadbalancer redirects requests which belong to the same sync session to different servers, e.g. when the database (calendar for e.g.) is large the sync is divided into more then one request: Here is a log for such a case: DEBUG: Backend of class SyncML_Backend_Horde created DEBUG: Started at 2008-09-25 20:49:48. Packet logged in /tmp/sync/client_10.wbxml DEBUG: New session created: 3baf28e3f425a8aa4cb8281f978e16ee DEBUG: Checking authentication for user riederd DEBUG: Authenticated: yes; version: 1.2; message ID: 1; source URI: IMEI:xxxxxxxxxxxxxx; target URI: .... [....] DEBUG: Return message completed DEBUG: Finished at 2008-09-25 20:49:48. Packet logged in /tmp/sync/server_10.wbxml DEBUG: Backend of class SyncML_Backend_Horde created DEBUG: Started at 2008-09-25 20:49:55. Packet logged in /tmp/sync/client_11.wbxml DEBUG: Existing session continued: 3baf28e3f425a8aa4cb8281f978e16ee DEBUG: Authenticated: yes; version: 1.2; message ID: 2; source URI: IMEI:xxxxxxxxxxxxxx; target URI: .... DEBUG: Created device class SyncML_Device_Nokia DEBUG: Created device class SyncML_Device_Nokia DEBUG: Received from client. DEBUG: Handle for state Sync DEBUG: Creating output for server changes in database ./notes DEBUG: Created device class SyncML_Device_Nokia DEBUG: Compiling server changes from 1970-01-01 01:00:00 to 2008-09-25 20:49:48 [...] DEBUG: Sending add from server: 20070628193702.@webmail.tugraz.at DEBUG: Sending add from server: 20070628193732.@webmail.tugraz.at DEBUG: Maximum message size 65535 approached during add; current size: 65072 DEBUG: Return message completed DEBUG: Finished at 2008-09-25 20:50:00. Packet logged in /tmp/sync/server_11.wbxml the next request is switched to another server, which results into a authentication error: DEBUG: Backend of class SyncML_Backend_Horde created DEBUG: Started at 2008-09-25 20:50:16. Packet logged in /tmp/sync/client_10.wbxml DEBUG: New session created: 3baf28e3f425a8aa4cb8281f978e16ee DEBUG: Invalid authentication DEBUG: Authenticated: no; version: 1.2; message ID: 3; source URI: IMEI:xxxxxxxxxxxxxx; target URI: ERR: Not authenticated while processing MapItem ERR: Not authenticated while processing MapItem [....] ERR: Not authenticated while processing MapItem DEBUG: Return message completed DEBUG: Finished at 2008-09-25 20:50:17. Packet logged in /tmp/sync/server_10.wbxml It seems that $_SESSION['SyncML.state'] gets lost between the requests when they go from different servers. In other Horde applications (like IMP) it is no problem when different requests go to different servers. Would it be possible to keep the Session active for different servers?

Synching uses regular PHP sessions, with the only difference

Thu, 25 Sep 2008 19:34:12 +0000

Synching uses regular PHP sessions, with the only difference that the session id is not passed by cookie or url parameter, but in the sync packet. Is synching not using the configured session handler in your case?

> Synching uses regular PHP sessions, with the only differen

Thu, 25 Sep 2008 20:08:44 +0000

> Synching uses regular PHP sessions, with the only difference that the > session id is not passed by cookie or url parameter, but in the sync > packet. > Is synching not using the configured session handler in your case? well I think/hope that it is using the same session handler as all the other Horde parts. Here is what I've set in conf.php for sessions: $conf['session']['name'] = 'Horde'; $conf['session']['use_only_cookies'] = true; $conf['session']['cache_limiter'] = 'nocache'; $conf['session']['timeout'] = 0; [...] $conf['sessionhandler']['params']['track'] = true; $conf['sessionhandler']['type'] = 'memcache'; $conf['sessionhandler']['memcache'] = true; I can not think at anything how syncing should use something else?

The _backendMode is not set on the server when a loadbalance

Thu, 02 Oct 2008 21:13:19 +0000

The _backendMode is not set on the server when a loadbalancer directs the client to another machine. The attached patch fixes the problem.

Committed, thanks.

Tue, 07 Oct 2008 12:47:30 +0000

Committed, thanks.

Changes have been made in CVS for this ticket: http://cvs.h

Tue, 07 Oct 2008 12:47:32 +0000

Changes have been made in CVS for this ticket: http://cvs.horde.org/diff.php/framework/SyncML/SyncML/Backend/Horde.php?r1=1.8.2.14&r2=1.8.2.15&ty=u