Z-Push Zimbra Backend (Push Email, Contacts, Appointments & Tasks)

Quote:

---

Originally Posted by janc

I tried upgrading to release 54. And approx. 6 hours later my server crashed. So perhaps I'm seeing the same problem as you. After the crash I downgraded back to release 53. Which I have had running for long a long time without any problems.

---

Which z-push version are you using ? Did you use the matching set of files from the release 54 distribution (it contains 3 different sets) ?

If it happened just once, it could also just be an unfortunate co-incidence. Perhaps your server would have crashed anyway.

Take a look at your htppd.conf file and pay attention to these settings.

KeepAlive Off

# prefork MPM
# StartServers: number of server processes to start
# MinSpareServers: minimum number of server processes which are kept spare
# MaxSpareServers: maximum number of server processes which are kept spare
# ServerLimit: maximum value for MaxClients for the lifetime of the server
# MaxClients: maximum number of server processes allowed to start
# MaxRequestsPerChild: maximum number of requests a server process serves
<IfModule prefork.c>
StartServers nn - should be a few more than the number of users you have
MinSpareServers 2 - if all servers processes are in use, more will be started up until there are at least this many idle processes
MaxSpareServers 3 - idle servers are killed off until there are just this many left
ServerLimit nn+10 - Maximum number of processes to run
MaxClients nn+10 - Maximum number of clients to handle simultaneously
MaxRequestsPerChild 200 - how many requests each server process should be allowed to run before getting killed/restarted
</IfModule>

You can do your own reading on this configuration - this is just a ballpark. For each process under apache/php/z-push/zimbrabackend you can expect it to run to approx 16-20MB depending on how much email the phone is typically syncing - check your own server for how much memory is allocated to each process using "top". Note that this will list the highest amount the process used at any time - not how much it is currently actively using. Apache/PHP does not release memory back to the OS once it is no longer needed - it holds onto it in anticipation of possibly needing it again.

The critical configuration if you suspect a memory leak is affecting you is the last one (MaxRequestsPerChild). The default value of 0 means that apache will never stop a child process. That is until all memory is exhausted and the server crashes. Setting this value to a real value (200 in the example above) will tell apache to kill off a child process after it has handled 200 requests. A new child process will be forked to replace it (assuming it is needed based on MinSpareServers) Obviously setting this too low will cause overhead for the OS stopping/starting processes - but on a z-push server this is not as critical as it would be on a high traffic web server. Setting it too high would allow memory leaks to build up in the process - possibly enough to crash your server.

Armed with the ballpark figure per-client figure you can work out how many clients your server could handle from a memory point of view.

Arghhh! Maybe I spoke to soon. I am back to restarting httpd every couple of hours :/ I am pretty sure its those darned samsung Galaxy S3's but until I deploy my z-push 2.x in parallel I will not be able to tell for sure.

Quote:

---

Originally Posted by sangamc

Arghhh! Maybe I spoke to soon. I am back to restarting httpd every couple of hours :/ I am pretty sure its those darned samsung Galaxy S3's but until I deploy my z-push 2.x in parallel I will not be able to tell for sure.

---

According to the z-push developers, pretty much anything with android 4 is a no-no with z-push 1.5

All you have to do is search their forums for issues people have with Galaxy S2, S3, and several HTC phones.

I would try to get up and running with z-push 2 if you can at all. There are still some minor issues but they are getting through them.

Its unfortunate since they are all using activesync that there is no consistency. And when I ask my users what kind of phones they have, its either iphone "or the one that isnt iphone" without them having the box with them, they have no clue what they carry in their pockets :)

I see the following warning in the debug.txt for z-push 1.5.x and also for z-push 2.x and I am not sure if it is a problem or something that can be safely ignored.

zimbra.php:792 Undefined index: change_attribute_token (8)

Any ideas what it means?

Quote:

---

Originally Posted by sangamc

I see the following warning in the debug.txt for z-push 1.5.x and also for z-push 2.x and I am not sure if it is a problem or something that can be safely ignored.

zimbra.php:792 Undefined index: change_attribute_token (8)

Any ideas what it means?

---

This indicates that your SOAP sessions are getting closed prematurely - and when z-push is assigned a new session it does not contain a change token on the first response it gets back from the backend.

Another common symptom you might see in this scenario is that if you are looking at an email in the WEB Client, it will close all of a sudden, and you will be back to looking at just a list of messages with a blank reader pane rather than the message in the reader pane.

This relates to the number of allowed SOAP sessions per user. This was highlighted to in a ticket (not by me) and was fixed in release 7.2 - where you can now increase the number.
Bug 75338 &ndash; Providing the ability to increase the number of Soap sessions
I set the value to 7 on my server and typically don't see the issue any more.

In my experience it is usually seen where a user (me) has more than one device doing "push" syncing at the same time. The other factor that causes it is using push with wi-fi on iPhones/iPads/etc. as the wi-fi policy on many of them keeps dropping the connection, and the mail client keeps opening new ones, and the server keeps having to kill the oldest session for that user to create a new one for them.

Thanks for the reply! most of the users where that message pops up have both iphone & ipads. I was worried that that error was the reason that there is such a huge delay in email sync. Most if not all of my users have to either manually sync or mail gets to their device 30 - 40 minutes after it gets to the zimbra server. I'll keep digging and see if I can resolve the delay issue.

:confused:

I have a strange problem wth my test z-push2 server that I am prepping for deployment. All the standard tests work. I can connect via browser and login getting the z-push version page. I can connect a device (ipad or iphone) with no problems, but when I open the inbox it is usually empty. If I hit refresh in the folder view, The inbox will populate again, but when I try to go into it, all the mail will disappear again :(

I have used z-push-admin to verify there are no errors with the device. Using z-push-top I can see the device connecting and polling successfully. But still empty inbox and missing folders (except inbox folder and outbox folder remain)

The only error in z-push.log is the following

LoopDetection->ProcessLoopDetectionPreviousConnectionFailed(): Command 'Sync' at 12/11/2012 21:15:22 with pid '21339' terminated unexpectedly or is still running.

Any ideas or pointers would be fantastic. I want to jump to z-push 2 to help support more devices and get past memory problems in 1.5
Thanks

Does your web server have write permissions for creating sub folders in the state folder, and permission to create/update files in those folders ?

Yes Apache user and Apache group is the owner of /var/lib/z-push/ and had 777 permissions set. I can see device folders being created and the users file being updates as I add more test devices.

Quote:

---

Originally Posted by liverpoolfcfan

Does your web server have write permissions for creating sub folders in the state folder, and permission to create/update files in those folders ?

---