Problem with Plesk + Nginx on VMware guest

[ridwan]

I have several VMs on VMware 5.5
All have the same problem with nginx.

I manage to prepare VMs with CentOS 7, install Plesk 12, then start moving clients from old servers to these VMs. Old server was Cloud Linux 6 with Plesk 11.5

First 15-20 clients transfered fine. There's warning about python, but none of my client use that, so I just disable mod_python on new Plesk instances.
Problems start occured after +/- 20 domains (one server at the 18th, the other at the 20th, etc). All with the same error message:
Error during CLI Gate interface usage: transfer closed with outstanding read data remaining.

I tried increasing file open limit as on http://kb.odin.com/en/123174. No luck.

Then I found that the migration process always stuck while it restarts nginx, Because nginx always crashed with message:
*** glibc detected *** /usr/sbin/nginx: malloc(): memory corruption: 0x000000000nnnnnn ***
Searching on Odin KB, and I found this http://kb.odin.com/en/123691
That article says that it was software issue and this had been fixed on Plesk 12.

df -h shows that space was not a problem, all partition only used 2-4%, same as df -i says inode used only 3%. All VMs assigned 16GB ram, and there's a lot of free memory available.

What did I miss?
Any clue would be appreciated.

* I ended up disabling nginx proxy and only use apache. Hundreds of domains can be transferred without problems now to each VMs. Although it's okay to only use apache, but it would be nice if I can speed up things with nginx.

 

[IgorG]

You can try to use new (not officially released yet) Migration Tool. Please read documentation in attach.

 

[ridwan]

Thanks Igor.
But unfortunately I'm still seeing that error message.

I think the problem was not in migration process, but on Nginx. Even if in a server with +/- 20 domains, it will hit that error if I add domains manually (new domains, not migrating from other server).
I got error messages similiar to http://kb.odin.com/en/123691
The warn message about server_names_hash_max_size increases at /var/log/nginx/error.log as I add more domains, from 732, 756, 844, and so on.
I see plesk is actually adding tweak at /etc/nginx/conf.d/aa500_psa_tweaks.conf for nginx hash table, but that warning message still appear. Something like
could not build optimal server_names_hash, you should increase either server_names_hash_max_size: 844 or server_names_hash_bucket_size: 64; ignoring server_names_hash_bucket_size
and then it will hit that memory allocation error that caused Nginx to crash.
*** glibc detected *** /usr/sbin/nginx: malloc(): memory corruption: 0x000000000nnnnnnn ***
*** glibc detected *** /usr/sbin/nginx: malloc(): memory corruption: 0x000000000nnnnnnn ***

Sometimes I can run Nginx after disabling and enabling it
plesk sbin nginxmng -d
plesk sbin nginxmng -e
** it will hit the error here, but if I restart the server from another ssh session, Nginx will eventually start sometimes.

I'm still looking for the possibilities to be able to run Nginx on my Plesk servers.

 

[IgorG]

As far as I see bugreport PPPM-2315 from mentioned KB article has been fixed long time ago. I can assume that this issue may be caused by some type of VMWare specific resources (memory) allocation.
I can only recommend to create a request to support team to do in-depth investigation to find the reason and to fix it. Please create a ticket to support with reference to this discussion at https://www.odin.com/support/request/ .
You may have free support, please check what kind of Plesk license you use for available support options at http://kb.odin.com/en/121580 .
If there’s no free support in your case, you can order Plesk per-incident support at http://www.odin.com/support/buy-support/ Support team will contact you as soon as purchase is processed, and they will do the best to resolve it.
If it is found that your problem was caused by product bug w/o available solution or workaround from Odin, then your purchase will be re-funded.