Fog using a huge amount of memory and failing

[jstephen]

I’ve been seeing a lot of sluggishness in the web UI for the past several updates, but didn’t think much of it. I updated to 1.5.3 earlier this week, and the memory usage from Fog is huge, and seems to now be effecting not just the web UI but also the base Fog functions - imaging is slower, tftp connections are timing out, etc.

I’ve assigned the Debian virtual machine that is running Fog 24576 MB of memory, and even with that much it is still trying to use more. Looking in top, apache2 has dozens of processes running and is sucking up all the memory. This is just when the server is sitting idle.

Any suggestions on how to troubleshoot this problem?

[george1421]

Please upgrade up to 1.5.4. There has been several fixes that will impact the memory usage (in a good way).

[jstephen]

OK, I’ve updated to 1.5.4, and I’d say it is at least marginally better. In the web interface, I tried opening 10 host pages, and it immediately bogged down again and wouldn’t load anything for a minute or two.

I can’t really say whether the problem is resolved yet since it always runs a bit better after restarting apache.

I’ll check it throughout the day and post back if there are continued problems.

Thanks!

[george1421]

@jstephen How much vMemory do you have on this VM as well as the vCPU?

Lastly how many fog clients connect to this fog server?

[Quazz]

Is that 24.5GB of RAM?

And apache2 is using most of it?

What about php or php-fpm?

Did you manually change some of the apache settings?

What does apache2ctl -l show?

[jstephen]

@george1421 It has dynamic memory in a range from 4096 MB - 24576 MB.

For the vCPU, I just went with the defaults from Hyper-V I think - this was setup about 4 years ago. The description in the settings says:
Processors: 1
NUMA nodes: 1
Sockets: 1
Numa Topology:
Maximum number of processors: 12
Maximim amount of memory 13348 MB

We’ve got right around 500 clients.

[jstephen]

@quazz Yes, 24.5GB of RAM.

Sitting idle, there are 25 apache2 processes using a total of 17% of the available memory. Apache is also using a lot of the cpu percentage as well - 64%.

php-fpm also has about 3 processes, but looks to be taking up only about 1.5% of the RAM.

I think a few things have likely been changed in the apache settings. The server was upgraded from Debian 7 last year which necessitated a few changes I think.

apache2ctl -l shows: core.c, mod_so.c, mod_watchdog.c, http_core.c, mod_log_config.c, mod_logio.c mod_version.c, and mod_unixd.c

[Wayne Workman]

@jstephen Can you add this to your apache configuration file and see if the issue resolves itself:

MaxRequestWorkers 100
MaxConnectionsPerChild 25

You’ll need to restart apache after adding this.

[jstephen]

@wayne-workman I added those to mpm_event.conf which seemed like the right place to put them, but I’m not entirely sure!

I can’t really see any difference… maybe it is a tad more stable? What I’ve noticed more than anything with this testing though is that when I restart apache, it runs way, way better for a while. Like this morning, each page in the Fog UI was taking about a minute to load; I restarted apache and it was down to 5 seconds per page even when opening multiple pages at once.

[Wayne Workman]

@jstephen Try adding the configuration lines to the bottom of this file: /etc/apache2/apache2.conf and then restart apache with systemctl restart apache2 and see what happens. Also, how many cores does your fog server have?

[george1421]

@jstephen said in Fog using a huge amount of memory and failing:

Yes, 24.5GB of RAM.
Sitting idle, there are 25 apache2 processes using a total of 17% of the available memory. Apache is also using a lot of the cpu percentage as well - 64%.
php-fpm also has about 3 processes, but looks to be taking up only about 1.5% of the RAM.

There has to be something else going on here. These settings are a bit out of wack.

First on the questions I asked you. You might consider 2 vCPUs at 4GB of ram. That is all FOG should use. If you have a 1000 or more hosts then we might need to bump up the numbers a bit.

With FOG 1.5.1 and later FOG now uses pgp-fpm for its pgp engine and moved away from apache’s built in php engine. This should lower the cpu requirements of apache by off-loading php requests to php-fpm. With apache using the majority of the CPU still, it almost sounds like the php activities have not shifted to php-fpm.

What I would like you to do is create a file in the root of the web server directory, typically /var/www/html name that file info.php and make it world readable. In that file place the following code:

<html>
<body>

<?php
 phpinfo();
?>

</body>
</html>

Once its created navigate to that file by using this url http://<fog_server_ip>/info.php

I’m expecting to see a this field set to

and not this

The first screen shows that php-fpm is managing the php process. The second screen is the way fog was setup before version 1.5.1. I would expect php-fpm to have the workload here, not apache.

ref: https://forums.fogproject.org/topic/10717/can-php-fpm-make-fog-web-gui-fast

[Joe Gill]

@george1421
I have also been experiencing high memory usage and sluggishness. I completely lost my interface and had to restart PHP in order for the interface to return. I will also try your suggestion here to see what is going on.

[george1421]

@joe-gill If you are not on FOG 1.5.4, but are using 1.5.1 to 1.5.3 there is a fix that is needed to php-fpm to address memory bloat under certain conditions.

[Joe Gill]

@george1421
I actually upgraded to FOG 1.5.4 and the issue persisted.

[george1421]

@joe-gill ok we need to collect a bit more info here. It sounds like a trend is starting.

1. What OS is the host OS for FOG?
2. What is your memory installed in the fog server?
3. What process is using the most memory?
4. What process is using the most CPU utilization?
5. Number of clients you have that has fog installed
6. Fog check in interval Fog Configuration -> Fog Settings -> Fog Client -> FOG_CLIENT_CHECKIN_TIME (if you have less than 300 [5 minutes] you might want to bump that up to 300 or 600 seconds).

[Joe Gill]

@george1421

Let me know if their is anything else you would like me to check. Thanks!

[george1421]

@joe-gill Ok you are using php-fpm. That is good. Now we need to understand what is consuming the memory. I had a previous post where I was asking specific questions about your current setup. Please respond to that… and

Depending on your distro there is a php-fpm file that fog creates it should be in /etc (in a sub directory) called www.conf. Under centos 7 its in /etc/php-fpm.d Lets review that file, towards the middle there should be this line uncommented.

pm.max_requests = 2000

Let make sure its not commented out.

[Joe Gill]

@george1421
Running on a Hyper V Virtual Machine
FOG OS - Debian 9.4
Memory - 4 GB
CPU - 2 CPUs allotted

I used Top to determine these. If their is a better command let me know and I will certainly run that and report back.

Memory Usage (in order from most to least) –
Apache2 (20 percent)
mysql
FOGImageSize
FOGPingHosts
FOGSnapinHash

CPU Usage –
Apache2

[george1421]

@joe-gill Can I get to bits of info I forgot to ask?

1. Top CPU usage service
2. Number of clients you have in your FOG environment
3. Fog check in interval Fog Configuration -> Fog Settings -> Fog Client -> FOG_CLIENT_CHECKIN_TIME (if you have less than 300 [5 minutes] you might want to bump that up to 300 or 600 seconds).

[Joe Gill]

@george1421
Top CPU usage service is Apache2.
Total hosts is ~500.

Where is the www.conf file located in Debian? I searched high and low and could not find it. I found something in the /etc/php/7.0/ directory but it didn’t have pm.max_requests.

Thanks!