NPROC limit on every account?: Page 2 - Enhance Control Panel

NPROC limit on every account?

AdamM

PradeepC
https://community.enhance.com/d/269-stress-test

If the enhance team releases a fix for proper setting and saving it should help. You will be able to set, for example, 1 php process that will not return errors anymore and will simply handle less requests but will not crash the server.

In OpenLiteSpeed this setting works. It means a only change untill next ols reboot because it is not saved permanently. You can test it, setting example . max connections to 1

PradeepC

AdamM The max connections set to 1 meaning this will be per user or for the overall system?

AdamM

PradeepC No. For virtual host - domain

Zoinkies

I’m using LSE on cPanel and a lot of my configuration isn’t working on Enhance yet. I believe LS spawns a lot of processes, I have a test server running Apache and weirdly enough seeing less CPU utilisation.

PradeepC

@Adam and @Darren this seems to be a problem that is being faced by mostly everyone.
It would be great if you could help fix the issues soon!

Adam

PradeepC We allocate only 1 website/account and then rest are add-on domains. If we set it to 10 even a single website/account using WordPress will start showing 503 errors with a few visitors and some updates going on in the backend. Never happened with cPanel and Cloudlinux when we were using it previously.

If you use addon domains then they will share a unix user and php-fpm/lsphp instance (like with other control panels) and will be subject to the same limit. Nproc between 10 and 20 in my experience is enough for shared hosting with moderate traffic but this will be affected by the virtual cpu limit as well as the CPU speed.

Adam

PradeepC Adam One more important thing, why is the hard resource limit per website and not per subscription? I firmly believe that it should be per subscription model.

On Enhance the websites for a subscription can be placed on different servers. In future we may be able to offer a setting to apply an optional per-subscription limit by restricting website placement for a subscription to a single server. However you would probably want to use this in conjunction with the per-website limit if your goal is to manage server resources and mitigate high load since the per-subscription limit would need to be much higher.

Adam

AdamM This is the PHP issue I mentioned in another thread. There are too many php processes being created. By default in litespeed it is 100...

The LSAPI worker limit is set to 100 and currently can't be overridden in Enhance - this functionality is coming soon. However it will only spawn workers as required (it will start with 1) and is subject to the nproc limit. If the nproc limit is set, lsphp will not be able to fork processes above that limit.

Zoinkies

Adam What settings would you need for a website to access full CPU speed, but only 2 core?

Adam

Zoinkies

vCPU of 2 will allow CPU usage up to the equivalent of 2 cores - that could be a single process at 200% CPU or 4 processes at 50%. The more processes the website has, the less CPU each process will have available. This is why, if you are limiting CPU, you should also limit nproc to prevent processes from building up.

prasad0889

@Zoinkies and @Adam Is there some different implementation in cPanel/Cloudlinux, because we have never seen the same problem happening there. Same resources same processors but never faced an issue.

Adam

prasad0889 @Zoinkies and @Adam Is there some different implementation in cPanel/Cloudlinux, because we have never seen the same problem happening there. Same resources same processors but never faced an issue.

CloudLinux I believe has a very different implementation of resource limits however if it's the nproc limit you're hitting and it's a hard limit with CloudLinux too then the behaviour should be the same. I would suggest the difference could be down to how long the request takes to execute or potentially the PHP SAPI being used?

prasad0889

@Adam, You are right, there is a longer time in Enhance to execute the request as compare to cPanel. Even when we have the same server, with the same resource limitation, in Enhance when we watch the system load, it increases heavily, whereas in cPanel it didn't happen like that. Also, the other sites are not affected here, but the processing time for the website becomes very long with lots of 503 errors.

I am failing to understand why lsphp requests go in D mode for so long?

Zoinkies

prasad0889 I’ve noticed the same thing, my cPanel performance is a tad-faster than Enhance performance, on identical servers. I’ve also noticed that LS Enterprise doesn’t behave the same either.

Adam

prasad0889 @Adam, You are right, there is a longer time in Enhance to execute the request as compare to cPanel. Even when we have the same server, with the same resource limitation, in Enhance when we watch the system load, it increases heavily, whereas in cPanel it didn't happen like that. Also, the other sites are not affected here, but the processing time for the website becomes very long with lots of 503 errors.

I am failing to understand why lsphp requests go in D mode for so long?

D state means the process is waiting for I/O. Did you possibly set the I/O bandwidth or IOPS resource limits? If not then it's likely to be related to the speed of the server's hard disk which Enhance doesn't influence.

prasad0889

@Adam we run PCIe 4.0 NVMe so I am pretty sure that it's not the bottleneck. We have set it to 10,000 for IOPS and 100 MB/s Disk I/O which is double of the cPanel value where we had 5000 IOPS and 50 MB/s disk I/O. Whenever a website is stuck in D mode, we have to manually increase its CPU to a few more cores and then the load comes down. The same website/websites worked perfectly fine earlier with 2 cores and 2 gb ram on cPanel and is now giving many problems.

Adam

prasad0889 @Adam we run PCIe 4.0 NVMe so I am pretty sure that it's not the bottleneck. We have set it to 10,000 for IOPS and 100 MB/s Disk I/O which is double of the cPanel value where we had 5000 IOPS and 50 MB/s disk I/O. Whenever a website is stuck in D mode, we have to manually increase its CPU to a few more cores and then the load comes down. The same website/websites worked perfectly fine earlier with 2 cores and 2 gb ram on cPanel and is now giving many problems.

It's possible that CloudLinux allows more burst whereas the cgroups limits Enhance applies are max values. If it happens again can you try increasing the IOPS and I/O settings rather than the CPU to see if that helps?

If you're stuck and it's a reproducible problem please feel free to send a support ticket, we can investigate further.

paul-bob

Any news here abou this issue?

Adam

paul-bob Any news here abou this issue?

There are quite a few separate issues being discussed in this thread. Can you elaborate?

If you have a site which is hitting the nproc limit with very low traffic, the other possibility is that it has malware and it's doing something nefarious with every request. Try (as root) attaching strace to one of the PHP processes.

strace -s 2048 -f -p 12345

The -s flag sets the limit of the string length that will be printed. The -f flag will follow any child processes that are spawned.

Where 12345 is the pid of the process you want to examine.

gmakhs

I am facing the same issue as mentioned above, worst thing is that account with 35 nproc limit in DA with CL used to run properly when in enhance they hit 503, it's clearly an issue with how enhance handles limits,

twest

gmakhs yeah CL has "burstable" limits, it's forgiving of overages. If the site can't run properly with 35 nproc AND they don't get a huge amount of traffic, then it's an issue with the code on their site = buggy plugin or theme. Fix the buggy code and they won't hit limits anymore. Fix the buggy code or increase the limits.

« Previous Page Next Page »