Server becoming temporarily non-responsive

[TimR]

I have multiple webservers (mostly the same exe's, with a few unique webservers)  running under nginx, and they become non-responsive for anywhere from 30-120 seconds (more or less). Anyone have tips on how to figure out what is going on? They might run for weeks(this is the norm) without incident or it could be days(this is unusual), I see no sign of the webservers restarting, via the performance page. I have a monitoring script, that emails me when they become non-responsive, but I'd like to figure out what the issue is to remove the false positives. The unique webservers, just run and I never have an issue with them.

[urayoan]

Hello Tim. I'm having similar issues, watching this post if anyone gives a clue for what may be happening.

In my case, im using NetTalk 14.36 and Clarion 11.1

What version you using?

[TimR]

I'm using Nettalk version 14.38 and Clarion 11.1.13788

[seanh]

Are you using the IMDD?  That did have a slight problem with memory that would cause odd behavior.  It showed up for me mostly as a secwin issue. 
This has been corrected at some point, I just can't remember when.

There was also a very small memory leak in the window event system for some versions of C11, It only leaked like 5 bytes or something and so took many events to cause a problem.  But a heavy webserver would do that.  That I believe is fixed in C12

[AtoB]

Hi Tim,

have you found out what might be the culprit with this issue (I'm about the ship an api-server to my client, almost same config (nettalk 14.38 in a server farm behind an nginx load balancer, allthough I'm using c10)?

[TimR]

No I haven't figured out what is going on, it doesn't happen frequently enough, for me to prioritize. I did, sort of, recreate this problem by flooding the server with requests using Apache Bench. There appeared to be some sort of staircase pattern going on with slowdowns, that got worse over time. It the server was given a chance to catch up, the slowdown went away.
I wasn't convinced my testing was actually chasing the problem, because I don't think my servers were getting flooded with requests. Also the server responded to all my testing requests, it just got slower and slower.