I actually don't believe it is apache, but rather a script that is causing
the problem. The site is basically a web application that doesn a number of
things from pdf generation to statistical calculations on work being done.
I don't think that any one script causes it in general, but perhaps when the
data being pulled is somehow out of the ordinary. There is really no more
than 15 people on it at the most. I think that knowing which parts to look
at in the proc status files will allow me to write a script to create a
snapshot of what is happening. I'm close, I can feel it. :-)
-----Original Message-----
From: flalug@nks.net [mailto:flalug@nks.net]On Behalf Of Bill Triplett
Sent: Wednesday, March 10, 2004 4:43 PM
To: flalug@nks.net
Subject: Re: [flalug] more memory questions...
On Mar 10, 2004, at 2:18 PM, Larry Brown wrote:
> OK, top is not really what I'm looking for because I'm wanting to
> script a
> process taht upon vm overload approach a script checks the pid numbers
> from
> the httpd access log, then calls something like ps and captures the
> response
> and evaluates which are the highest etc.
If you suspect apache's httpd process, there is a tool called 'ab'
(apache bench) that will allow you to simulate all kinds of traffic to
your web server (complete with cookies, form posts, etc..)
You could fire up top with a high refresh rate then unleash ab on the
server and watch the httpd processes jump around ;-)
Good luck...
> Server sits at rest with 0 swap in use (according to free and top)
> Then 5
> minutes later, I have 50 to 60% swap in use. I have looked at top
> using M
> to sort by memory usage and nothing seems excessively high. (although
> httpd
> and mysql are at the top) I chased this a couple of weeks ago without
> finding anything solid. Since it did not noticeably affect the
> performance
> of the server, I wrote it off to some internal process of swapping
> items by
> the kernel that are normal. However, today the server locked up from
> excessive memory usage. I have to get to the bottom of this and know
> how to
> sluth this in the future. I'm hoping that by creating a script that
> pulls
> together all the relavent pieces of information, I can simply run the
> script
> and get a readout of exactly what is happening.
>
> I have done a lot of googling today and came accross a pretty good
> article
> on linux-tutorial.info.
> http://ww.linux-tutorial.info/cgi-bin/display.pl?90&0&322&0&3
>
> However, I'm still trying to get this down. All I really need to know
> is
> how to get...
>
> memory used by pid in physical memory
> memory used by pid in swap memory
>
> From the article (from my understanding) ps -p <processid> -o "sz vsz
> rss"
> gives the process core image size [size of the process at its inception
> before growing as it performs its functions?], size of total memory
> for the
> process [size of the process including virtual and physical memory?],
> and
> size of physical memory taken up by the process.
>
> The problem is that when I run free I get...
>
> Swap: 265032 2296 262736
>
> and when I run ps -p 3331 -o "sz vsz rss" I get ...
>
> 5491 21964 9428
>
> according to that the difference between the physical memory 9428 and
> the
> total memory size 21964 would have to be 12536 which is larger than
> the 2296
> reported in use in swap by the "free" command.
>
> The only descriptions I come accross in postings asking about memory
> use
> come back with an explination of what free returns to explain how much
> memory is actually available etc. I need to single out these
> processes and
> be able to account for their usage.
>
> Sorry for the book, but I'm looking for something pretty specific here
> and I
> don't know how else to explain it.
>
> -----Original Message-----
> From: flalug@nks.net [mailto:flalug@nks.net]On Behalf Of Eben King
> Sent: Wednesday, March 10, 2004 1:15 PM
> To: Linux
> Subject: Re: [flalug] more memory questions...
>
>
> On Wed, 10 Mar 2004, Larry Brown wrote:
>
>> Does anyone know how to get the exact amount of memory used for a
>> process?
>> In ps it renders a percentage number.
>
> Is "top -b -n 1 | grep '^ *$pid ' | awk '{ print $5 }'" what you want?
> That's RSS. I suppose you could take top's percentage and multiply it
> by
> the total amount of memory, but the percentage is at most three
> significant digits, usually fewer.
>
> Also, /proc/<pid>/status is interesting. For an xterm, I get:
>
> ...
> VmSize: 2448 kB
> VmLck: 0 kB
> VmRSS: 1436 kB
> VmData: 348 kB
> VmStk: 24 kB
> VmExe: 588 kB
> VmLib: 1420 kB
> ...
>
> So I guess it kinda depends what you mean by "memory".
>
> --
> -eben ebQenW1@EtaRmpTabYayU.rIr.OcoPm home.tampabay.rr.com/hactar
>
> This message was created using recycled electrons.
>
>
>
>
This archive was generated by hypermail 2.1.3 : Fri Aug 01 2014 - 20:15:21 EDT