[mapserver-users] Slow and degrading performance with radar (but not satellite) on mapserver-6.0.1-3_0.el6.x86_64

[Stephen Woodbridge]

Hi Paul,

So you have a complex historical "mess" and its not clear where the 
performance issue is. So you need to divide the problem into small 
problems that you can verify are or are not contributing. I would start 
with something like this:

Take one slow image request and try that as cgi or cli and not fcgi and 
turn on debugging.

   - copy and rename you mapfile so it doesn't mess with the production 
requests

   - turn on the debugging and send it to stderr in the debug mapfile

  - you can manually run that image request from the commandline like:

if mapserv is not in your path you might need to find it and your the 
path to it below

mapserv -nh QUERY_STRING="everything after the ? in the original query" 
 >junk.png 2>error.txt

-nh suppresses headers from being output before the image data

error.txt will be stderr output and should contain the debug messages

Then report back what you find out.

-Steve W

On 7/22/2020 1:47 PM, English Paul wrote:
>
> Hi,
>
>    Newbie here, just got a work assignment to look into some 
> performance issues with mapserver-6.0.1-3_0.el6.x86_64 – specifically, 
> at one time, it rendered radar images very fast, then it degraded and 
> seemed to consume a lot more CPU. So – the EC2 instance side was 
> upgraded significantly – and it still performs badly and uses a lot of 
> CPU. It also seems to be getting slowly worse over time (days/months, 
> not seconds/minutes). I asked this on IRC, but it looks like email 
> might be a better route.
>
>   During all of this, the same mapserver instance renders satellite 
> images quickly. These seem to be a similar, or in some cases larger 
> size png to start with, rendered onto the same final map/size.
>
>   So – the obvious answer is that it isn’t using the CPU to 
> render/re-render, but rather spending it on something else – I/O most 
> likely – eg: a network request, disk I/O, SQL query?
>
>   The previous person working on it tried turning on debug at various 
> levels – but unfortunately that made it even slower, making it tricky 
> to answer “what is making it slow when debug is turned off?”
>
>  My first instinct was to try an strace and nothing was obvious. Next 
> up – a flame graph from strace, and/or trying dtrace – but my 
> understanding is that dtrace is a little weak on RHEL/CentOS/Amazon 
> Linux 6.0 **and** I’m not particularly good at that. Also, we’re using 
> fcgi, so attaching to the correct process is a bit tricky.
>
>   My next instinct was to look at release notes and see if the current 
> stable has anything fixes/improvements that directly address this – 
> there aren’t any that are obvious to my eyes, but you developers have 
> been busy! So many things! Including some performance fixes and one 
> “significant” performance fix.
>
> Current config file:
>
> AddHandler fcgid-script fcgi
>
> FcgidIPCDir /var/run/mod_fcgid
>
> FcgidProcessTableFile /var/run/mod_fcgid/fcgid_shm
>
> FcgidMaxProcesses 10
>
> FcgidMaxProcessesPerClass 10
>
> FcgidMaxRequestInMem 196608
>
> FcgidInitialEnv PROJ_LIB /usr/share/proj
>
> FcgidInitialEnv LD_LIBRARY_PATH "/usr/local/lib:/usr/pgsql-9.1/lib"
>
> So – suggestions for my next move? I currently plan to take a quick 
> swing at building 7.6 for RHEL 6.0, knowing there might be old 
> libraries and whatnot that make that a non-starter. Of course – we’ve 
> got other infra running on this same instance, so upgrading everything 
> is a much bigger task.
>
>  Strace flame graph?
>
>   Stretch and try dtrace?
>
>   A better way to use debug?
>
>   Something else I’m missing – eg: differences between the image types 
> that make them perform so differently?
>
> Thanks,
>
> Paul
>
>
> _______________________________________________
> mapserver-users mailing list
> mapserver-users at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/mapserver-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/mapserver-users/attachments/20200722/585ad646/attachment.html>