Cobbler API - slow with many hosts

[Sean Carolan]

I wrote a basic python CGI script to parse cobbler data and generate
some graphs.  The script works fine but it takes a really long time to
run due to the large number of hosts stored in cobbler's *.json files.
 Any suggestions for improving performance?  I know I can convert the
*.json data and import into a mysql database, but thought I'd check
here first before rewriting the code.

[James Cammarata]

On Thu, Jun 21, 2012 at 1:55 PM, Sean Carolan <scarolan(a)gmail.com&gt; wrote:
>  I wrote a basic python CGI script to parse cobbler data and generate
>  some graphs.  The script works fine but it takes a really long time to
>  run due to the large number of hosts stored in cobbler's *.json files.
>   Any suggestions for improving performance?  I know I can convert the
>  *.json data and import into a mysql database, but thought I'd check
>  here first before rewriting the code. 
How many systems? What kind of disk is the system running on? If it's
a slower SATA type, you might look at relocating the /var/lib/cobbler
directory to another partition, especially if your server is busy
(lots of http requests to /var/www for instance).

Beyond that, you might look at one of the cobbler nosql backends
(couch and mongo), however those are not widely used and probably not
ready for prime time yet. Dumping the data into one of the various
nosql solutions would probably be easier anyway, since JSON is well
supported there and you wouldn't have to worry about mysql schemas
(unless you plan to use mysql as a bastardized key/value store, which
is quite common too).

A sample of the script might be useful too, to see how you're
requesting the data.

[Sean Carolan]

>  How many systems? What kind of disk is the system running on? If
> it's
>  a slower SATA type, you might look at relocating the /var/lib/cobbler
>  directory to another partition, especially if your server is busy
>  (lots of http requests to /var/www for instance). 
It's about 3000 systems.  This is actually a sync'd copy of a
production system that doesn't get any http requests at all.  It's
running in VMware, and I have a feeling that the bottleneck here is
the back-end storage.  It takes about 15 seconds just to do this:

allservers = xmlrpclib.Server("http://localhost/cobbler_api").get_systems()

I managed to resolve this by using the cPickle module to suck
allservers into a pickle dump file.  To speed it up even further we
put the dump file on a ram disk.  Now the graphs that were previously
taking 30 seconds to render come up in 3-4 seconds.

I would like to publish the code for this at some point, I just have
to check with the corporate overlords to make sure it's ok.  ;-)

[James Cammarata]

On Fri, Jun 22, 2012 at 10:28 AM, Sean Carolan <scarolan(a)gmail.com&gt; wrote:
> > How many systems? What kind of disk is the system running on? If
> it's
> > a slower SATA type, you might look at relocating the /var/lib/cobbler
> > directory to another partition, especially if your server is busy
> > (lots of http requests to /var/www for instance).
>
>  It's about 3000 systems.  This is actually a sync'd copy of a
>  production system that doesn't get any http requests at all.  It's
>  running in VMware, and I have a feeling that the bottleneck here is
>  the back-end storage.  It takes about 15 seconds just to do this:
>
>  allservers = xmlrpclib.Server("http://localhost/cobbler_api").get_systems()
Yeah, if you have the ability I'd recommend a raw device mapped LUN.
That will speed things up considerably.