Ganglia GPU Monitoring Enhancements

[ranacseruet]

• This aims to generalize the number of custom php graph files required to represent custom metric graphs. This eliminates the limit of hard coded number of GPUs ganglia-web can handle.

[ranacseruet]

Before, for cluster node with more than one gpus, custom graph for a metric(say ‘util’) had to be named as gpu0_util.php, gpu1_util.php, gpu2_util.php , which was limited to the hard coded numbers added up to. Thus, to make this process simpler, I have added an small modification under ‘ganglia-web’ module, which shouldn’t effect any additional functionality, but facilitate this particular issue to all other modules(which could have multiple device attached in single node) as well

[dpocock]

@vvuksan do you think this is OK to merge? It is not a very big change.

[dpocock]

@ranacseruet can you please contact @vvuksan on IRC to see if he is happy to merge this? I already accepted the other pull request.

[vvuksan]

Problem I have with this is that it relies on optional graph name having a number in it so if for example someone calls their graphs apache2.2_report it may get caught. I would prefer we implement this differently. We should do this in the actual optional graph config and not in graph.php. There is no reason why you shouldn't be able to traverse the RRD directory to find out all the GPU names and numbers

[ranacseruet]

@vvuksan , as per the design document that I shared earlier(http://goo.gl/kLeumX), I have modified the implementation, which reads mapping settings from json file. Please review this change. Also, if this is OK, to get the gpu module working, there will be a json file needed, which I will create pull request separately. You can review that json settings as well: http://goo.gl/MgHRfB . Thanks.

[ranacseruet]

I found it was showing some conflict due to recent other changes on master branch. Just merged the conflicts