From: Phil R. <P.R...@rl...> - 2003-04-08 11:10:27
|
Hi all, I'm about to leave the job where I've been playing with ganglia, so I thought I'd check if there are any modifications I've made which are more widely useful. Unfortunately I forked off at 2.5.2 and modified heavily, so getting patches isn't plausible, but you can view the pages at http://ganglia.gridpp.rl.ac.uk/ (feel free to add it to the website if you want!) and I've dropped the source at http://frogtop.com/temp/ganglia-webfrontend.tgz for anybody interested. My modifications include the following; - lots of aesthetic changes(!) - lots of fixes so that PHP doesn't generate 1MB of errors per page! - quoting of hash indices (now also in mainline) - wrapping many code chunks in 'if (isset())' for better behaviour when data structures are 'real-world' incomplete - more use of htmlspecialchars() or rawurlencode() as appropriate, and other cleansing of untrusted input before use [not guaranteed complete, but improved!] - metric related changes - new default metric '[none]' for fast render of cluster pages - ability to choose 'special' metrics (cpu, load, mem) on the cluster pages (single page view for all nodes in cluster) - new 'special' metric 'network' (combines bytes in/out) [multiplied up to bit/s to avoid confusing network people!] - omit dull metrics (such as those combined into special metrics) - different auth framework - users with restricted view can only access summary data - no per host data (so can't view host page, or get metrics on cluster page) - connections from particular addresses always get full view, otherwise you get asked for a password if you try and access restricted pages (marking clusters as private is still in there, too) - if a cluster only contains one system, you go straight to the host view, instead of the cluster view (unless restricted) - the gmetad sources for each cluster are highlighted in the cluster view - slightly different handling of up/down nodes, eg. to handle the case where a box moves IP address (and so has one up and one down entry) slightly more sensibly - low resolution month graph bugfix (see mail from last week) - extra 'special pages' for clusters (eg. our OpenPBS stuff, which gives additional graphs showing the state of the batch system queues), and report pages (including a quick utility for scooping up archive copies via wget and mangling them so they can be used online) I should of course emphasize that this lot was grown entirely for the requirements here, and I've almost certainly broken things like the grid_tree stuff which we don't use; and I apologise now for any grim code in there! Feel free to mail me any questions (this address will work for a while); I'm likely to be using ganglia at the next place too, so I'll probably be back ;) Cheers, Phil |