Performance portal/planning

From mediawiki.org

proposed at possible projects.

software[edit]

  • new tool or just reconfigure gdash with a bunch more dashboards? can gdash currently do anything besides graphite?

potential data sources[edit]

  • icinga (aka nagios)
  • ganglia
  • graphite/diamond/statsd
  • logstash (stats for searches in a way that it could be public. or maybe also included in a potential private perf dash). e.g. top X mediawiki exceptions by file/linenumber.
  • tendril/dbtree
  • watchmouse
  • smokeping?
  • torrus
  • analytics output? e.g. signpost traffic report type stuff. or anything from x-analytics header, eventlogging, ttfb/total time per req according to varnish/protoproxy. Extension:NavigationTiming data
  • ishmael maybe. should there also be a private perf dash?

interface[edit]

  • overview(s) and then also dash per service? (e.g. swift, redis, hhvm, php5, apache, apc, )
  • need an interface where people can test out potential new dashboards before they are deployed for all dash users and shares syntax with the on disk conf (or at least generates syntax for checkin to git)
    • user's custom creations should be permalinkable so that they can be bookmarked or shared
  • ????

configuration[edit]

  • should be relatively language agnostic
    • avoid embedded scripting in the middle of config
    • we should potentially (depending on the new tool) be able to keep using the same config even if we switch tools
  • simple flat files? maybe yaml. maybe compile at deploy-time with jinja if needed
  • compare to gdash format (which I haven't looked at in a while)

See also[edit]