Skip to content
Erik Mackdanz edited this page Oct 3, 2013 · 11 revisions

New Aspects

  • flapping (open bounty for solid testing method)
  • connections
    • This is a specialization of what Flip calls "discoverhood": ironcuke should test when a component announces that it discovers something. Beyond testing TCP connections, I'm not sure how this can be testable.
  • users
  • mounts - a system with a failed mount can run healthy for quite a while, until the local disk runs out of space. Cleanup is no fun when this happens, but the condition is detectable.
  • ports
    • response times (not sure how to test this in a generic way, see proofs)
    • non-declared ports aren't listening. This would require some meta-capability, above the component level
    • test access from external machine. This tests security groups, iptables, etc. Would require some external reflector service, and probably rolls in with "confirm ports not listening" item
  • logs
    • Get some data
    • Don't get too much data
  • http access. This is probably the :dashboard mentioned in the silverware README
  • :proof, a line of script that a component can announce with an additional component-specific test (e.g. :proof => { :hdfs_writable => "hadoop fs put /tmp/foo 'bar'" })

Missing SRP Components

  • chef-server (low pri, since hosted chef is the common case)
  • command-scheduler
  • backup-center

Codebase changes

  • add back in a little bit of gorillib to the no_gorillib branch for validating announcements and consistency with ironfan

Always nice

  • make all announcements and discoveries consistent and sensible
  • Redo README
Clone this wiki locally