Page MenuHomePhabricator

Make daemons work correctly no matter where they are or how many copies are running
Closed, ResolvedPublic

Description

In the bold new world of high availability, bin/phd start should do the right thing by default no matter where you launch it. In particular:

  • The PullLocal daemon needs to become smart about making sure the repository is supposed to exist on the current host, and we should drop the --no-discovery and repository selection flags.
    • This is probably not difficult, but needs some finesse/communication for existing installs.
    • The bin/repository subcommands can still force action on a repository on the current host (although maybe they should prompt if the repository is not naturally local to the host).
    • This runs into T4292 to some degree.
  • The Trigger daemon is fine to launch everywhere, but its locking may need to be more fine-grained or forgiving. This won't scale up cleanly with the number of instances, but no install should ever be pushing the trigger daemon to its limits.
  • The Taskmaster daemons seems reasonable to boot up as an autoscale group of the default size.
  • The documentation should be updated to reflect this state of the world.

We should also revisit the UI, particularly with an eye toward making it more useful for figuring out which host a daemon is running on (T6768) and improving usability (T7307). There are a handful of cheap adjacent tasks which make sense to pick up here since testing them is cheap (T9385, T7475, probably T6615 now that key changes are cheap).

Event Timeline

eadler added a project: Restricted Project.Apr 9 2016, 6:00 PM

T6806 might be ripe to resolve one way or another, too.

eadler moved this task from Restricted Project Column to Restricted Project Column on the Restricted Project board.Apr 17 2016, 6:25 PM
eadler moved this task from Restricted Project Column to Restricted Project Column on the Restricted Project board.May 13 2016, 9:39 PM
eadler moved this task from Restricted Project Column to Restricted Project Column on the Restricted Project board.Jun 1 2016, 10:45 PM
eadler moved this task from Restricted Project Column to Restricted Project Column on the Restricted Project board.Jul 4 2016, 9:00 PM

I believe this is effectively resolved now -- we're running all the daemons on multiple hosts on secure. There's remaining followup work (T6768, T6615, T10753, T10754) but I think the core is in reasonable shape.