Page MenuHomePhabricator

Show a queue utilization statistic in the Daemon console
ClosedPublic

Authored by epriestley on Dec 9 2013, 4:46 AM.
Tags
None
Referenced Files
F14062999: D7747.diff
Mon, Nov 18, 3:55 PM
F14036585: D7747.diff
Sun, Nov 10, 11:00 AM
F14025819: D7747.id17513.diff
Thu, Nov 7, 8:42 PM
F14025545: D7747.id17520.diff
Thu, Nov 7, 6:47 PM
F14025452: D7747.id17513.diff
Thu, Nov 7, 6:04 PM
F14025442: D7747.diff
Thu, Nov 7, 5:55 PM
F14025440: D7747.id.diff
Thu, Nov 7, 5:53 PM
F14025360: D7747.diff
Thu, Nov 7, 5:03 PM

Details

Summary

This came up recently in a discussion with @lifeihuang, and then tangentally with @hach-que. Make it easier for users to get a sense of whether they might need to add more daemons. Although we've improved the transparency of daemons, it's not easy for non-experts to determine at a glance how close to overflowing the queue is.

This number is approximate, but should be good enough for determining if your queue is more like 25% or 95% full.

If this goes over, say, 80%, it's probably a good idea to think about adding a couple of daemons. If it's under that, you should generally be fine.

Test Plan

{F88331}

Diff Detail

Lint
Lint Skipped
Unit
Tests Skipped

Event Timeline

src/applications/passphrase/view/PassphraseCredentialControl.php
5

Unrelated bugfix for rendering this control with no options, as we do when creating a new file:/// repository. I hit this while testing the change above.

Just curious - do you envision this working in the Phacility case?

(One giant queue for all the installs, or per-install queue? Administration of those by owners of a given install, or by us, or by automagicalness? I'd assume a per-install queue with some automagical administration going down.)

The dumb/launch plan is to run a group of daemons per install -- basically execute phd stop and then phd start for each install every time we push. These commands would be distributed across some homogenous daemon tier, and we can get smarter about load balancing installs across machines based on data like this. I think that will get us pretty far, and then maybe we write a meta-daemon, since presumably the big opportunity for optimization will be sleeping more often on dead installs. This is fuzzier than the other scale plans, but we have a lot more headroom since it's basically OK for the daemons to be down for a few minutes on a push.

i just found that image funny and wanted to say hi. it has been a ridiculous morning here.

sorry about your pipes

At least it was outside?

Aaaand we're at 2% utilization.

{F88569, size=full}