Skip to content

Latest commit

 

History

History
55 lines (35 loc) · 2.95 KB

README.md

File metadata and controls

55 lines (35 loc) · 2.95 KB

Heroku H12 Monitor

This daemon is designed to watch for "stuck" dynos reporting constant H12s (i.e. HTTP 503's) and selectively restart them.

Why

Occasionally, a dyno can hang and get into a state where it's no longer responding to requests. Due to Heroku's routing mechanics, that stuck dyno will still receive requests from the router, which will then continue to report H12s.

A dyno can get into this state for a number of reasons, though getting to the root of problem is the only way to a lasting solution.

How it works

When run, it will connect to the Heroku API and open a streaming HTTP connection to the Heroku logplex, monitoring router events. Each log message is parsed and the code is looked at to determine if it's an H12. The alogorithm it follows is:

  • If a dyno reports an H12, increment a counter for that dyno
    • If the dyno exceeds 3 consecutive H12s, issue a restart.
  • Whenever a dyno successfully processes a request, reset the counter.

Thus, a dyno must report three consecutive H12s in order to be considered "stuck". The script will then issue a restart via the Heroku API.

How to use it

  1. Log into Heroku and get your API key. It's on your My Account page.
  2. Edit h12-monitor.rb and add your API key to line 14. Set your heroku app name on line 15.
  3. Run bundle install.
  4. Start the app by running ruby h12-monitor.rb run

Example output

When running it in the foreground (i.e. with the run option), you'll see output similar to the following:

$ ruby h12-monitor.rb run
h12-monitor.rb: process with pid 40464 started.
I, [2013-02-18T14:38:46.540868 #40464]  INFO -- : Connecting to Heroku logplex for myapp-production.
I, [2013-02-18T14:52:07.955868 #40464]  INFO -- : myapp-production: monitored 50 lines, 18 dynos reporting
I, [2013-02-18T15:06:13.414261 #40464]  INFO -- : myapp-production: monitored 100 lines, 20 dynos reporting
W, [2013-02-18T15:06:14.038075 #40464]  WARN -- : web.12 reports H12 (#1)
W, [2013-02-18T15:06:14.038225 #40464]  WARN -- : web.12 reports H12 (#2)
W, [2013-02-18T15:06:14.038402 #40464]  WARN -- : web.12 reports H12 (#3)
W, [2013-02-18T15:06:14.038443 #40464]  WARN -- : restarting dyno web.12
I, [2013-02-18T15:06:18.677463 #40464]  INFO -- : myapp-production: monitored 150 lines, 20 dynos reporting

The monitor will print a status line every 50 log lines it processes. If it detects an H12 on a dyno, it will print a WARN message. After three consecutive H12s for the same dyno, it will restart it.

This isn't a solution...

This monitor is designed as a stop-gap until you can debug and solve the underlying problem of why your dyno's get stuck. However, it can buy you some time while you debug.

TODO

  1. Could use some more coverage, mostly around the network connection stuff with the log streamer.

License

h12-monitor is Copyright © 2013 Ryan Twomey. It is free software, and may be redistributed under the terms specified in the MIT-LICENSE file.