Joseph Schorr
52fa9aad5b
Fix etcd watching
...
Etcd can miss events on watches if they are occurring fast enough, so if we can get an exception indicating that we've missed an index, we reset the state of our local tracking structures by re-reading the *full* list and starting a new watch at HEAD
2015-06-25 21:22:39 -04:00
Joseph Schorr
b4c39e8ec0
Fix ephemeral build manager to ask for watches in index order with no gaps
2015-06-23 17:11:46 -04:00
Joseph Schorr
c2dc1c9b75
Handle case where etcd key is already removed on job complete
2015-06-17 15:02:58 -04:00
Jake Moshenko
c435f5c127
Add a comment about why we are taking a lock when terminating a builder machine.
2015-06-10 16:19:51 -04:00
Jake Moshenko
f767fc4d03
Track whether builders ever came online in etcd. Mark builds which never successfully heartbeated as incomplete.
2015-06-10 16:19:51 -04:00
Jake Moshenko
884fedd229
Improve the log messages in the buildman.
2015-06-10 16:19:51 -04:00
Jake Moshenko
d31e25d5cd
Allow the individual build manager types to specify how long the queue should wait before retring a job that fails to schedule.
2015-06-10 16:19:50 -04:00
Joseph Schorr
f82831bff6
Log the etcd exception so we can debug this issue
2015-06-09 20:33:55 -04:00
Jake Moshenko
0349f3f1a3
Handle the case where YAML config returns a list not a tuple.
2015-03-26 14:53:56 -04:00
Jimmy Zelinskie
8589871f43
buildman: rm unused imports
2015-03-09 13:04:16 -04:00
Jake Moshenko
5c68e52fce
Really really fix the exception handling.
2015-02-27 17:33:46 -05:00
Jake Moshenko
cf5bc6f0be
Properly catch multiple exceptions.
2015-02-27 17:32:10 -05:00
Jake Moshenko
857c3e2959
Start catching etcd key errors as well.
2015-02-27 17:10:15 -05:00
Joseph Schorr
4551b3a957
Remove the boto timeout set (doesn't work anyway) and add some better logging to the scheduler
2015-02-25 16:00:14 -05:00
Joseph Schorr
5dd78f76c7
Add additional logging, timeouts, and exception checks
2015-02-25 15:15:22 -05:00
Joseph Schorr
afe7e14254
Add better exception handling and logging to the ephemeral build manager
2015-02-25 12:09:14 -05:00
Joseph Schorr
524705b88c
Get dashboard working and upgrade bootstrap. Note: the bootstrap fixes will be coming in the followup CL
2015-02-17 19:15:54 -05:00
Jake Moshenko
a4b0c8698d
Allow the key prefixes in etcd to be configurable.
2015-02-02 12:00:19 -05:00
Jake Moshenko
ef0806bd9d
Make the logs for the build manager more bearable.
2015-01-26 15:27:39 -05:00
Jake Moshenko
86852da4ba
Catch exceptions when ELB times out a connection to etcd.
2015-01-23 11:29:38 -05:00
Jake Moshenko
265aeabf60
We need to tell the etcd client which protocol to use.
2015-01-22 16:59:04 -05:00
Jake Moshenko
f2471a86f6
Fix the python requirements. Add the ability to map in etcd client certs and ca.
2015-01-22 10:53:23 -05:00
Jake Moshenko
fc757fecad
Tag the EC2 instances with the build uuid.
2015-01-05 15:35:14 -05:00
Jake Moshenko
8037962716
Change the severity of a log message which is actually expected in the happy case.
2015-01-05 14:44:54 -05:00
Jake Moshenko
f58b09a064
Remove the loop argument from the call to build_component_ready.
2015-01-05 13:08:25 -05:00
Jake Moshenko
320ae63ccd
Handle the case where there are no realms registered.
2015-01-05 12:23:54 -05:00
Jake Moshenko
b33ee1a474
Register existing builders to watch their expirations.
2015-01-05 11:21:36 -05:00
Jake Moshenko
a9839021af
When the etcd key tracking realms is first created the action is create, not set.
2014-12-31 11:46:02 -05:00
Jake Moshenko
cc70225043
Generalize the ephemeral build managers so that any manager may manage a builder spawned by any other manager.
2014-12-31 11:33:56 -05:00
Jake Moshenko
3ce64b4a7f
We must yield from stop_builder.
2014-12-23 16:12:10 -05:00
Jake Moshenko
2ed9b3d243
Disable the etcd timeout on watch calls to prevent them from disconnecting the client.
2014-12-23 14:54:34 -05:00
Jake Moshenko
4e22e22ba1
We have to serialize our build data before sending it to etc.
2014-12-23 14:09:04 -05:00
Jake Moshenko
709e571b78
Handle read timeouts from etcd when watching a key.
2014-12-23 12:13:49 -05:00
Jake Moshenko
055a6b0c37
Add a total maximum time that a machine is allowed to stick around before we terminate it more forcefully.
2014-12-23 11:18:10 -05:00
Jake Moshenko
34bf92673b
Add support for adjusting etcd ttl on job_heartbeat. Switch the heartbeat method to a coroutine.
2014-12-22 17:24:44 -05:00
Jake Moshenko
2b6c2a2a50
Improve tests for the ephemeral build manager.
2014-12-22 16:22:07 -05:00
Jake Moshenko
12ee8e0fc0
Switch a few of the buildman methods to coroutines in order to support network calls in methods. Add a test for the ephemeral build manager.
2014-12-22 12:14:16 -05:00
Jake Moshenko
2d7e844753
First implementation of ephemeral build lifecycle manager.
2014-12-16 13:41:30 -05:00
