nanny
nanny copied to clipboard
uber-archive
Cluster management for Node processes
Worker state change events need to be hooked up before the process is started. This is achieved by initializing the process on next tick. However, this breaks the assumption that...
Nanny seems to have an issue dealing with multiple concurrent connections. Incoming requests to my service end up timing out. Will need to be investigated further.
Calling worker.start(cb) in the "running" state does not appear to issue the callback, &c.
After which a worker that has been placed in debug mode would stop with intent to restart. Alternately, we just monitor debug session duration and set up alerts for long...
The health report loop should not prevent a worker from stopping gracefully. Perhaps the health report system should be tied to whether the worker has any active servers.
@kriskowal noted that he had a set of child processes floating around without a master process. There is a possibility that a child process might get orphaned and float around...
e.g., In standby state, for the stop command, call callback on next tick. https://github.com/uber/nanny/commit/5f44addd7f2a82cd5509d0dd09ef516a5b6fa878#commitcomment-8209862
At time of writing, I’m killing and restarting. In the future, we may want to automate a perf capture.