Renewing address of remote actor after process restart

[celluloid]

Renewing address of remote actor after process restart

[jasonayre]

Although the "infinite loop" problem exists, its not always infinite, as its conditional on whether process 1 comes back up, if it does (with same address and port it was originally attempting to connect to), the request that process 2 was attempting to make, will succeed.

Also to expound on the registry problem, it isnt that the registry isn't being updated, its that node2 is trying to call node1 after it dies, before the registry gets updated (and it thinks that the remote node is still there). If you run your examples again with 10 second intervals between, it should work (throw an error that remote node doesent exist).

However, even if "live update via zk watchers" did update the registry in near real time, there could still be issues when client makes a request nearly same time the requested node goes down. The underlying issue isn't the registry being updated, its the lack of failure on the request side. I.e. its just sitting for a potentially infinite amount of time waiting for node1 to come back up to execute the request.

It seems to me that the there needs to be a failure timeout/retry mechanism and the client node needs to fail after x amount of attempts. I'll open a separate issue with proposal.

[HoneyryderChuck]

@jasonayre +1 for the timeout feature. I've (dumb-assedly!) tried to use ruby timeout on top of it:

timeout(5) { cell.do_something_that_takes_time }

only to see it break everything.

I had noticed the pattern of bringing the process back to life with the same address, and we have implemented this for now until we get a better solution (the sysadmins don't like this port attribution, though).

[niamster]

I believe it's addressed in pull request #88

[niamster]

Please reopen if you still have this issue in master.