workers don't wake up

[nbari]

when launching the workers, they poll the gearmand server for 60 seconds but then they sleep and don't wake up when a job is ready on the server.

If a job is submitted, instead of been processed immediately, the workers take the project after around also 1 min.

Any idea of how to make the workers process all the jobs immediately or to wake them up when a job arrives ?

Jobs I am creating are in background mode

[nbari]

If I am right there could be something wrong with the handling of 'NOOP' the version of gearmand that I am using the lastest version 1.1.7

[klange]

We are aware of this issue and have a possible fix in testing.

[nbari]

Can you share the testing version so that I can also use it and help debugging.?

[klange]

Please see the worker-coaxing branch in this repository, which does not rely on NOOP responses to request workers. In addition to this, we have a batched worker class that is still in development (which I can not make available at this time) that uses a different workflow and should not be susceptible to issues handling NOOPs.

[nbari]

is there a stable working version ? maybe older versions of germand with any python client release ?

[klange]

You may have some luck with an older gearmand, yes.

[nbari]

sorry to bother again with this, but any know gearmand release /version that work fine with the current python client ? if I am right is more an issue with gearmand that with python right ?

[nbari]

I am testing branch worker-coaxing with gearmand 1.1.7 and working very nice, just curious to know how you handle the 'NOOP's ?

[chrisvaughn]

Any update on how testing is going for these changes? I have run into the same problem and will be using the worker-coaxing branch in our dev environment to see how things go.

[klange]

The branch has proved promising, but we would like to take a cleaner and more effective approach as there are still some edge cases that lead to the same end result (stuck workers).

[ninjapenguin]

Just to add to this thread - we are also experiencing problems with this issue - are you able to elaborate on the known edge cases with worker-coaxing? Is the general consensus that it is stable?

[yunjianfei]

There are some bugs with worker-coaxing that work with mutiple gearman servers, noop received ,will get job lock, There only have one lock, but there are many noop for mutiple servers.

So. only one server can received GRAB_JOB_UNIQ . Other NOOP handler can't get lock, and go to sleep.

I have resolved this problem, gather some code from the branches. Add some code to grab job from servers.

My fork is https://github.com/yunjianfei/python-gearman