Maximum timeout for python functions

[andresmgot]

Issue Ref: #365

Summary:

This PR adds a timeout handler for the Python runtime. If a function takes more than 3 seconds on its execution the subprocess get killed. The timeout is configurable using the environment variable FUNC_TIMEOUT.

Also, to avoid duplicated kubeless.py files (from python-2.7 and python-3.4 folders) they are now hard links.

TODOs:

• Ready to review:
• Tests
• Docs (Done in Add maximum timeout for ruby runtime #434)

[murali-reddy]

FUNC_TIMEOUT env not being set in deployment spec. How timeout can be configured?

[andresmgot]

the idea is to execute kubeless function deploy --env FUNC_TIMEOUT=10 .... It is not specified in the dpm spec because the function uses a default value if it is not set.

[murali-reddy]

i see. Was not aware of --env flag. Changes look good to me. Document the behaviour.

[anguslees]

Multiprocessing is a pretty heavyweight way to add a timeout - you're forking a new process and reserialising the request and response objects for every function call.

A. Do we care about performance?
B. Can we just do this with a thread? Or if we have to handle the case where the entire process is deadlocked - perhaps we can just run the watcher in a persistent separate process so we can pass the timeout and not the much larger request/response bodies across processes.

[murali-reddy]

With Python's Global interpretter Lock and bottle, threading is useless anyway. If not for timeout we do need better way for parallel execution.

[andresmgot]

I will try to do it with threads

[murali-reddy]

Perhaps this is out of conext of this bug. I opened #435 for consideration pythons multiprocessing

[andresmgot]

@anguslees I tried with threads but there is not an official way of terminating threads (and seems that is something wrong to do: https://stackoverflow.com/questions/323972/is-there-any-way-to-kill-a-thread-in-python). If we don't terminate the function it will keep using the container resources until it fails.

I tried as well using a Timer but this has the issue that the timeout function is executed in a different thread so the main thread remains executing the function anyway.

I am reverting to the previous approach with processes, it is the only way of properly terminating the process if it takes more than N seconds.