The timeout duration for starting/stopping a switch may be a little bit short.

[tang-mouren]

I've encountered a problem when creating/stopping a topology containing 144 hosts and 13 switches with p4utils. There is a high probability that the program will raise a ChildProcessError which saying "P4 swich xx did not start/stop correctly.". I digged into the source code and found that the error was from the if sentence: "if not wait_condition(self.switch_status, True, timeout=SWITCH_START_TIMEOUT".
To my understanding, the time limit for a child process to create or stop a P4 switch may not be long enough when creating a larger topology, so I changed the global variables "SWITCH_START_TIMEOUT" and "SWITCH_STOP_TIMEOUT" from 10 to 60. This solution worked and the problem didn't occured again after I changed the code.
I'm not certained whether my understanding is correct, so I make this pull request. I will appreciate it if you can give me a reply, and I hope this adjustment to the code can help others meeting the same problem

[tang-mouren]

In my commit, I made a compromise, so the variables value are chagned into 30.

[edgar-costa]

HI @tang-mouren,

Thanks for looking into this. I never had this problem, even with topologies of that size (probably with fewer hosts). I guess that is the problem, although given everything is started sequentially, I am surprised that 10 seconds are not enough.

The only problem I see with raising the timeout is that when the problem comes from something else (i.e., a bug in the switch code), the waiting times might be too high. Was that the reason why you went from 60 to 30 in the pull request?

[tang-mouren]

Emm, later I changed the value from 60 to 30 and the problem didn't occur. I haven't done enough experiments to test the problem to get the most suitable time so I choose the value 30, in case that a bug might cause the infiniteloop letting developers waste more time waiting for the timeout.
Up till now, in my opinion, we can give the variable a larger size. And in the future, we can add more information in the error report, telling developers that changing the value of the time limit may address the problem

[tang-mouren]

I found that the program may stick at a moment when allocating more interfaces to a switch, and this may cause some time. The timeout event happens randomly. It doesn't always occur when I run the program.

[tang-mouren]

So, I suggest the value of the time limit can be changed larger to avoid this problem to a certain degree.