Remove containers grace period

[Vlatombe]

Nothing interesting should be running in agents containers when it is requested to be terminated, so don't wait these 30 seconds to kill remaining processes inside it.

Should speed up a lot tests hopefully.

Subsumes #689 and #690, please review this commit.

[kerogers-cloudbees]

There's a test failure in ContainerExecDecoratorTest.testCommandExecution() Is that existing or was that test relying on the grace period to execute?

[Vlatombe]

It's also on master. I never hit it when running tests locally but it seems to happen systematically when running on kind 0.7.0 (#685), not sure why.

[Vlatombe]

Saves about 14 minutes (40 minutes total vs. 54 min)

[Vlatombe]

Test failure already seen in #553

[Vlatombe]

Looks like buffer from other threads is printed instead.

[4] pid is 20
[0] pid is 21
Executing command: "sh" "-c" "cd /tmp; echo [1] pid is \$\$ > test; cat /tmp/test" 
exit
Executing command: "sh" "-c" "cd /tmp; echo [2] pid is \$\$ > test; cat /tmp/test" 
exit
Executing command: "sh" "-c" "cd /tmp; echo [3] pid is \$\$ > test; cat /tmp/test" 
exit
Executing command: "sh" "-c" "cd /tmp; echo [9] pid is \$\$ > test; cat /tmp/test" 
exit
[7] pid is 51
[5] pid is 53
[5] pid is 53
[5] pid is 53
Thread 0 finished
Thread 4 finished
[2] pid is 81
Thread 8 finished
Thread 6 finished
Thread 5 finished
Thread 2 finished
Thread 7 finished
[9] pid is 82
Thread 3 finished
[9] pid is 82

See duplicated [5] and [9]

[Vlatombe]

@jglick looks like some synchronization issue when running in cpu constrainted environment.

[Vlatombe]

@jglick Don't review this right now, cleaning up the patch first.

[jglick]

(trying to ignore changes from #690)

[jglick]

Is this really safe?

[Vlatombe]

I was wondering as well in case some process really need to do cleanup. Probably safer to set it for all tests.

[jglick]

safer to set it for all tests

Do you mean set it only for tests?

[Vlatombe]

Yes this is what I meant

[jglick]

Fine here in tests.