fix(cli): shut down timers at the end of execution

[flrgh]

Summary

With timer-ng, the CLI has become slow to terminate as nginx attempts to wait for all pending timers to stop before exiting:

$ time kong roar >/dev/null

real    0m1.033s
user    0m0.024s
sys     0m0.016s

This commit explicitly shuts down the timer-ng instance so that the CLI can quickly exit:

$ time kong roar >/dev/null

real    0m0.132s
user    0m0.027s
sys     0m0.013s

Full changelog

• explicitly stop timer-ng instance on completion of CLI commands

[bungle]

@ADD-SP can you confirm this is alright?

[Tieske]

wondering; shouldn't this be fixed in timer-ng instead?

[flrgh]

wondering; shouldn't this be fixed in timer-ng instead?

I think timer-ng is mostly behaving as expected here. I believe the 1s shutdown delay observed currently is due to a semaphore:wait(1) call somewhere (as controlled by this constant). Delaying worker shutdown by as much as 1s seems totally fine in the regular nginx context--it's just that the CLI is a whole different beast.

[Tieske]

wondering; shouldn't this be fixed in timer-ng instead?

I think timer-ng is mostly behaving as expected here. I believe the 1s shutdown delay observed currently is due to a semaphore:wait(1) call somewhere (as controlled by this constant). Delaying worker shutdown by as much as 1s seems totally fine in the regular nginx context--it's just that the CLI is a whole different beast.

this is an old problem caused by a lack of IO in the nginx control loop, workarounds are available. See openresty/lua-resty-core#337 That said; from that issue I see that I already pointed that out before...