High cpu usage

[paveldvorak5]

Hello, trap --ui use a lot of cpu even if it's doing nothing.
Here's the cause, it's checking something every 50ns but I don't know which resource is not available.
Running with -vvv does not help.

strace log

clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=50000}, NULL) = 0
accept(4, 0x7ffea2aef0a0, [128])        = -1 EAGAIN (Resource temporarily unavailable)
accept(5, 0x7ffea2aef0a0, [128])        = -1 EAGAIN (Resource temporarily unavailable)
pselect6(9, [8], [8], [8], {tv_sec=0, tv_nsec=0}, NULL) = 1 (out [8], left {tv_sec=0, tv_nsec=0})
accept(6, 0x7ffea2aef0a0, [128])        = -1 EAGAIN (Resource temporarily unavailable)
accept(7, 0x7ffea2aef0a0, [128])        = -1 EAGAIN (Resource temporarily unavailable)

[roxblnfk]

Hi.
Looks like just sockets checking.
Is it Linux?

[wesamls]

Same for me, using macOS 12.3.

[paveldvorak5]

Hi. Looks like just sockets checking. Is it Linux?

yes, it it
What socket are we talking about? Maybe it's too often?

[roxblnfk]

What socket are we talking about? Maybe it's too often?

Socket accepts:

trap/src/Socket/Server.php

Line 85 in 17f299e

while (!$this->cancelled and false !== ($socket = \socket_accept($this->socket))) {

And socket select

trap/src/Socket/Client.php

Line 80 in 17f299e

if (\socket_select($read, $write, $except, 0, 0) === false) {

It happens every tick.

I don't experience such issues on Windows, probably because the time for usleep() on Windows cannot be less than 10ms. Linux ticks at a smaller interval (50 microseconds).

trap/src/Application.php

Line 127 in 17f299e

\usleep($sleep);

[paveldvorak5]

Yes, that's it. I "solved" the issue but changing the interval. Can setting a low interval break things?

[roxblnfk]

The higher this value, the slower the content will be delivered to the browser, and it directly affects TTFB.
What value are you using? I think it can be used for socket polling, while delivering content at the minimum interval.

[paveldvorak5]

I changed it to 5000. I use it for local dev so the delay does not matter. Maybe put the value to config?

[roxblnfk]

Could you check 1000 and 500? I think 1ms is comfortable value for developer and CPU :)

[paveldvorak5]

As expected, the usage went up from 1% to 4%.
When changed to 10000, the app seems to work fine, only it takes 30 s to start the port 8000.

[roxblnfk]

I will release a patch with #140. It includes personal throttling for polling each socket. I would appreciate any feedback here about it.

[wesamls]

@roxblnfk Thanks for the update.
After installing the latest version, the cpu usage still high but lower than last time (was ~ 23, now ~ 18).
Not sure how to change the "wait period"? any command parameter to pass or config file?

[roxblnfk]

@wesamls see #141

It would be great if you could find the optimal values for yourself and share them here.

[wesamls]

Thanks for adding the environment variables.
I've used TRAP_MAIN_LOOP_INTERVAL=1000 and cpu went down to ~ 4.

[paveldvorak5]

Can you check the optimal values?
I tried on local machine TRAP_TCP_POLLING_INTERVAL=10000 TRAP_MAIN_LOOP_INTERVAL=10000 vendor/bin/trap --ui to get to 0.5% CPU and it still works fine. So I think the default values are too high.