General slowness after a while, how to debug?

[Madd0g]

I noticed that after a few hours/days, HS becomes sluggish until I reload the config, then it's fast again.

I'm doing tons of things, using many parts of the api with a lot of custom code. HS is my command center.

It's running a low-traffic internal webserver, IPC, running processes every few secs, drawing on the screen, running applescript, using axuielement every few secs, updating menus every few secs, watching battery/power, parsing json/xml, lots of timer, hotkey, window, app-related handling.

I'm wondering if I can somehow pinpoint the areas of my code that are causing the sluggishness? Should I convert hs.drawing to hs.canvas if I'm putting things on the screen 10-15 times a minute? Any cleanup related advice for variables, timers, processes, other things? I'm pretty much a beginner in lua.

Thanks

[cmsj]

is there a significant difference in RAM usage between the sluggish state and the reloaded state?

[Madd0g]

I see the Hammerspoon process gaining about 100MB a day. No sluggishness so far.

but as I was watching the process I noticed other things:

1. I'm calling hammerspoon from Karabiner-Elements with the /usr/local/bin/hs -c "<LUA CODE>" command line, but I see dozens of hanging hs processes in activity monitor. Some don't seem to close after the code they run? Can I somehow force them to quit? (I just started doing this since it's much much faster than doing the same with applescript)
2. Same thing with processes I start from HS, I see tons of them even though the code is supposed to kill the ones that don't quit on their own:

local mod = {}
local timer = nil

function mod.start()
    timer = hs.timer.doEvery(8, function ()
        -- I copied this termination example from somewhere
        if mod._task and mod._task:isRunning() then
            print("+++ powermetrics task taking too long; killing it!")
            mod._task:terminate()
        end
        local cmd = "/usr/bin/sudo"
        mod._task = hs.task.new(cmd, function(ret, out, err)
            if ret == 0 then
                -- do things
            else
                print(err)
            end
        end, {
            "-A", "/usr/bin/powermetrics", "-n", "1", "-i", "1", "-f", "plist"
        }):start()
    end)

end

function mod.stop()
    timer.stop(timer)
    timer = nil
end

return mod

Can you see anything wrong in the above? I do see the message about killing the process in the console once in a while.

[Madd0g]

I spoke too soon, it's around 700MB (starts around 170) now and animations are getting slower and slower... also got this in the console (after calling HS from Karabiner):

ERROR:   LuaSkin: hs.ipc:callback - error during callback for Hammerspoon: ...rspoon.app/Contents/Resources/extensions/hs/ipc/init.lua:416: ipc port is no longer valid
stack traceback:
    [C]: in method 'sendMessage'
    ...rspoon.app/Contents/Resources/extensions/hs/ipc/init.lua:416: in function 'hs.ipc.internal.__defaultHandler'

Also, I see in activity monitor that the hammerspoon process has a big number of ports, 4th on the list of processes sorted by ports, don't know what that means.

after reload I see 1.7k but just before I had to reload it was showing 8k.

[luckman212]

@Madd0g that code sample above... what exactly are you hoping it will do? One thing I noticed there is hs.task.new() calls sudo with the -A parameter (askpass mode). If your environment doesn't have the SUDO_ASKPASS variable set to a command that returns your root password, then this will fail.

[Madd0g]

@luckman212 - it does work, I put the powermetrics command in the sudoers file based on advice in this thread and it doesn't ask for a password.

It's really strange though, the higher the CPU usage, the more this process gets stuck and never exits. I wonder if _task:terminate() doesn't work because of the sudo trick

[luckman212]

Ok, well the -A flag is for askpass so I would take that out and retest. I don't think it's needed when adding exceptions in sudoers.

[Madd0g]

I retested without the -A flag, no difference.

But now I know exactly how to repro the powermetrics hanging process - I run one of my project builds that creates very high CPU while building for about 40 seconds. During that time the timer runs 3-4 times but leaves behind 1-2 processes.

I tried running the same command from the terminal under the same high CPU conditions and I don't see it get stuck, it always quits.

[luckman212]

I'm still not sure exactly what this Spoon is supposed to do, but:

• if you start a process with sudo, you need to be root to kill it, so I would expect hs.task:terminate() to fail there

• it looks like powermetrics has its own interval timer, and is designed to run continuous sampling. The -i 1 args above would have this happen every 1s, and then the hs.timer piling another one on top of that every 8s. Seems like something's not right with that logic but again I am not sure exactly what the desired output is here.

[Madd0g]

yeah powermetrics does its polling on interval (5s) by default and never stops, but if you pass -n 1 -i 1 it samples for just 1ms and quits.

I would actually prefer to only launch it once, but the API docs say hs.task is not meant for tasks that keep streaming data.

I am not sure exactly what the desired output is here

I'm reading some values from the output, fan speed and process list.

[Madd0g]

yeah, that's a good point about mod._task:terminate(), I wasn't even thinking that in order to send a SIGTERM to the process you need any kind of additional auth.

[gineer01]

I used to have a similar slowness issue (Hammerspoon could take up to multiple GB of memory) and managed to track it down to #1980 . You might want to check if your config or any Spoon you use have Applescript. I have to ban hs.osascript and any module/spoon using it with this code in my config.

hs.osascript = "DO NOT USE. MEMORY LEAK."
hs.itunes = "DO NOT USE. MEMORY LEAK." -- it uses osascript

[Madd0g]

@gineer01 - that's interesting, I recently switched the most frequent thing I was doing with applescript to use axuielement. so basically I reduced my applescript usage by at least 90%, but the slowness got more frequent.

the only applescript that is left are things that I trigger maybe 2-3 times a day.

[Madd0g]

so I've been checking activity monitor every time before restarting HS. memory doesn't not seem like a big indicator, but again the "Ports" number is always high when I check (about 4x than the number at startup).

from this page it says:

These are Not network ports.

macOS is built on top of the Carnage Mellon MACH kernel, and processes talk to the MACH micro kernel and between each other using 'ports'.

I recently started using IPC and the hs command line a lot (I trigger hammerspoon with it from Karabiner-Elemenents), could that be related? it opening some connection but never closing it?

EDIT: another thing about my usage of the hs CLI tool - after I noticed the hanging processes, I changed my code like this and that eliminated the hanging hs process.

- /usr/local/bin/hs -c \"%s\"
+ /usr/local/bin/hs -c \"return %s\"

just adding that return seemed to allow the hs processes to close, I'm not sure where I got this solution from, but it solved that problem.