Memory use from home assistant keeps growing

Configuration
1

i never really looked at it before, untill i got a problem recently that my beebox ran out of memory.
now i keep a closer look at it and i notice that HA’s memory use isnt constant, but constantly growing.

after restart it takes up around 600 mb but after a while it takes up 1,2 Gb of memory.
i already took out 6 from my 10 cams, because i suspected the problem there (got some errors)
but that doesnt solve the problem.

i got these components active:

api
auth
binary_sensor
binary_sensor.mysensors
binary_sensor.workday
calendar
calendar.google
camera
camera.foscam
device_tracker
emulated_hue
frontend
google
group
http
input_boolean
input_number
input_select
input_text
light
light.limitlessled
logger
lovelace
mqtt
mysensors
notify
notify.pushetta
onboarding
sensor
sensor.fastdotcom
sensor.fritzbox_callmonitor
sensor.fritzbox_netmonitor
sensor.mysensors
sensor.season
sensor.systemmonitor
sensor.template
sensor.time_date
sensor.yr
sun
switch
switch.broadlink
switch.command_line
switch.mysensors
system_log
tts
websocket_api
zone

so nothing special.
running HA 0.82.1

anyone else noticed anything like it? or has a clue what can cause this?

i am repairing the problem for now by going from 4 to 16 Gb on my beebox. still i like a better solution obviously.

2

oof, memory leak. Those are hard to find w/o mem profilers. I’ll watch my ram usage and get back to you.

EDIT: I have to actually start tracking it. Apparently I forgot to add it to recorder.

1 Like
3

thx, i hope you find something because at this moment i dont have a clue where and how to start looking.

4

Well, maybe after I collect some data, I can tell you what I have implemented and you can narrow the list down. It could also be coming from outside HA.

EDIT: for what it’s worth, my memory hasn’t moved. When I get home I’ll have more info.

5

from outside HA?
that can only be 1 of the components.
unless changing the entities inside from AD would grow HA?

but thanks for looking into it.

6

You might want to look into this

It may point to where the leak is coming from. But like any memory profiler, it’s hard to use because these problems are hard to find.

7

You can probably eliminate these components as I have them and my memory use is constant;

auth
binary_sensor
binary_sensor.workday
camera
device_tracker
frontend
group
http
input_boolean
input_number
input_select
input_text
light
logger
lovelace
mqtt
8

i got all those components running for months, years even.
never noticed any trouble (but didnt look)

i havent updated for a while, didnt add any components recently.

whats your memory use?

9

That’s what I’m saying. Take those ones off your list and see what’s left as I use them and my memory use is constant at 50% (mariadb uses most of this).

1 Like
10

50% doesnt say me much. :wink:
how much MB for HA only?

11

i think i found out the problem, and it troubles me more then before.

because i still experienced problems after i installed 16 Gb in my beebox, i did start debugging on my own way yesterday.

taking out component by component from the config and with all AD disabled trying to find out where the trouble was.
i saw errors about connecting problems, but didnt understand them and google didnt help either.

untill i had deleted every component and the problem was still there.
so i updated HA, but that didnt help either.

then a brainwave hit me, i didnt close down my dashboards.
after i did close all of them the error was gone.
in the meantime i found out that the error appeared over 600 times a minute!!
my memory trouble was gone at that point also, so the heap of errors were filling up the memory.

it was a connection error from the cameras on my dashboard that resulted in this.
and thats what troubles me. such a small thing causing to break down everything isnt a good thing.

allthough i am glad that problem is gone, i still got trouble.
i have had this over and over in the past, and its there again.

every few seconds a warning that a component takes more time then needed.
something is slowing things down inside HA.
off course not any error that gives a clue

ill probably will start a second HA and let that run stuff thats important only so i can start debugging this.

off course the error that is an aiohttp issue for over 1,5 year on github still appears once in a while.
i have given up hope that someone looks at that. (also because they wanted to close it several times)

@tom_l and @petro thanks for at least trying to help.
to bad that my trouble always seem to be from a level, where noone has a clue.

1 Like
12

Have same problems on raspberry pi. Many errors in logs (broadlink, yeelight and others), a HA always on at my tab. Every day after update raspberry pi need to reboot.

13

errors or warnings?
to get those problems on a RPI is way more expected, then gettin it on a powerfull device.
but i found some issues about it on github. off course its around there for months and no solution.

and my guess is that there will never come a solution, because noone looks at it.
some like the other error i got, that issue is already 1,5 year old.

and my gut tells me that it all boils down to aiohttp trouble.

14

the problem was never really gone. only it was less visible because i did split things up, and because it sometimes took hours, sometimes days before the problem was there.

after months of searching, splitting HA up into different instances, etc. i found the problem.

1 camera has a bad network connection.
as soon as i took that 1 out of the configuration the problem was gone.

it still troubles me that 1 device with a bad connection can hang up HA completely.
without any errors or visible warnings (except that a connection is lost sometimes)
debugging something like that is almost impossible when you got hundreds of devices like i do.

2 Likes