Garbage collection tuning in HA

[rlippmann]

Describe your core improvement

It would probably help performance to increase/manually manage python garbage collection in Home Assistant.

The ways I've been thinking of improving this are:

1. bumping the default. I've seen 50k object threshold tossed around as a more realistic number for larger projects
2. temporary disabling the gc during startup
3. since HA uses its own event loop, manually perform gc during idle periods

Current limitations

Currently, HA appears to just use the default of 700 objects for gc threshold.

Technical benefits

Better performance, less "stop the world" issues when trying to perform actions.

Additional context

The first method of just bumping the threshold is very easy.

I'm not sure the second (pausing gc during startup) may or may not be beneficial. Would need to do some benchmarking. Just bumping the threshold might be enough.

The third method might be difficult or perhaps impossible to implement, but could potentially yield less overall memory usage as well as a more responsive system.

[rlippmann]

@bdraco Any thoughts on this, since you seem to be the HA performance guru?

My one concern would be for devices with lower RAM like Raspberry Pi that are running HA. Maybe conditioning using a larger threshold based upon what psutil returns and an "average" HA object size?

[mib1185]

Please don't tag people out of the blue if they were not involved in the current discussion to demand attention! This is intrusive and should be avoided!
One will look into this request if they have time and are familiar with topic.

[rlippmann]

Please don't tag people out of the blue if they were not involved in the current discussion to demand attention! This is intrusive and should be avoided! One will look into this request if they have time and are familiar with topic.

Sorry :(

[jvmahon]

Python 3.14 (currently in Release Candidate stage) will add incremental garbage collection which may address any concerns around this issue (assuming HA moves to 3.14), so it seems the best approach for now is wait-and-see rather than try to manually implement. https://docs.python.org/3.14/whatsnew/3.14.html

[rlippmann]

It will only incrementally gc the oldest generation, though. python/cpython#116206