Tiedoksi, että 2025.9.2 päivityksessä on bugi joka aiheuttaa korkean CPU käytön.
Totesin itsekin saman ja downgradasin.
Suosittelen odottamaan 2025.9.3 versiota.
The problem After upgrading to 2025.9.2 i get high load on the homeassistant core container. so i downgraded back to 2025.9.1. This is just to see if others expierencing the same issue. What versio...
github.com
"I've identified the root cause of the high CPU usage in 2025.9.2. This is caused by a kernel bug that was exposed by a new capability check we added in this version.
What Changed in 2025.9.2
We added a permission probe to check if Home Assistant has the necessary capabilities (NET_ADMIN/NET_RAW) for Bluetooth management operations. This was intended as a low-risk improvement to provide better fallback behavior when permissions are missing.
The Hidden Kernel Bug
On many systems, the Bluetooth management socket has a kernel bug where sendto() returns 0 instead of the number of bytes sent, even though the data is successfully transmitted. When asyncio sees this 0 return value, it interprets it as "nothing was sent, retry later" and keeps the data in its buffer, attempting to send it repeatedly. This creates an infinite loop consuming 100% CPU.
The irony is that the capability check actually succeeds - the kernel delivers the packet and returns a valid response. But asyncio doesn't know this due to the incorrect return value.
Why This Affects So Many Systems
This kernel bug appears across various platforms and virtualization environments, which explains the widespread impact. The bug was always present but remained hidden until our new capability check triggered it.
The Fix
We've implemented a workaround that:
Bypasses asyncio's transport layer for Bluetooth management sockets
Writes directly to the socket
Treats a return value of 0 as success (which it actually is for these sockets)
This prevents the infinite retry loop while maintaining full functionality.
Why Downgrading Works
Version 2025.9.1 doesn't include the capability check, so it never triggers the kernel bug. That's why downgrading resolves the issue.
A fix is being prepared and will be included in the next release. Thank you all for your patience and detailed reports - they were invaluable in tracking down this tricky issue."
