FFMPEG crashing with no config changes. 0.17.0 working for weeks, then broken. #22519
-
|
I have been running the 0.17.0 frigate for over two weeks. This morning the cameras stopped working and the CPU is spiking with the following errors in the logs for all the cameras. They are all wired and H264. It's exactly the same for all of them. Using the I am running all of them using go2rtc and RSTP, then using the restream input for the detect (substream) and record (main stream). Also, the RSTP streams work in the web, but none of the recording or detecting is working because of the libva or ffmpeg or some other issue. Again, nothing has changed, and I can't find any other issue that has the same errors like this across all cameras suddenly. Thoughts? Anything to try would be helpful! |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
The errors indicate Intel QSV hardware acceleration is failing to initialize the MFX video decoder ( Immediate WorkaroundSwitch from cameras:
backyard_cam:
ffmpeg:
hwaccel_args: preset-vaapiEnsure your Docker container has the correct device passed through: (2) devices:
- /dev/dri/renderD128:/dev/dri/renderD128Things to Check on the Host
Testing Hardware AccelerationYou can verify VAAPI is working inside the container with: (3) If VAAPI also fails, the issue is definitively at the host/driver level rather than Frigate configuration. 📚 Sources: Have another question? Just tag @inkeep. |
Beta Was this translation helpful? Give feedback.
-
|
Restarting the entire host finally worked. Still not sure what caused this, or why after 2 years without issue, and 2 weeks on 0.17 without issue, it happened. |
Beta Was this translation helpful? Give feedback.
Restarting the entire host finally worked. Still not sure what caused this, or why after 2 years without issue, and 2 weeks on 0.17 without issue, it happened.