Jellyfin hardware transcoding setting not applied after overnight update
Posted by ukonvasara
My question: I had to manually run these commands after sparkbox auto-update in order for jellyfin to have access to igpu at /dev/dri: sparkbox set-env JELLYFINHWACCEL vaapi and then sparkbox restart media SparkBox AI replied: Everything's up and healthy right now, and mounts are stock — /mnt/user/Media → /data like they should be, nothing custom broken. That matches what you described: this isn't an ongoing outage, it's that sparkbox update recreated the media containers without re-applying the docker-compose.hw.yml GPU overlay, so Jellyfin came back up minus /dev/dri until you manually forced it with set-env + restart media. That overlay is supposed to get re-added automatically on every up/update whenever JELLYFINHWACCEL is set and /dev/dri exists on the host — so if it dropped out after an auto-update, the most likely explanation is the update ran before that env value got read back in, or something about the auto-update path skipped the overlay step that a manual sparkbox up/restart does correctly. Since you already fixed it by hand and everything's healthy now, there's nothing left for me to bounce or repair on your box today. This is exactly the kind of thing that needs Tom's eyes rather than another restart from me — auto-update silently losing a hardware overlay is a real bug candidate, not a one-off misconfig. I'd click Share to demox below so he can check
4 replies
Chris wrote:
Hey ukonvasara -- you didn't do anything wrong, and what you did (set-env + sudo sparkbox restart media) is exactly the right fix. We closed a bug like this a while back where the GPU device could drop off during an update, so seeing it happen again on an overnight auto-update is worth a closer look. Can you run sparkbox status over SSH and tell us the version number at the top? If you're already on the newest build and it drops again after this, that tells us it's a fresh issue rather than the old one, and I've flagged it either way so Tom can dig in.
ukonvasara wrote:
v1.6.536
Chris wrote:
Thanks for confirming — v1.6.536 already includes the fix for that earlier bug where an update could drop /dev/dri from Jellyfin (that one was closed a little while back). So by Tom's own test here, seeing it happen again on your box means this is a fresh recurrence, not the old issue coming back. Good news is you already found the right workaround (set-env + restart media), so you're not stuck in the meantime. I've got this flagged for Tom's team to dig into why the overlay didn't reattach on its own this time — no further steps needed from you for now. Thanks for the extra detail, it'll help track it down.
Chris wrote:
Good news — Tom's team tracked this down and v1.6.541 ships today with the real fix. A single false negative on the GPU check during an update could get baked in as the box's 'normal' state, so every later nightly update matched that broken baseline and skipped reattaching the device — that's exactly what you hit. Run sudo sparkbox upgrade to grab it, and you shouldn't need the manual set-env/restart workaround going forward. Appreciate you sticking with this and confirming your version, that detail is what nailed it down.