feat(stacks): pick GPU services from the stack's own services, not free text - #72
Merged
Conversation
…ee text The GPU control asked for a service name in a free-text field sitting under a "GPU passthrough" heading, so it read as "type a GPU here" — and even read correctly, it asked an operator to recall a compose service name. Watchtower knows the stack's services from its containers' labels, so the control becomes a toggle per service. - stacks.services: the stack's compose service names from com.docker.compose.service labels of its project's containers (one cheap Docker call, all states); a Docker outage is an empty list, not an error, because this is an input aid - GpuServiceEditor is a toggle per known service; a selected service the engine no longer reports still gets a row marked "not deployed", so a stored setting is never silently dropped - an "another service…" row keeps every case configurable, including a stack that has never been deployed (no containers, no known services) - the literal device editor's service column gets the same names as a datalist, and both sub-sections gained a line saying what they take
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Follow-up to ADR-0031 (merged in #71) from testing it on real hardware. The GPU control asked for a service name in a free-text field that sat under a "GPU passthrough" heading, directly above a line listing the detected GPUs — so it read as "type a GPU here". Even read correctly, it asked an operator to recall a compose service name from memory.
Watchtower already knows the stack's services, so the control is now a toggle per service — the checkbox originally asked for.
stacks.services— the stack's compose service names from its containers'com.docker.compose.servicelabels (one cheap Docker call, all states, so a stopped stack still lists them). Deliberately the deployed services rather than the repository's compose file: the file would have to be cloned and resolved per keystroke. A Docker outage returns an empty list rather than an error — this is an input aid, and failing it would take the whole Settings tab down with it.GpuServiceEditoris a toggle per known service. A selected service the engine no longer reports still gets a row, marked "not deployed" — silently dropping it would erase a stored setting the operator cannot see (profile-gated services, or a stack not currently up).datalist, and both sub-sections gained a line saying what they take.Why not select the GPUs themselves
Unchanged by design (ADR-0031 decision 5): the stored intent is "map the host's GPUs", resolved by probing on each deploy, which is what makes one setting portable to a host whose render node has a different number. Per-GPU selection only becomes meaningful on a multi-GPU host, and doing it properly needs the stable
/dev/dri/by-path/pci-…-renderalias rather thanrenderD128, since probe order can shuffle across reboots. Still deferred until a real multi-GPU need appears.Reviewer notes
rpc-schema.jsonregenerated (156 methods).Testing
Verified live against the dev instance: created two containers labelled into the stack's compose project so it reported real services, confirmed the picker listed
apiandtranscoder, toggled one on, saved through the actual Save button, and read the stored row back; then deleted that container and confirmed the row returned checked and marked "not deployed". Test containers and the row were cleaned up afterwards.Both backend suites green (1910 Application + 339 Api), frontend
tscand lint clean.