Repository navigation
Promote agents to dev (2026-10-22, helper routing) - #789
Merged
Merged
Conversation
Use the caller's existing helper route for evaluation and planning instead of inheriting VM build capabilities. Preserve the full build expression, matrix mappings and reserved-lane selection; cover the real planner bindings in the capability gate.
Refresh only the canonical self revision and integrity after the helper-routing fix. Preserve every solved graph field and the complete Nix input graph.
Contributor
|
Thanks for your Pull Request! Below you will find a summary of the cachix status of each package, for each supported platform.
|
Render the local Incus wrappers and immutable client package from one producer script. Preserve existing daemon routing, tool selection and default-off behavior while allowing an explicit target Nix override.
Pin the qualified producer source through the ordinary canonical lock lifecycle without changing the dependency graph or selected toolchain.
The JIT .runner that GARM serves carries GitHub's workFolder ("_work",
relative to RUNNER_ROOT), so sandbox jobs ran in runner/_work instead of the
payload's work directory (HOME and the runner's cwd). The payload now
rewrites workFolder to the work directory before starting the runner, as
configure --work already does on the registration-token path. Other fields
are kept unchanged; a .runner without the field, or an empty one, gets it
added.
Covered by TestPayloadPointsRunnerWorkFolderAtWorkDir (GitHub's shape, an
absent field, an empty object), by the end-to-end fake listener, which now
refuses to start unless workFolder is its cwd, and by a ninth negative
control in t_garm_provider_agentharbor.
The devops-modules release train batches the provider changes into one promotion and so into one GARM restart window. AH3-F1 (JIT jobs run in the payload work directory) is part of that batch, so agents is merged in; this restarts the stabilisation. It also brings in 4dc9100/206d405d (shared host Nix client wrappers), which landed on agents after the cut.
Member
Author
|
Landing at c21ccb5. Every Linux check is green. The one red check is |
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
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.
Route flake matrix evaluation and planning through the caller's existing helper-runner input. Planning no longer inherits capabilities needed only by later VM builds. The complete build job, matrix mapping, and Foundry override are retained.
The capability gate checks the actual planner bindings; the previous production bindings fail the new control. The existing downstream caller placement contract and its 23 negative/vacuity controls also passed.
Verified locally on Linux at cc41f18:
Original CI on this promotion remains mandatory, including Darwin. These source checks do not establish deployed runner or shared-store reuse acceptance.
Update 2026-10-23: agents merged in (c21ccb5), restarting the stabilisation. Brings in the release-train batch so the provider changes reach dev, and therefore the fleet, in one GARM restart window:
.runnerworkFolder now points at the work dir instead of RUNNER_ROOT/_work).Verified locally on Linux x86_64 at c21ccb5:
checks.x86_64-linuxbuild. deployment-reconciler-timer-retry-vm and t_garm_capability_placement timed out at host load ~250 and passed on re-run.internal/backendpackage runs. It fails the same way on dev's own source, and passes 5/5 on its own.prek run --all-filespasses.Not verified locally: Darwin. The darwin batch failures on the previous head (nixos-rebuild-ng tmpdir test, lsof missing, durable-state test on m3 runners) also appear on dependabot PR #577 against dev, so they are runner environment, not this change.