Every controller in this package woke up on every update of the resources it
owns, including the ones that only touch fields it never reads. An
EphemeralRunner status carries readiness, failure bookkeeping and the job
details written by the listener, a pod reports IPs, its node, a start time and
several conditions, and an EphemeralRunnerSet rewrites
Status.FinishedRunnerCleanupPatchID for every listener patch id. None of that is
input to the owner, yet all of it enqueued a reconcile.
Add an update predicate to each owned watch. A predicate is only allowed to be
an optimisation, so each one is written as the projection of the fields its
reconciler actually reads and drops an update only when all of them are equal:
- AutoscalingRunnerSet -> EphemeralRunnerSet: object metadata, the whole spec,
and the Status.Phase and Status.AppliedActionableRevision pair that
ephemeralRunnerSetOutdatedForAppliedRevision consults.
- EphemeralRunnerSet -> EphemeralRunner: object metadata, spec, Status.Phase
and Status.RunnerID.
- EphemeralRunner -> Pod: UID, deletion timestamp, pod phase, reason and
message, the container and init container statuses, and the Ready
condition.
An event of an unexpected type is always delivered, and create, delete and
generic events are untouched. The primary watches keep seeing every update, so
the status patches a reconciler makes to hand work to its own next pass, such
as marking a runner Failed or Outdated before cleaning up its resources, still
re-enqueue.
The Ready condition lookup is extracted into podReady so that the predicate and
updateRunStatusFromPod cannot disagree about what readiness means.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The controller derived Status.Phase directly from the pod phase, so a runner
became Running as soon as its pod started, whether or not it had picked up a
job. That made Running mean "the pod is up" instead of "the runner is busy",
and it left the EphemeralRunnerSet scale-down path unable to tell an idle
runner from one that is executing a job.
The listener already knows when a job is assigned to a specific runner, so
move the transition there. HandleJobStarted now reads the runner first and
only promotes it to Running when it is not terminal (Failed, Succeeded or
Outdated) and not being deleted, then patches the phase alongside the job
fields it already writes. The listener role gains "get" on ephemeralrunners
for that read.
On the controller side updateRunStatusFromPod keeps publishing the initial
Pending phase while the pod is starting, and no longer promotes to Running.
Runners waiting for work now stay Pending, so scale-down picks them before
runners that are actually executing a job.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Reconcilers deep copied the object they had just fetched on every single
reconcile, purely so a merge patch could be computed on the rare pass
that actually changes something. The copy is a full recursive walk and
allocation of the object, and the overwhelming majority of reconciles
throw it away untouched.
Introduce lazyCopy, which takes the snapshot on the first call to
Mutate and hands back the live object. Because Mutate is the only way
to reach the object, the snapshot cannot be taken after the mutation it
is supposed to be diffed against, which is the way this optimization is
usually gotten wrong.
Apply it to the four Reconcile entry points, and move the two
EphemeralRunnerSet status copies inside the branch that patches, so they
are only paid for when the status really changed.
While here, drop the short circuit in the EphemeralRunner finalizer
block: `addedFinalizers || AddFinalizer(...)` skipped adding the actions
finalizer whenever the first finalizer was added.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
