Documentation
¶
Overview ¶
Package scheduler is the generic, in-process recurring-evaluation seam on the agent runtime. It runs registered jobs at a fixed interval inside the worker lifecycle, so a consumer can re-pull signals and derive standing state on a schedule (e.g. learn metric baselines, evaluate SLO burn) without adding any new mutation path.
It is deliberately unopinionated and tier-neutral: the scheduler owns timing and lifecycle only; the content of every job belongs to the consumer. Registration mirrors the middleware.SetOrgResolver / SetAuthMiddleware seam — a consumer (the enterprise hooks.Register) registers jobs once at boot, before the worker starts. OSS registers none, so the scheduler is a no-op and community behaviour is byte-for- byte unchanged (no goroutines, no timers).
READ-ONLY CONTRACT (enforced by review + the import-graph guard in guard_test.go): a Job MUST be read-only / analyze-kind. It may pull signals and derive state, but it MUST NOT mutate cluster state and MUST NOT bypass the one emission path (services.CreateIncident) — findings it raises flow through that single path exactly like on-demand analyze. This package itself imports no write/emit/storage path, so a registered job cannot acquire write capability through the scheduler.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func Register ¶
Register adds a job to the process-wide registry. Call once at boot, before the worker starts. Returns an error for an invalid job or a duplicate name. This is the setter the enterprise hooks.Register attaches to (mirror of middleware.SetOrgResolver).
func Reset ¶
func Reset()
Reset clears the registry. Test-only helper to isolate cases; not used in production.
func SetOwnership ¶
SetOwnership installs the process-wide job-ownership predicate. A job runs on this instance only when pred(job.Name) is true. Passing nil clears the slot (back to own-everything). Last-wins: a second call replaces the first. Call at boot, before the worker starts. This is the entry point the enterprise cluster.Identity attaches to (mirror of middleware.SetOrgResolver); community OSS installs none.
Types ¶
type Job ¶
type Job struct {
// Name uniquely identifies the job (used in logs and dedup). Required.
Name string
// Interval is the period between runs. Required; must be > 0.
Interval time.Duration
// Run executes one evaluation. It MUST honour ctx cancellation and MUST
// be read-only / analyze-kind (see the package contract). Required.
Run func(ctx context.Context) error
}
Job is one registered recurring evaluation.
type Scheduler ¶
type Scheduler struct {
// contains filtered or unexported fields
}
Scheduler runs a fixed set of jobs until its context is canceled. Construct it with New (explicit jobs) or NewFromRegistry (the process-wide registry snapshot).
func NewFromRegistry ¶
func NewFromRegistry() *Scheduler
NewFromRegistry returns a Scheduler over a snapshot of the process-wide registry. The worker calls this; in community mode the snapshot is empty and Run is a no-op.
func (*Scheduler) Len ¶
Len reports how many jobs the scheduler holds in its snapshot. Run starts a goroutine and timer only for the subset this instance owns under the installed ownership predicate; with no predicate (community / single-instance) the owned set is every job, so Run behaves exactly as before.
func (*Scheduler) Run ¶
Run drives every owned job until ctx is canceled, one goroutine per job, and returns once all jobs have stopped. With no owned jobs it returns immediately and starts nothing — the OSS runtime is unaffected, and an un-owned job (one the ownership predicate rejects) starts no goroutine and no timer.
Source Files