cron

A modern, focused Go cron scheduler with no third-party dependencies.

Features

• Standard 5-field cron expressions plus @hourly / @daily / @every 10s descriptors, a per-spec TZ= prefix, and POSIX 7 as Sunday.
• Optional seconds field via WithSecondsField() (or a custom parser).
• Quartz tokens (L, L-n, LW, nW, N#M, NL, names like FRI#3) via the parserext subpackage.
• Schedule combinators: OnceAt, Union, and Filter (calendar exclusions).
• DAG jobs with conditional dependencies, data flow between steps, and per-step timeout/retry via the workflow subpackage.
• Job wrappers in wrap: Recover, Timeout, Retry, SkipIfRunning, DelayIfRunning. Job panics are recovered by default (ErrJobPanic).
• Per-event hooks and recorders so you can plug in metrics and tracing.
• Missed-fire policies (MissedSkip, MissedRunOnce, MissedRunAll) with a configurable tolerance window, overridable per entry; WithLastRun seeds cross-restart catch-up.
• Lifecycle control: Pause/Resume, in-place Update, graceful Drain or cancelling Stop.
• Distributed coordination: Locker (exactly-once per fire, clock-skew-proof fire-scoped keys) and Elector (leader-only scheduling), with Redis and Postgres backends as separate zero-impact modules.
• Manual Trigger, TriggerAndWait, and TriggerByName, with concurrency and entry limits.
• DST-correct scheduling (differentially fuzzed against a wall-clock scan). Per-entry timeout, jitter, retry, missed-fire policy, name, and chain.

Install

go get github.com/libtnb/cron

Requires Go 1.25+ (uses iter.Seq, sync.WaitGroup.Go, and slog).

Quick start

package main

import (
	"context"
	"fmt"
	"log/slog"
	"os/signal"
	"syscall"
	"time"

	"github.com/libtnb/cron"
	"github.com/libtnb/cron/wrap"
)

func main() {
	// Cancel on SIGINT / SIGTERM.
	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
	defer stop()

	// Build a scheduler. Wrappers in WithChain apply to every entry.
	c := cron.New(
		cron.WithLogger(slog.Default()),
		cron.WithChain(wrap.Recover(), wrap.Timeout(30*time.Second)),
	)

	// Add a job. Add returns an EntryID and a parse error (if any).
	_, err := c.Add("@every 5s", cron.JobFunc(func(ctx context.Context) error {
		fmt.Println("tick", time.Now())
		return nil
	}), cron.WithName("heartbeat"))
	if err != nil {
		panic(err)
	}

	// Start the loop. Idempotent while running.
	if err := c.Start(); err != nil {
		panic(err)
	}

	<-ctx.Done()

	// Drain in-flight jobs and hooks. The deadline caps the wait.
	shutdownCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
	defer cancel()
	_ = c.Stop(shutdownCtx)
}

Subpackages

Path	Purpose
github.com/libtnb/cron	Scheduler, parser, schedules, hooks, recorders, retry policy.
github.com/libtnb/cron/wrap	Job wrappers: Recover, Timeout, SkipIfRunning, DelayIfRunning, Retry.
github.com/libtnb/cron/workflow	DAG jobs with OnSuccess, OnFailure, OnSkipped, OnComplete.
github.com/libtnb/cron/parserext	Quartz tokens (L, N#M, NL).
github.com/libtnb/cron/lockers/redis	Redis Locker/Elector (separate module).
github.com/libtnb/cron/lockers/postgres	Postgres Locker/Elector (separate module).
github.com/libtnb/cron/contrib/prometheus	Prometheus metrics recorder (separate module).
github.com/libtnb/cron/contrib/otel	OpenTelemetry tracing wrapper (separate module).

Cron expressions

The default parser takes five fields:

minute hour day-of-month month day-of-week

Names are accepted (mon, MON, mon-fri, jan). Step (*/5), range (1-5), list (15,45), and combinations are supported. A spec may carry a TZ=Europe/Berlin or CRON_TZ=... prefix to override the scheduler's timezone for that entry.

The descriptors @yearly, @monthly, @weekly, @daily, @midnight, @hourly, and @every <duration> are also accepted. @every 90s is the canonical fixed-interval form; the interval has a 1s floor.

To use seconds, enable it on the built-in parser (keeps WithLocation working):

cron.WithSecondsField()

Or install a custom parser, which then owns the timezone:

// Optional seconds: 5- and 6-field specs both parse.
cron.WithParser(cron.NewStandardParser(cron.WithSeconds()))

// Strict: 6 fields required.
cron.WithParser(cron.NewStandardParser(cron.WithSeconds(true)))

Building a scheduler

c := cron.New(
	cron.WithLocation(time.UTC),
	cron.WithSecondsField(), // the spec below has a seconds field
	cron.WithMissedFire(cron.MissedRunOnce),
	cron.WithMaxConcurrent(32),
	cron.WithRetry(cron.Retry(3, cron.RetryInitial(time.Second))),
)

id, err := c.Add(
	"0 0 9 * * *",
	emailJob,
	cron.WithName("daily-digest"),
	cron.WithTimeout(time.Minute),
)

AddSchedule registers a programmatic Schedule instead of a string:

id, err := c.AddSchedule(cron.ConstantDelay(time.Hour), job)

Built-in schedules compose:

c.AddSchedule(cron.OnceAt(deployTime), job)          // fire exactly once
c.AddSchedule(cron.Union(weekdayNine, weekendTen), job)
c.AddSchedule(cron.Filter(daily, notHoliday), job)   // skip filtered firings

Entries can be paused, resumed, and updated in place — the ID and Prev survive; Update re-parses the spec, UpdateSchedule swaps a programmatic schedule:

c.Pause(id)  // manual Trigger still works while paused
c.Resume(id) // reschedules from now
err = c.Update(id, "*/10 * * * *")

Missed fires

When a firing runs more than WithMissedTolerance (default 1s) late, WithMissedFire decides what to do:

• MissedSkip (default) drops the missed firing and waits for the next scheduled time.
• MissedRunOnce runs the job once at the most recent missed time, then resumes the regular schedule.
• MissedRunAll runs the job once per missed firing (the newest 1000 are kept), then resumes.

The policy can be overridden per entry with WithEntryMissedFire, and jitter with WithEntryJitter. By themselves the policies cover in-process stalls (VM suspend, clock jumps, a backlog while the loop was blocked). To catch up across restarts, persist the last run time and seed it back:

c.Add("0 2 * * *", job,
	cron.WithLastRun(lastRunFromDB),             // first fire computes from here
	cron.WithEntryMissedFire(cron.MissedRunOnce)) // -> the 02:00 you missed runs once

Distributed coordination (multi-instance)

Running the same scheduler on N instances runs every job N times. Two first-class primitives fix that; the core stays dependency-free and backends live in separate modules:

go get github.com/libtnb/cron/lockers/redis      # package redislock
go get github.com/libtnb/cron/lockers/postgres   # package pglock

Locker — exactly once per fire. Every automatic fire claims the key <name>@<scheduledAt-unixnano>; one instance in the fleet wins, the rest skip and emit EventSkipped. Because the key identifies the fire (not the job), dedup depends on neither lock hold time nor clock agreement, and catch-up fires (MissedRunOnce/MissedRunAll) each claim their own key. Claims expire via TTL (default 10m) — set it to exceed your jitter plus clock skew.

client := redis.NewClient(&redis.Options{Addr: "localhost:6379"})
c := cron.New(cron.WithLocker(redislock.NewLocker(client)))
// Locked entries must be named (ErrLockerRequiresName otherwise): the name
// is the cross-instance key component, so keep it unique per job.
c.Add("0 * * * *", job, cron.WithName("hourly-report"))

Elector — one active instance. IsLeader gates every fire; on any error (including backend outage) the fire is skipped — fail-closed, nobody runs.

c := cron.New(cron.WithElector(redislock.NewElector(client)))

Postgres works through database/sql with any driver; create the tables once with pglock.Migrate(ctx, db). Table names are customizable via pglock.MigrateTables plus WithLocksTable/WithLeaderTable.

• Use schedules whose instants match across instances: cron expressions, OnceAt, or AlignedDelay. @every's phase is per-process (the scheduler warns).
• Override per entry with WithEntryLocker(l); WithEntryLocker(nil) opts an entry out of the global locker.
• Manual Trigger/TriggerAndWait bypass coordination: manual means "run it here, now".
• Observe skips with SkipHook (EventSkipped: not-leader, lock-held, or lock-error) and JobSkippedRecorder.
• cron.MemoryLocker/cron.MemoryElector (in the core, dependency-free) serve tests and single-process composition.

Triggering and removal

Trigger runs the job immediately. The returned error tells the caller why dispatch was rejected:

if err := c.Trigger(id); err != nil {
	switch {
	case errors.Is(err, cron.ErrEntryNotFound):
	case errors.Is(err, cron.ErrSchedulerNotRunning):
	case errors.Is(err, cron.ErrConcurrencyLimit):
	}
}

err = c.TriggerAndWait(ctx, id) // blocks and returns the job's error

count, err := c.TriggerByName("daily-digest") // err joins per-entry failures

c.Remove(id) // false if id is unknown

Remove blocks future automatic fires and future Trigger calls for that entry. Jobs already dispatched keep running. Two shutdown modes:

• Stop(ctx) cancels in-flight jobs (ErrCronStopping as the cause) and waits for the loop, jobs, and hook dispatcher, capped by ctx.
• Drain(ctx) stops scheduling but lets in-flight jobs finish naturally.

Job panics are recovered into ErrJobPanic-wrapped errors by default, so one bad job cannot crash the process; opt out with WithoutRecover().

Reading entries

Entry and Entries return copies and never block on the scheduler's internal lock, so they are safe to call from a hot path (HTTP handler, debug endpoint).

if entry, ok := c.Entry(id); ok {
	fmt.Println(entry.Name, entry.Next)
}

for e := range c.Entries() {
	fmt.Println(e.Name, e.Prev, e.Next)
}

NextN and Between operate on a Schedule directly, without a running scheduler:

next := cron.NextN(schedule, time.Now(), 10)
window := cron.Between(schedule, start, end)

Hooks and recorders

Hooks and recorders are split per event so a subscriber implements only the methods it cares about:

• Hooks: ScheduleHook, JobStartHook, JobCompleteHook, MissedHook.
• Recorders: JobScheduledRecorder, JobStartedRecorder, JobCompletedRecorder, JobMissedRecorder, QueueDepthRecorder, HookDroppedRecorder.

type metrics struct{}

// Implements JobCompleteHook only; the other 3 events are skipped automatically.
func (*metrics) OnJobComplete(e cron.EventJobComplete) {
	// record duration, error, etc.
}

c := cron.New(cron.WithHooks(&metrics{}))

Hooks are delivered on a buffered channel and dropped when the buffer is full. The size is configurable via WithHookBuffer and the drop count is exposed through HookDroppedRecorder.

Recorders, unlike hooks, are not serialized: their methods are called inline and concurrently from job goroutines, the scheduler loop, and Add/Remove/Trigger callers. Implementations must be concurrency-safe and non-blocking.

Ready-made observability lives in separate contrib modules, so the core stays dependency-free:

// go get github.com/libtnb/cron/contrib/prometheus
rec, _ := cronprom.New() // cron_jobs_{started,completed,missed,skipped}_total,
c := cron.New(cron.WithRecorder(rec)) // durations, lateness, queue depth...

// go get github.com/libtnb/cron/contrib/otel
c := cron.New(cron.WithChain(otelcron.Wrapper())) // one span per run, named
// after the entry via cron.EntryInfoFromContext, error status recorded

Jobs and wrappers can identify their own invocation through cron.EntryInfoFromContext(ctx) (ID, Name, ScheduledAt).

Workflow DAGs

workflow.Workflow is a cron.Job, so a DAG can be scheduled with Add or AddSchedule like any other job. workflow.New validates the graph and returns an error (ErrDuplicateStep, ErrUnknownDep, ErrCycle); workflow.MustNew panics on misconfiguration and is convenient for static graphs.

w := workflow.MustNew(
	workflow.NewStep("download", downloadJob).
		WithTimeout(2*time.Minute).
		WithRetry(cron.Retry(3, cron.RetryInitial(time.Second))),
	workflow.NewStep("transform", transformJob,
		workflow.After("download", workflow.OnSuccess)),
	workflow.NewStep("notify_failure", notifyJob,
		workflow.After("transform", workflow.OnFailure)),
)
_, _ = c.Add("@hourly", w, cron.WithName("etl"))

Conditions: OnSuccess, OnFailure, OnSkipped, OnComplete (any terminal state). A step is skipped when one of its dependencies didn't match the requested condition.

NewStepFunc steps pass data through the DAG — each receives its dependencies' outputs and Execution.Steps reports outputs and timings:

extract := workflow.NewStepFunc("extract", func(ctx context.Context, _ workflow.Inputs) (any, error) {
	return fetchRows(ctx)
})
load := workflow.NewStepFunc("load", func(ctx context.Context, in workflow.Inputs) (any, error) {
	return nil, store(ctx, in["extract"].([]Row))
}, workflow.After("extract", workflow.OnSuccess))

w := workflow.MustNew(extract, load).WithOnComplete(func(e *workflow.Execution) {
	log.Printf("run %s took %v", e.ID, e.Duration)
})

Quartz tokens

parserext.NewQuartzParser accepts standard 5/6-field specs and adds the Quartz day tokens:

Token	Field	Meaning
L	dom	last day of the month
L-3	dom	3 days before the last day
LW	dom	last weekday of the month
15W	dom	weekday nearest the 15th (never crosses month)
5#3	dow	third Friday (FRI#3 also works)
5L	dow	last Friday (FRIL also works)

c := cron.New(cron.WithParser(parserext.NewQuartzParser(time.UTC)))

_, _ = c.Add("0 0 18 L * ?", reportJob)      // last day of every month
_, _ = c.Add("0 0 9 ? * FRI#3", standupJob)  // third Friday
_, _ = c.Add("0 30 22 ? * FRIL", payrollJob) // last Friday
_, _ = c.Add("0 0 9 15W * ?", invoiceJob)    // weekday nearest the 15th

? is accepted in the day-of-month and day-of-week fields per the Quartz convention. Numeric day-of-week stays cron-style 0-6 Sunday-first (not Quartz's 1-7); the name forms are unambiguous.

Migrating from robfig/cron

robfig/cron	libtnb/cron
cron.New(cron.WithSeconds())	cron.New(cron.WithSecondsField())
Job.Run()	Job.Run(context.Context) error
c.AddFunc(spec, func())	c.Add(spec, cron.JobFunc(func(ctx) error { ... }))
cron.WithLogger(custom)	cron.WithLogger(*slog.Logger)
cron.Recover(logger)	wrap.Recover(wrap.WithLogger(logger))
cron.SkipIfStillRunning(logger)	wrap.SkipIfRunning()
cron.DelayIfStillRunning(logger)	wrap.DelayIfRunning()
c.Start()	c.Start() error
c.Stop()	c.Stop(ctx) error
c.Entries()	c.Entries() iter.Seq[Entry]

Credits

• robfig/cron

License

See LICENSE.