go-again

go-again provides:

• again: a thread-safe retry helper with exponential backoff, jitter, timeout, cancellation, hooks, and a temporary-error registry.
• pkg/scheduler: a lightweight HTTP scheduler with pluggable state storage (in-memory by default, SQLite built-in) that reuses the retrier for retryable requests and optional callbacks.

Status

As of February 27, 2026, the core retrier hardening work and the scheduler extension described in PRD.md are implemented and covered by tests, including race checks.

Features

Retrier (github.com/hyp3rd/go-again)

• Configurable MaxRetries, Interval, Jitter, BackoffFactor, and Timeout
• Do and DoWithContext retry APIs
• Temporary error filtering via explicit error list and/or Registry
• Retry-all behavior when no temporary errors are supplied and the registry is empty
• Cancellation via caller context and Retrier.Cancel() / Retrier.Stop()
• Errors trace (Attempts, Last) plus Errors.Join()
• DoWithResult[T] helper
• Optional slog logger and retry hooks

Scheduler (github.com/hyp3rd/go-again/pkg/scheduler)

• Interval scheduling with StartAt, EndAt, and MaxRuns
• HTTP request execution (GET, POST, PUT)
• Retry integration via RetryPolicy
• Optional callback with bounded response-body capture
• URL validation by default (via sectools) with override/disable support
• Custom HTTP client, logger, concurrency limit, and scheduler-state storage backend

Installation

go get github.com/hyp3rd/go-again

Requires Go 1.26+ (see go.mod).

Retrier Quick Start

package main

import (
 "context"
 "errors"
 "fmt"
 "net/http"
 "time"

 again "github.com/hyp3rd/go-again"
)

func main() {
 retrier, err := again.NewRetrier(
  context.Background(),
  again.WithMaxRetries(3),                 // retries after the first attempt
  again.WithInterval(100*time.Millisecond),
  again.WithJitter(50*time.Millisecond),
  again.WithTimeout(2*time.Second),
 )
 if err != nil {
  panic(err)
 }

 retrier.Registry.LoadDefaults()
 retrier.Registry.RegisterTemporaryError(http.ErrAbortHandler)

 var attempts int
 errs := retrier.Do(context.Background(), func() error {
  attempts++
  if attempts < 3 {
   return http.ErrAbortHandler
  }

  return nil
 })
 defer retrier.PutErrors(errs)

 if errs.Last != nil {
  fmt.Println("failed:", errs.Last)
  return
 }

 fmt.Println("success after attempts:", attempts)
 _ = errors.Join(errs.Attempts...) // equivalent to errs.Join()
}

Retrier Notes

• MaxRetries counts retries after the first attempt (total attempts = MaxRetries + 1).
• If temporaryErrors is omitted and Registry has entries, the registry is used as the retry filter.
• If temporaryErrors is omitted and the registry is empty, all errors are retried until success/timeout/cancel/max-retries.
• Do checks cancellation between attempts. For long-running work, use DoWithContext.
• Cancel() and Stop() cancel the retrier's internal lifecycle context; they are terminal for that retrier instance.

Context-Aware Retrying

Use DoWithContext when the operation itself accepts a context and should stop promptly on cancellation:

// assuming `retrier` was created as in the previous example
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()

errs := retrier.DoWithContext(ctx, func(ctx context.Context) error {
 select {
 case <-time.After(250 * time.Millisecond):
  return nil
 case <-ctx.Done():
  return ctx.Err()
 }
})
defer retrier.PutErrors(errs)

The retryable function should observe ctx.Done(); if it ignores context cancellation, the work may continue running after the retrier returns.

Scheduler Quick Start

The scheduler runs jobs immediately when scheduled (or at StartAt if set), then continues every Schedule.Every until MaxRuns, EndAt, removal, or Stop().

Request and callback URLs are validated by default using sectools (HTTPS only, no userinfo, and no private/localhost hosts unless configured otherwise).

package main

import (
 "context"
 "net/http"
 "time"

 again "github.com/hyp3rd/go-again"
 "github.com/hyp3rd/go-again/pkg/scheduler"
)

func main() {
 retrier, _ := again.NewRetrier(
  context.Background(),
  again.WithMaxRetries(5),
  again.WithInterval(10*time.Millisecond),
  again.WithJitter(10*time.Millisecond),
  again.WithTimeout(5*time.Second),
 )

 s := scheduler.NewScheduler(
  scheduler.WithConcurrency(8),
 )
 defer s.Stop()

_, _ = s.Schedule(scheduler.Job{
  Schedule: scheduler.Schedule{
   Every:   1 * time.Minute,
   MaxRuns: 1,
  },
  Request: scheduler.Request{
   Method: http.MethodPost,
   URL:    "https://example.com/endpoint",
   Body:   []byte(`{"ping":"pong"}`),
  },
  Callback: scheduler.Callback{
   URL: "https://example.com/callback",
  },
  RetryPolicy: scheduler.RetryPolicy{
   Retrier:          retrier,
   RetryStatusCodes: []int{http.StatusTooManyRequests, http.StatusInternalServerError},
 },
})
}

Scheduler Examples

Example: Schedule Once + Callback

Runnable version:

go run ./__examples/scheduler

Source: __examples/scheduler/scheduler.go

s := scheduler.NewScheduler(
 scheduler.WithHTTPClient(server.Client()),
 scheduler.WithURLValidator(nil), // allow local endpoints for example usage
)
defer s.Stop()

jobID, err := s.Schedule(scheduler.Job{
 Schedule: scheduler.Schedule{Every: 10 * time.Millisecond, MaxRuns: 1},
 Request: scheduler.Request{Method: http.MethodGet, URL: server.URL + "/target"},
 Callback: scheduler.Callback{URL: server.URL + "/callback"},
})
if err != nil {
 panic(err)
}

payload := <-callbackCh
fmt.Println("job:", jobID, "success:", payload.Success, "status:", payload.StatusCode)

Example: Query Status and History

// after Schedule(...)
status, ok := s.JobStatus(jobID)
if ok {
 fmt.Println("state:", status.State, "runs:", status.Runs, "active:", status.ActiveRuns)
}

history, ok := s.JobHistory(jobID)
if ok {
 for _, run := range history {
  fmt.Println("run#", run.Sequence, "status:", run.Payload.StatusCode, "success:", run.Payload.Success)
 }
}

filtered := s.QueryJobStatuses(scheduler.JobStatusQuery{
 States: []scheduler.JobState{scheduler.JobStateRunning, scheduler.JobStateScheduled},
 Offset: 0,
 Limit:  50,
})
fmt.Println("filtered statuses:", len(filtered))

recentRuns, ok := s.QueryJobHistory(jobID, scheduler.JobHistoryQuery{
 FromSequence: 10,
 Limit:        5,
})
if ok {
 fmt.Println("recent retained runs:", len(recentRuns))
}

Example: Durable Scheduler State with SQLite

Runnable version:

go run ./__examples/scheduler_sqlite

Source: __examples/scheduler_sqlite/scheduler_sqlite.go

dbPath := filepath.Join(os.TempDir(), "go-again-scheduler-example.db")
storage, err := scheduler.NewSQLiteJobsStorageWithOptions(
 dbPath,
 scheduler.WithSQLiteHistoryMaxAge(24*time.Hour),
 scheduler.WithSQLiteHistoryMaxRowsPerJob(100),
)
if err != nil {
 panic(err)
}
defer storage.Close()

s := scheduler.NewScheduler(
 scheduler.WithJobsStorage(storage),
 scheduler.WithURLValidator(nil),
)
defer s.Stop()

jobID, err := s.Schedule(scheduler.Job{
 Schedule: scheduler.Schedule{Every: 20 * time.Millisecond, MaxRuns: 1},
 Request: scheduler.Request{Method: http.MethodGet, URL: target.URL},
})
if err != nil {
 panic(err)
}
fmt.Println("scheduled job:", jobID)

pruned, err := storage.PruneHistory()
if err != nil {
 panic(err)
}
fmt.Println("pruned rows:", pruned)

Example: Fail-Closed Scheduler Construction

Use NewSchedulerWithError(...) when constructor-time URL validator initialization errors must fail startup.

s, err := scheduler.NewSchedulerWithError(
 scheduler.WithConcurrency(8),
)
if err != nil {
 // fail startup instead of warning + degraded mode
 return err
}
defer s.Stop()

Scheduler Options

• WithHTTPClient(client) sets the HTTP client used for requests and callbacks.
• WithLogger(logger) sets the scheduler logger.
• WithConcurrency(n) limits concurrent executions when n > 0.
• WithJobsStorage(storage) sets pluggable scheduler state storage (active jobs plus status/history; default: in-memory).
• WithHistoryLimit(limit) sets retained per-job history length (default 20).
• WithURLValidator(validator) overrides URL validation. Pass nil to disable validation.
• NewSchedulerWithError(...) returns constructor errors (including startup state reconciliation failures and default URL validator initialization failure).

Scheduler Behavior Notes

• Supported methods for requests and callbacks: GET, POST, PUT.
• Callbacks are skipped when Callback.URL is empty.
• Callback method defaults to POST.
• Callback.MaxBodyBytes defaults to 4096.
• Request.Timeout and Callback.Timeout apply per HTTP request/callback (not the schedule lifetime).
• If RetryPolicy.Retrier is nil, the scheduler creates a default retrier and loads registry defaults.
• Calling Schedule after Stop() returns scheduler.ErrSchedulerStopped.
• Schedule returns scheduler.ErrStorageOperation when required scheduler-state persistence fails.
• NewSchedulerWithError(...) should be preferred for fail-closed startup behavior in security-sensitive paths.
• JobCount() and JobIDs() provide lightweight read-only scheduler introspection.
• JobStatus(id), JobStatuses(), and JobHistory(id) provide status and retained run history snapshots.
• QueryJobStatuses(JobStatusQuery) adds ID/state filters with pagination (Offset, Limit) over status snapshots.
• QueryJobHistory(id, JobHistoryQuery) adds history filtering (FromSequence) and tail limiting (Limit) while preserving ascending sequence order.
• Default InMemoryJobsStorage is process-local; use WithJobsStorage(...) for custom durable/backed storage.
• NewSQLiteJobsStorage(path) provides a built-in durable storage implementation for WithJobsStorage(...); call Close() when finished.
• NewSQLiteJobsStorageWithOptions(path, ...) configures SQLite retention controls: WithSQLiteHistoryMaxAge(duration), WithSQLiteHistoryMaxRowsPerJob(n), and WithSQLiteHistoryRetention(...).
• SQLiteJobsStorage.PruneHistory() and PruneHistoryWithRetention(...) provide manual pruning for periodic cleanup jobs.
• SQLite retention is also applied on write for new history records (age-based and max-rows-per-job), in addition to scheduler WithHistoryLimit.
• On scheduler startup, recovered active-job registrations from storage are reconciled: scheduled/running states are marked canceled, then active-job IDs are cleared. Jobs are not auto-resumed.
• Non-fatal storage write failures during runtime transitions are logged (warn) and execution continues.
• Non-fatal request/callback response body read/close failures are logged (warn) and execution continues.
• NewSchedulerWithError(...) fails constructor-time reconciliation errors; NewScheduler() logs a warning and continues.
• NewScheduler() logs a warning and continues if default URL validator initialization fails; use NewSchedulerWithError() to fail closed.

Custom URL Validation (Allow Local HTTPS)

validator, _ := validate.NewURLValidator(
 validate.WithURLAllowPrivateIP(true),
 validate.WithURLAllowLocalhost(true),
 validate.WithURLAllowIPLiteral(true),
)

s := scheduler.NewScheduler(
 scheduler.WithURLValidator(validator),
)

Disable URL Validation (Allow Non-HTTPS)

s := scheduler.NewScheduler(
 scheduler.WithURLValidator(nil),
)

Examples

Run the example programs directly:

go run ./__examples/chan
go run ./__examples/context
go run ./__examples/scheduler
go run ./__examples/scheduler_sqlite
go run ./__examples/timeout
go run ./__examples/validate

Development Commands

make test
make test-race
make lint
make sec

Benchmark (direct Go command):

go test -bench=. -benchtime=3s -benchmem -run=^$ -memprofile=mem.out ./...

Known Limitations / Gaps

• Scheduler.Stop() cancels the scheduler lifecycle; the same instance is not intended to be reused afterward.
• Retrier.Cancel() / Retrier.Stop() are terminal for the retrier instance.
• DoWithContext can only stop work promptly if the retryable function respects the provided context.
• NewScheduler() (non-error constructor) intentionally degrades to warning-only behavior if default URL validator initialization fails; use NewSchedulerWithError() when you need constructor-time failure.

Performance

go-again adds retry orchestration overhead but is designed to keep allocations low. See the benchmark in tests/retrier_test.go and run the benchmark command above in your environment for current numbers.

Documentation

• API docs: https://pkg.go.dev/github.com/hyp3rd/go-again
• Product/status notes: PRD.md

License

The code and documentation in this project are released under Mozilla Public License 2.0.

Author

I'm a surfer, a crypto trader, and a software architect with 15 years of experience designing highly available distributed production environments and developing cloud-native apps in public and private clouds. Feel free to hook me up on LinkedIn.