asynqpg

Distributed task queue for Go, backed by PostgreSQL.

Consumers pull tasks concurrently without contention – Postgres hands each worker its own batch and skips rows already claimed by others, so throughput scales with workers. Producers can enqueue inside the same transaction as your business logic, so a job appears if and only if your write commits.

Features

Postgres-native – tasks stored in a single table; no Redis or external broker required
Safe concurrent processing – Postgres hands each worker its own batch and skips rows already claimed by others, so multiple consumers never touch the same task
Transactional enqueue – pass *sqlx.Tx to EnqueueTx so jobs commit atomically with your business logic
Flexible retries – exponential or constant backoff, snooze without consuming attempts, skip-retry for permanent errors
Delayed & scheduled tasks – WithDelay or direct ProcessAt for future processing
Idempotency tokens – EnqueueMany deduplicates by (type, idempotency_token) at the DB layer
Per-type worker pools – independent concurrency, timeout, and middleware per task type
Leader-elected maintenance – a single node holds the lease for stuck-task rescue and cleanup
Web dashboard – embedded React SPA with Overview, Tasks, Workers, and Maintenance pages; ⌘K palette and j/k navigation
Pluggable auth – Basic Auth out of the box; OAuth providers (GitHub included)
OpenTelemetry – built-in counters, histograms, and distributed tracing for all operations

Installation

go get github.com/yakser/asynqpg

Requires Go 1.25+ and PostgreSQL 14+.

Database Setup

Apply migrations to create the asynqpg_tasks table:

make up         # start PostgreSQL in Docker
make migrate    # apply migrations

Or use testcontainers-go in tests – no manual setup needed.

Quick Start

Producer

package main

import (
    "context"
    "encoding/json"
    "log"
    "time"

    "github.com/jmoiron/sqlx"
    _ "github.com/lib/pq"
    "github.com/yakser/asynqpg"
    "github.com/yakser/asynqpg/producer"
)

func main() {
    db, err := sqlx.Connect("postgres", "postgres://postgres:password@localhost:5432/asynqpg?sslmode=disable")
    if err != nil {
        log.Fatal(err)
    }

    p, err := producer.New(producer.Config{Pool: db})
    if err != nil {
        log.Fatal(err)
    }

    payload, _ := json.Marshal(map[string]string{"to": "user@example.com", "subject": "Hello"})

    _, err = p.Enqueue(context.Background(), asynqpg.NewTask("email:send", payload,
        asynqpg.WithMaxRetry(5),
        asynqpg.WithDelay(10*time.Second),
    ))
    if err != nil {
        log.Fatal(err)
    }
}

Consumer

package main

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

    "github.com/jmoiron/sqlx"
    _ "github.com/lib/pq"
    "github.com/yakser/asynqpg"
    "github.com/yakser/asynqpg/consumer"
)

func main() {
    db, err := sqlx.Connect("postgres", "postgres://postgres:password@localhost:5432/asynqpg?sslmode=disable")
    if err != nil {
        log.Fatal(err)
    }

    c, err := consumer.New(consumer.Config{Pool: db})
    if err != nil {
        log.Fatal(err)
    }

    if err := c.RegisterTaskHandler("email:send",
        consumer.TaskHandlerFunc(func(ctx context.Context, task *asynqpg.TaskInfo) error {
            fmt.Printf("Processing task %d: %s\n", task.ID, task.Type)
            // process task...
            return nil
        }),
        consumer.WithWorkersCount(5),
        consumer.WithTimeout(30*time.Second),
    ); err != nil {
        log.Fatal(err)
    }

    if err := c.Start(); err != nil {
        log.Fatal(err)
    }

    sigCh := make(chan os.Signal, 1)
    signal.Notify(sigCh, os.Interrupt, syscall.SIGTERM)
    <-sigCh

    if err := c.Stop(); err != nil {
        log.Printf("shutdown error: %v", err)
    }
}

Architecture

Task Lifecycle

flowchart LR
    %% Styles
    classDef pending   fill:#f59e0b,stroke:#92400e,color:#000
    classDef running   fill:#3b82f6,stroke:#1e40af,color:#fff
    classDef completed fill:#22c55e,stroke:#166534,color:#000
    classDef failed    fill:#ef4444,stroke:#991b1b,color:#fff
    classDef cancelled fill:#6b7280,stroke:#374151,color:#fff
    classDef edge      fill:#f3f4f6,stroke:#d1d5db,color:#374151

    %% States
    P["Pending"]:::pending
    T["Running"]:::running

    subgraph Final
        C["Completed"]:::completed
        F["Failed"]:::failed
        Ca["Cancelled"]:::cancelled
    end

    %% Entry points
    Enqueue(["Enqueue"]):::edge --> P
    EnqueueDelayed(["Enqueue (with delay)"]):::edge --> P

    %% Normal processing
    P -- "fetched by consumer" --> T
    T -- "handler success" --> C
    T -- "handler error, attempts_left > 0" --> P
    T -- "handler error, attempts_left = 0" --> F

    %% Rescuer (stuck tasks)
    T -- "stuck, attempts_left > 0 (Rescuer)" --> P
    T -- "stuck, attempts_left = 0 (Rescuer)" --> F

    %% Manual actions via client API
    F  -- "manual retry" --> P
    Ca -- "manual retry" --> P
    P  -- "manual cancel" --> Ca
    F  -- "manual cancel" --> Ca

    %% Cleaner (maintenance, leader-only)
    C  -. "auto-delete (Cleaner)" .-> Deleted(["Deleted"]):::edge
    F  -. "auto-delete (Cleaner)" .-> Deleted
    Ca -. "auto-delete (Cleaner)" .-> Deleted

    %% Manual delete via client API
    P  -- "manual delete" --> Deleted
    C  -- "manual delete" --> Deleted
    F  -- "manual delete" --> Deleted
    Ca -- "manual delete" --> Deleted

Tasks are fetched using SELECT ... FOR NO KEY UPDATE SKIP LOCKED, enabling safe concurrent processing across multiple consumers. The blocked_till column serves as both a delay mechanism (delayed enqueue, retry backoff) and a distributed lock expiry – a task re-appears for fetching only after blocked_till passes.

Packages

Package	Purpose
`producer/`	Enqueue tasks: `Enqueue`, `EnqueueTx`, `EnqueueMany`, `EnqueueManyTx`
`consumer/`	Fetch and process tasks with configurable worker pools
`client/`	Task inspection and management (get, list, cancel, retry, delete)
`ui/`	HTTP handler serving REST API + embedded React SPA

Producer

Create a producer with producer.New:

p, err := producer.New(producer.Config{
    Pool:            db,              // required: *sqlx.DB
    Logger:          slog.Default(),  // optional
    DefaultMaxRetry: 3,               // optional, default: 3
    MeterProvider:   mp,              // optional: OTel metrics
    TracerProvider:  tp,              // optional: OTel traces
})

Enqueue Methods

// Single task
id, err := p.Enqueue(ctx, task)

// Within an existing transaction (atomic with your business logic)
id, err = p.EnqueueTx(ctx, tx, task)

// Batch enqueue (skips duplicates by idempotency token)
ids, err := p.EnqueueMany(ctx, tasks)

// Batch within a transaction
ids, err := p.EnqueueManyTx(ctx, tx, tasks)

Task Options

asynqpg.NewTask("type", payload,
    asynqpg.WithMaxRetry(5),                       // max retry attempts
    asynqpg.WithDelay(10*time.Second),              // delay before first processing
    asynqpg.WithIdempotencyToken("unique-token"),   // deduplicate enqueues
)

// Schedule a task for a specific time
task := asynqpg.NewTask("report:generate", payload)
task.ProcessAt = time.Date(2026, 1, 1, 9, 0, 0, 0, time.UTC)

Consumer

Create a consumer with consumer.New:

c, err := consumer.New(consumer.Config{
    Pool:              db,               // required
    ClientID:          "worker-1",       // optional, for leader election
    RetryPolicy:       retryPolicy,      // optional, default: exponential backoff
    FetchInterval:     100*time.Millisecond,
    ShutdownTimeout:   30*time.Second,
    // Retention for completed/failed/cancelled tasks
    CompletedRetention: 24*time.Hour,
    FailedRetention:    7*24*time.Hour,
    CancelledRetention: 24*time.Hour,

    // DisableMaintenance: true,    // set to disable rescuer + cleaner
    // DisableBatchCompleter: true, // set to disable batch completions
})

Handler Registration

err := c.RegisterTaskHandler("email:send", handler,
    consumer.WithWorkersCount(10),          // goroutines for this task type
    consumer.WithMaxAttempts(5),            // override default
    consumer.WithTimeout(30*time.Second),   // per-task execution timeout
)

Implement consumer.TaskHandler or use the TaskHandlerFunc adapter:

type TaskHandler interface {
    Handle(ctx context.Context, task *asynqpg.TaskInfo) error
}

Retry Policies

Exponential backoff (default) – attempt^4 seconds with 10% jitter, capped at 24h:

&asynqpg.DefaultRetryPolicy{MaxRetryDelay: 24 * time.Hour}

Constant delay:

&asynqpg.ConstantRetryPolicy{Delay: 5 * time.Second}

SkipRetry

If a handler encounters a permanent error that should not be retried (e.g., invalid payload, business logic rejection), it can return asynqpg.ErrSkipRetry to immediately fail the task, skipping all remaining retry attempts:

func (h *EmailHandler) Handle(ctx context.Context, task *asynqpg.TaskInfo) error {
    var payload EmailPayload
    if err := json.Unmarshal(task.Payload, &payload); err != nil {
        // Invalid payload – retrying won't help
        return fmt.Errorf("bad payload: %w", asynqpg.ErrSkipRetry)
    }
    // process task...
    return nil
}

SkipRetry works with errors.Is, so it can be wrapped with additional context via fmt.Errorf("...: %w", asynqpg.ErrSkipRetry).

Snooze

Sometimes a handler needs to defer processing without counting it as a failure. TaskSnooze reschedules the task after a given duration without counting it as an attempt – attempts_left and attempts_elapsed remain unchanged:

func (h *MyHandler) Handle(ctx context.Context, task *asynqpg.TaskInfo) error {
    if !isExternalServiceReady() {
        // Try again in 30 seconds, doesn't count as a failed attempt
        return asynqpg.TaskSnooze(30 * time.Second)
    }
    // process task...
    return nil
}

If you want to reschedule and count it as a failed attempt (with error message stored), use TaskSnoozeWithError:

func (h *MyHandler) Handle(ctx context.Context, task *asynqpg.TaskInfo) error {
    if err := callExternalAPI(); err != nil {
        // Retry in 1 minute, counts as an attempt, error message is stored
        return fmt.Errorf("api unavailable: %w", asynqpg.TaskSnoozeWithError(1 * time.Minute))
    }
    // process task...
    return nil
}

Key differences:

	`TaskSnooze`	`TaskSnoozeWithError`
Counts as attempt	No	Yes
Stores error message	No	Yes
Respects max retries	No (unlimited snoozes)	Yes (fails when exhausted)
Use case	External dependency not ready	Transient error with custom delay

Both work with errors.As and can be wrapped with fmt.Errorf. Panics on negative duration; zero duration makes the task immediately available.

Task vs TaskInfo

The library uses two distinct structs to separate concerns:

Task (root package) -- the input struct for enqueueing. Contains only fields you set when creating a task: Type, Payload, Delay, MaxRetry, IdempotencyToken.
TaskInfo (root package) -- the runtime struct passed to handlers. Contains all database-assigned fields needed during processing:

func (h *MyHandler) Handle(ctx context.Context, task *asynqpg.TaskInfo) error {
    task.ID               // database ID
    task.Type             // task type
    task.Payload          // task payload
    task.AttemptsLeft     // remaining retry attempts
    task.AttemptsElapsed  // number of attempts already made
    task.CreatedAt        // when the task was first enqueued
    task.Messages         // error messages from previous failed attempts
    task.AttemptedAt      // when the current processing attempt started
    // ...
}

The client package has its own client.TaskInfo -- a full database read model returned by client.GetTask / client.ListTasks for inspection and management. It includes additional fields like Status, BlockedTill, UpdatedAt, and FinalizedAt.

Context Utilities

Task metadata is also available via context helpers, useful in middleware and utilities:

func (h *MyHandler) Handle(ctx context.Context, task *asynqpg.TaskInfo) error {
    id, _        := asynqpg.GetTaskID(ctx)         // database ID
    retry, _     := asynqpg.GetRetryCount(ctx)     // attempts already elapsed
    max, _       := asynqpg.GetMaxRetry(ctx)       // total max retry count
    createdAt, _ := asynqpg.GetCreatedAt(ctx)       // task creation time

    // Or get all metadata at once:
    meta, ok := asynqpg.GetTaskMetadata(ctx)
    // meta.ID, meta.RetryCount, meta.MaxRetry, meta.CreatedAt
    // ...
}

For testing handlers, use asynqpg.WithTaskMetadata to create a context with metadata:

ctx := asynqpg.WithTaskMetadata(context.Background(), asynqpg.TaskMetadata{
    ID: 42, RetryCount: 0, MaxRetry: 3, CreatedAt: time.Now(),
})
err := handler.Handle(ctx, task)

Middleware

The consumer supports composable middleware for cross-cutting concerns. Middleware wraps task handlers using the func(TaskHandler) TaskHandler pattern, similar to net/http middleware.

Global middleware applies to all task types:

c, _ := consumer.New(config)

_ = c.Use(func(next consumer.TaskHandler) consumer.TaskHandler {
    return consumer.TaskHandlerFunc(func(ctx context.Context, task *asynqpg.TaskInfo) error {
        slog.Info("processing task", "type", task.Type, "id", task.ID)
        err := next.Handle(ctx, task)
        slog.Info("task done", "type", task.Type, "id", task.ID, "error", err)
        return err
    })
})

Per-task-type middleware applies only to a specific handler:

c.RegisterTaskHandler("email:send", emailHandler,
    consumer.WithMiddleware(rateLimitMiddleware),
    consumer.WithWorkersCount(5),
)

Execution order: global middleware (outermost, first registered runs first) wraps per-task middleware, which wraps the handler. A middleware can short-circuit by returning an error without calling next.Handle.

Lifecycle

c.Start()               // start processing
c.Stop()                 // graceful shutdown (uses configured ShutdownTimeout)
c.Shutdown(timeout)      // graceful shutdown with custom timeout

Client

Inspect and manage tasks:

cl, err := client.New(client.Config{Pool: db})

// Get a single task
info, err := cl.GetTask(ctx, taskID)

// List tasks with filtering
result, err := cl.ListTasks(ctx, client.NewListParams().
    States(asynqpg.TaskStatusFailed, asynqpg.TaskStatusPending).
    Types("email:send").
    Limit(50).
    OrderBy(client.OrderByCreatedAt, client.SortDesc),
)
// result.Tasks, result.Total

// Manage tasks
_, err = cl.CancelTask(ctx, id)   // pending/failed → cancelled
_, err = cl.RetryTask(ctx, id)    // failed/cancelled → pending
_, err = cl.DeleteTask(ctx, id)   // remove from database

All methods have *Tx variants for transactional use.

Web UI

Mount the dashboard as an HTTP handler:

handler, err := ui.NewHandler(ui.HandlerOpts{
    Pool:   db,
    Prefix: "/asynqpg",
})
http.Handle("/asynqpg/", handler)

The dashboard is a single-page React app embedded in the Go binary via //go:embed. It ships as four operator pages plus a profile/appearance page, with a command palette (⌘K), keyboard navigation (j/k to move, x to select), and light/dark themes that follow the system preference.

Overview

Cluster snapshot at a glance: KPIs by status, per-task-type breakdown, current leader, and lease TTL. Auto-refreshes every 5 seconds.

Tasks

Live task list with saved views (All, Pending, Running, Failed, Needs retry, Dead-letter), filters by type / status / idempotency-token presence, full-text search by id / type / token, bulk retry / cancel / delete, and pagination. Clicking a row opens an inline drawer with tabs for payload, attempt history with error traces, timing breakdown, related tasks, the raw DB row, and a curl snippet to reproduce the call.

Workers

Shows the currently elected leader (with lease remaining + auto-renew countdown) and a live snapshot of every task in running status across the cluster, including current attempt and elapsed time.

Maintenance

Status mix at a glance, the most recent failures, and a description of the leader-elected background jobs (Rescuer, Cleaner, Leader election) that run only on the elected leader.

Authentication

Basic Auth:

ui.HandlerOpts{
    BasicAuth: &ui.BasicAuth{Username: "admin", Password: "secret"},
}

OAuth providers (e.g., GitHub):

ui.HandlerOpts{
    AuthProviders: []ui.AuthProvider{githubProvider},
    SecureCookies: true,    // for HTTPS
    SessionMaxAge: 24*time.Hour,
}

Implement the ui.AuthProvider interface to add any OAuth/SSO provider. Multiple providers can be configured in parallel — the login screen shows one button per provider. See examples/demo/github_provider.go for a complete GitHub OAuth implementation.

Observability

All public components accept optional MeterProvider and TracerProvider. When nil, the global OpenTelemetry provider is used.

Metrics

Metric	Type	Description
`asynqpg.tasks.enqueued`	Counter	Tasks enqueued
`asynqpg.tasks.processed`	Counter	Tasks finished processing
`asynqpg.tasks.errors`	Counter	Processing or enqueue errors
`asynqpg.task.duration`	Histogram	Handler execution duration (seconds)
`asynqpg.task.enqueue_duration`	Histogram	Enqueue latency (seconds)
`asynqpg.tasks.in_flight`	UpDownCounter	Currently executing tasks

All metrics are tagged with task_type and status attributes.

Tracing

Spans are created for all enqueue and processing operations with proper SpanKind (Producer/Consumer/Client).

Grafana Dashboard

A pre-built Grafana dashboard is included at deploy/grafana/dashboards/asynqpg-overview.json. It is provisioned automatically when you run make demo-up.

The dashboard refreshes every 10 seconds and includes:

Stat panels – total tasks enqueued, processed, errors, and currently in-flight
Enqueue Rate – tasks enqueued per second over time
Processing Rate by Status – throughput broken down by completed / failed / cancelled
Task Duration Quantiles – p50/p95/p99 handler execution latency
Enqueue Duration Quantiles – p50/p95/p99 enqueue latency
Error Rate by Type – error count per task type over time
Error Ratio – fraction of processed tasks that errored
In-Flight Over Time – concurrently executing tasks over time

All panels support a Task Type variable filter to drill down into a specific task type.

Observability Stack

The demo includes a full observability stack (Jaeger, Prometheus, Grafana, OTel Collector) in deploy/:

make demo-up   # start PostgreSQL + observability stack

Jaeger – http://localhost:16686
Prometheus – http://localhost:9090
Grafana – http://localhost:3000

Demo

An interactive TUI demo with producers, consumers, web UI, and observability.

Quick Start

make demo       # start infra, migrate, run demo
make demo-down  # stop all services

Or run steps individually:

make demo-up    # start PostgreSQL + observability stack
make migrate    # apply migrations
make demo-run   # run the demo app

CLI Flags

Pass flags via ARGS:

make demo-run ARGS="--tasks 50 --no-auto --log-level debug"

Flag	Default	Description
`--tasks, -n`	`100`	Number of initial tasks to seed
`--no-auto`	`false`	Disable automatic task generation
`--no-logs`	`false`	Hide the log viewport
`--log-level, -l`	`info`	Log level: `debug`, `info`, `warn`, `error`

TUI Commands

Command	Description
`enqueue <type> [N]`	Enqueue N tasks (types: `email`, `notification`, `report`)
`auto on\|off`	Toggle automatic task generation
`stats`	Show processing statistics
`clear`	Clear log viewport
`help`	Show available commands
`quit`	Graceful shutdown

Scroll logs with mouse wheel, arrow keys, or PgUp/PgDown. Press Tab to switch focus between input and log viewport.

Authentication

The demo supports optional authentication for the web UI via environment variables. Create a .env file in examples/demo/ or pass variables directly:

Basic Auth:

BASIC_AUTH_USER=admin BASIC_AUTH_PASS=secret make demo-run

GitHub OAuth:

GITHUB_CLIENT_ID=your_id GITHUB_CLIENT_SECRET=your_secret make demo-run

Without these variables, the UI runs without authentication.

Performance

The library includes Go benchmarks covering SQL-level operations, batch completion, producer throughput, and consumer processing.

make bench   # Run benchmarks (uses testcontainers, requires Docker)

Testing

make test              # unit tests (go test -race -count=1 ./...)
make test-integration  # integration tests (uses testcontainers, no manual DB needed)
make test-all          # both
make bench             # benchmarks (integration, requires Docker)
make lint              # golangci-lint

Contributing

Contributions are welcome, including bug reports, feature requests, documentation improvements, and code changes. See CONTRIBUTING.md for local setup, contribution workflow, testing expectations, and pull request guidelines.

Support

Website: yakser.github.io/asynqpg-landing
Bug reports & feature requests: GitHub Issues

TODO

Move testutils to a separate module to avoid pulling testcontainers-go into consumer projects

Project Status

Under active development. The API may change before v1.0. Bug reports and feature requests are welcome.

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 121 Commits
.github		.github
client		client
consumer		consumer
deploy		deploy
docs/images		docs/images
examples/demo		examples/demo
internal		internal
migrations		migrations
mocks		mocks
producer		producer
scripts		scripts
testutils		testutils
ui		ui
.gitattributes		.gitattributes
.gitignore		.gitignore
.golangci.yaml		.golangci.yaml
CHANGELOG.md		CHANGELOG.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
SECURITY.md		SECURITY.md
cliff.toml		cliff.toml
context.go		context.go
context_test.go		context_test.go
docker-compose.yaml		docker-compose.yaml
errors.go		errors.go
errors_test.go		errors_test.go
example_test.go		example_test.go
go.mod		go.mod
go.sum		go.sum
metrics.go		metrics.go
metrics_test.go		metrics_test.go
observability_integration_test.go		observability_integration_test.go
otel_attributes.go		otel_attributes.go
pool.go		pool.go
pool_adapter.go		pool_adapter.go
pool_integration_test.go		pool_integration_test.go
retry_policy.go		retry_policy.go
retry_policy_test.go		retry_policy_test.go
task.go		task.go
task_info.go		task_info.go
task_info_test.go		task_info_test.go
task_test.go		task_test.go
tracing.go		tracing.go
tracing_test.go		tracing_test.go

Folders and files

Latest commit

History

Repository files navigation

asynqpg

Contents

Features

Installation

Database Setup

Quick Start

Producer

Consumer

Architecture

Task Lifecycle

Packages

Producer

Enqueue Methods

Task Options

Consumer

Handler Registration

Retry Policies

SkipRetry

Snooze

Task vs TaskInfo

Context Utilities

Middleware

Lifecycle

Client

Web UI

Overview

Tasks

Workers

Maintenance

Authentication

Observability

Metrics

Tracing

Grafana Dashboard

Observability Stack

Demo

Quick Start

CLI Flags

TUI Commands

Authentication

Performance

Testing

Contributing

Support

TODO

Project Status

License

About

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages