Overview

Live state read from PostgreSQL and the running worker process.

—

Pending—queue depth

Running—claimed now

Completed—retained window

Dead—dead-lettered

Canceled—terminal

Workers—goroutines active

Completed / min—last 60 seconds

Reaped—stale recoveries

Controls

Every action creates real Dhara tasks. Nothing is simulated in the browser.

Burst jobs

PENDING → RUNNING

Enqueue echo jobs (2s each) and watch the worker pool claim them concurrently.

Batch size

Flaky jobs

RETRY

Retrying jobs fail temporarily, then succeed. “Always fail” exhausts attempts and dead-letters.

Mode Count

Slow tasks

TIMEOUT

Run slow_task. The handler exceeds its deadline and is retried, then dead-lettered.

Count

Worker pool

Dhara exposes process gauges and database leases, not per-worker heartbeats.

Configured—

Active—

Inflight—

Prefix—

Worker lease	Current task	Attempt	Last heartbeat
No running task leases.

Idle worker identities are not synthesized. Rows appear only when PostgreSQL shows a real locked_by lease.

Recent tasks

Click any task to inspect its real state and, for your own tasks, payload and logs.

Task	Type	State	Attempts	Worker	Next run	Duration	Created
Loading tasks…

— Polling pauses while this tab is hidden.

Prometheus metrics

Raw output from Dhara’s existing client and worker handlers. expand

These are process-local counters, not fabricated series.

Client / queue gauges `/metrics`

Not loaded.

Worker execution `/worker/metrics`

Not loaded.

What you are watching

Claiming

Workers claim rows with SELECT … FOR UPDATE SKIP LOCKED. Each PENDING task is claimed by exactly one worker.

Retries

Failed attempts are recorded in PostgreSQL and rescheduled with exponential backoff plus full jitter. The attempt counter is incremented on claim.

Heartbeat vs timeout

A handler timeout is an ordinary handler error. The reaper only acts when a RUNNING task stops refreshing locked_at — for example after the worker process dies.

Dead letters

Once attempts are exhausted, Dhara writes a dead_letters row. Retrying a DEAD task resets attempts and returns it to the ready queue.