Async Tasks Expert

Expert guidance for the Frappe Tweaks Async Task system — a managed, observable alternative to raw frappe.enqueue.

When to Use Async Tasks vs frappe.enqueue

Use Async Tasks when any of the following apply:

• You need to see task status, errors, or execution time in the UI
• Multiple agents/jobs can enqueue the same method and you need a cap on concurrency
• You want tasks to queue up and respect a priority order rather than flood workers
• The method is a whitelisted function or Server Script
• You need to be able to cancel tasks programmatically
• You want failed tasks to be automatically retried after a configurable delay

Stick with frappe.enqueue when:

• The job is fire-and-forget with no UI visibility required
• You are inside a framework hook that already handles retries/logging (e.g., Sync Jobs)

Public API

from tweaks.tweaks.doctype.async_task_log.async_task_log import (
    enqueue_async_task,
    enqueue_safe_async_task,
    bulk_enqueue_async_task,
    bulk_enqueue_safe_async_task,
)
from tweaks.tweaks.doctype.async_task_log.async_task_log_dispatch import (
    toggle_dispatcher,
    can_dispatch_now,
)

enqueueasynctask

task = enqueue_async_task(
    method=None,                    # dotted path str OR callable; optional when document_* fields are set
    queue="default",                # "default" | "short" | "long"
    timeout=300,                    # seconds; default 300
    job_name=None,                  # human-readable label (title); defaults to method when omitted
    # --- document action shorthand (alternative to method) ---
    document_type=None,             # DocType name  ┐ all three must be provided
    document_name=None,             # document name ┤ together; method is then auto-derived
    document_action=None,           # method to call on the document (e.g. "submit") ┘
    # --- retry options ---
    max_retries=0,                  # max automatic retry attempts on failure; 0 = no retries
    retry_delay=None,               # seconds to wait after failure before retrying; None = retry immediately
    # --- other options ---
    at_front=False,                 # jump ahead of other Pending tasks for this method
    call_whitelisted_function=False,# execute via call_whitelisted_function (Server Scripts)
    batch_id=None,                  # optional batch group label; tasks sharing a batch_id are ordered together
    batch_order=None,               # position within the batch; lower values dispatch first
    arguments=None,                 # explicit dict of method kwargs; overrides **kwargs on key collision
    **kwargs,                       # forwarded to method / document action as keyword arguments
)
# task.name  → Async Task Log document name

Rules:

• Pass either method or all three of documenttype + documentname + document_action. Passing neither raises ValueError.
• When the document fields are used, method is automatically derived as the doctype controller dotted path + ".{action}" (e.g. "erpnext.accounts.doctype.salesinvoice.salesinvoice.submit").
• jobname is an optional human-readable label stored as the task title. When omitted, it is set to method automatically in beforeinsert.
• Inner function resolution is applied at execution time: if the document controller defines submit, it is called instead of submit (mirroring Document.queueaction).
• Use arguments={"queue": "short"} (not queue= in kwargs) whenever a method argument name collides with an enqueueasynctask API parameter — arguments takes priority over kwargs and its keys are never intercepted by the function signature.
• max_retries=0 (the default) disables automatic retries. Set to a positive integer to enable them.
• retry_delay is the number of seconds to wait (measured from the task's modified timestamp on failure) before the next retry attempt. None or 0 means retry on the next dispatch cycle.

enqueuesafeasync_task

Shorthand for enqueueasynctask(..., callwhitelistedfunction=True). Use when calling whitelisted functions or Server Scripts by name. Accepts the same jobname, maxretries and retry_delay parameters.

task = enqueue_safe_async_task(
    "myapp.api.sync_customer",
    queue="short",
    customer_id="CUST-0001",
)

bulkenqueueasync_task

Create multiple Async Task Log documents in one call and dispatch them together as a single ordered batch.

tasks = bulk_enqueue_async_task(
    tasks=[                             # list of per-task dicts (same keys as enqueue_async_task)
        {"method": "myapp.utils.process_invoice", "invoice_name": "INV-001"},
        {"method": "myapp.utils.process_invoice", "invoice_name": "INV-002", "at_front": True},
    ],
    batch_id="my-import-run-abc",       # optional; auto-generated uuid4 when omitted
    queue="short",                      # kwargs here overwrite matching keys in every task dict
)

How it works:

1. If batch_id is not given, a random UUID is assigned so all tasks share the same batch.
2. Each task's batch_order is set sequentially (0, 1, 2 …) in insertion order.
3. The dispatcher is internally suspended while inserting to prevent partial dispatches, then resumed atomically once all documents are committed.
4. A single dispatchasynctasks job is enqueued at the end, respecting concurrency limits for all inserted tasks.

Key behaviour:

• Extra kwargs are merged into every** task dict (useful for shared fields like queue or timeout).
• Individual task dicts can still override merged values before calling enqueueasynctask.
• Returns the batchid string (auto-generated if not provided); use it to query Async Task Log by batchid to track completion.

bulkenqueuesafeasynctask

Shorthand for bulkenqueueasynctask(..., callwhitelisted_function=True). Use when each method is a whitelisted function or Server Script name.

bulk_enqueue_safe_async_task(
    tasks=[
        {"method": "myapp.api.sync_customer", "customer_id": "CUST-0001"},
        {"method": "myapp.api.sync_customer", "customer_id": "CUST-0002"},
    ],
    queue="short",
)

Passing a Callable

from myapp.utils import process_invoice

task = enqueue_async_task(process_invoice, invoice_name="INV-001")
# Internally resolves to "myapp.utils.process_invoice"

Document Action Shorthand

Use documenttype + documentname + documentaction instead of method when you want to call a method on a specific document in the background. This is the async-task equivalent of doc.queueaction().

# Submit a Sales Invoice in the background
task = enqueue_async_task(
    document_type="Sales Invoice",
    document_name="SINV-0042",
    document_action="submit",
    queue="long",
)

# Call a custom document method with kwargs
task = enqueue_async_task(
    document_type="My Doctype",
    document_name="MY-001",
    document_action="my_custom_method",
    some_arg="value",
)

Execution behaviour:

1. frappe.getdoc(documenttype, document_name) is called inside the worker.
2. doc.unlock() is called to release any file lock left from the caller.
3. Inner function resolution: if doc.submit exists and documentaction="submit", _submit is called instead.
4. getattr(doc, action)(**kwargs) is invoked with any extra kwargs.

Async Task Type concurrency is keyed on the derived method string, so you can set concurrencylimit on "erpnext.accounts.doctype.salesinvoice.sales_invoice.submit" as usual.

Async Task Type (optional configuration)

Create an Async Task Type document with method matching the dotted path to configure:

Creating via code (fixtures/patches):

frappe.get_doc({
    "doctype": "Async Task Type",
    "method": "myapp.utils.heavy_import",
    "priority": 10,
    "concurrency_limit": 2,
    "is_standard": 1,
}).insert(ignore_if_duplicate=True)

Status Lifecycle

Pending → Queued → Started → Finished
                           ↘ Failed ──(auto-retry)──→ Pending
          ↘ Canceled  (from Pending, Queued, or Started)

• Pending: Created, waiting for dispatcher to promote it.
• Queued: Pushed to RQ, waiting for a worker.
• Started: Worker picked it up.
• Finished / Failed: Terminal states. error_message populated on failure.
• Canceled: Manually canceled; RQ job stopped if already Queued/Started.
• Auto-retry: When a task has maxretries > 0 and retrycount < maxretries, the dispatcher automatically resets it to Pending after the retrydelay has elapsed since the last failure. retry_count is incremented each time a retry is triggered.

A realtime event asynctaskstatus is published to the creating user on every status transition (Queued, Started, Finished, Failed, Canceled). Use frappe.asynctasks.showprogress (see [JavaScript API](#javascript-api) below) for the idiomatic high-level approach. For raw access:

// Payload: { name, job_name, status, message, error, batch_id?, batch_done?, batch_total? }
// batch_id, batch_done, batch_total are only present when the task belongs to a batch.
// batch_done/batch_total are authoritative counters read from Redis — embedded in every batch task event.
// Always filter by name/batch_id — events are broadcast for ALL tasks.
frappe.realtime.on("async_task_status", ({ name, job_name, status, message, error, batch_id, batch_done, batch_total }) => { ... })

You can also push a custom message alongside a status update by calling notify_status() directly:

task.notify_status(message="Processing row 42 of 100...")

To send a progress notification from inside the executing method (where you don't have the task document), use the notifytaskstatus utility:

from tweaks.tweaks.doctype.async_task_log.async_task_log import notify_task_status

def my_long_running_job(items):
    for i, item in enumerate(items):
        process(item)
        notify_task_status(message=f"Processed {i + 1} of {len(items)}")

notifytaskstatus resolves the current RQ job, looks up the matching Async Task Log, and calls notify_status() on it. It is a no-op when called outside a worker context.

Structured progress messages

Pass a dict with a progress key to emit granular progress information. The frappe.asynctasks.showprogress handler understands this format and will drive the progress bar with exact count/total values instead of the default status-based percentages.

notify_task_status(message={
    "progress": {
        "count": i + 1,       # current step (used as the numerator)
        "total": len(items),  # total steps (used as the denominator)
        "description": f"Processing item {i + 1} of {len(items)}…",
    }
})

All three fields (count, total, description) are optional — include only the ones you need. When any field is absent the JS handler falls back to the default step-based value for that part of the progress bar.

JavaScript API

A thin JS namespace frappe.asynctasks is available on every desk page, provided by tweaks/public/js/tweaks/asynctasks.js.

frappe.asynctasks.showprogress(name, title?, handler?, hideoncompletion?)

High-level wrapper around frappe.realtime.on("asynctaskstatus", …) that drives a frappe.show_progress bar for a given task. Use this instead of wiring the raw realtime event by hand.

Behaviour:

• Immediately shows the progress bar at 10 % ("Pending").
• Registers the asynctaskstatus realtime listener so no transition event is missed.
• On Failed, calls frappe.throw(error) to surface the error message in a dialog.
• The realtime listener is automatically deregistered once a terminal status (Finished, Failed, Canceled) is received.
• When message is an object with a progress property, the handler reads progress.count, progress.total, and progress.description to drive the progress bar with exact values. Any missing field falls back to the default step-based value.

Important: handler is called on every matching event (Pending, Queued, Started, Finished, Failed, Canceled), not only on terminal ones. If your handler should only fire once (e.g. to transition to a second phase), guard it with a status check.

// Enqueue via a server call, then track progress
frappe.call({
    method: "myapp.api.start_import",
    callback(r) {
        frappe.async_tasks.show_progress(
            r.message,              // Async Task Log name
            __("Importing data"),   // progress bar title
            ({ status }) => {
                if (status === "Finished") frappe.show_alert(__("Import complete!"))
            }
        )
    },
})

Server-side structured progress (drives the bar with exact count/total):

# Inside the background method — no task document reference needed
from tweaks.tweaks.doctype.async_task_log.async_task_log import notify_task_status

def my_long_running_job(items):
    for i, item in enumerate(items):
        process(item)
        notify_task_status(message={
            "progress": {
                "count": i + 1,
                "total": len(items),
                "description": f"Processing item {i + 1} of {len(items)}…",
            }
        })

frappe.asynctasks.showbatchprogress(batchid, title?, handler?, hideoncompletion?)

Pure batch progress tracker. Shows a frappe.showprogress bar that counts individual task completions using server-authoritative Redis counters embedded in every asynctask_status event. The client owns no counter state — missed events are harmless.

Behaviour:

• Immediately shows an indeterminate bar (0 / 100, "Starting…") while waiting for the first event.
• Listens for asynctaskstatus events; filters by batchid and drops events where batchtotal == null (tasks arrived before Redis keys were written).
• Each event carries batchdone and batchtotal from Redis (set atomically by the server). The default bar label is "{jobname}: {statuslabel}" (or just "{statuslabel}" when no jobname). Structured msg.progress overrides count/total/description as usual.
• handler is called on every matching event (Pending, Queued, Started, Finished, Failed, Canceled). Guard with status === 'Finished' (or a count check) if you only want to act once on completion.
• The realtime listener is deregistered when a terminal-status event brings batchdone >= batchtotal.
• Does not handle the coordinator task — that is the caller's responsibility (see below).

Usage:

frappe.async_tasks.show_batch_progress(
    batch_id,
    __("Creating records"),
    ({ status, batch_done, batch_total }) => {
        if (status !== 'Finished' || batch_done < batch_total) return
        frappe.show_alert({
            message: __("{0} of {1} created", [batch_done, batch_total]),
            indicator: "green",
        })
        listview.refresh()
    },
)

Two-phase coordinator → batch pattern

When a coordinator task resolves a list of items and then calls bulkenqueueasync_task to fan out, orchestrate the two phases in the caller:

// Server returns { batch_id, coordinator_task_name } synchronously.
frappe.call({
    method: "myapp.api.bulk_create",
    callback(r) {
        const { batch_id, coordinator_task_name } = r.message

        const onBatchComplete = ({ status, batch_done, batch_total }) => {
            if (status !== 'Finished' || batch_done < batch_total) return
            frappe.show_alert({
                message: __("{0} of {1} created", [batch_done, batch_total]),
                indicator: "green",
            })
            listview.refresh()
        }

        // Phase 1: show coordinator bar. On Finished, transition to batch bar.
        frappe.async_tasks.show_progress(
            coordinator_task_name,
            __("Creating records"),
            ({ status }) => {
                if (status === "Finished")
                    frappe.async_tasks.show_batch_progress(batch_id, __("Creating records"), onBatchComplete)
            },
            false,  // don't auto-hide Phase 1 bar — Phase 2 will take over
        )
    },
})

Key point: Guard the showbatchprogress call with status === "Finished" because handler in showprogress is called on every event. Without the guard, showbatch_progress would be registered once per event (Pending, Queued, Started, Finished), leading to duplicate completion callbacks.

Server-side coordinator pattern:

import uuid
from tweaks.tweaks.doctype.async_task_log.async_task_log import (
    enqueue_async_task,
    bulk_enqueue_async_task,
    notify_task_status,
)

@frappe.whitelist()
def bulk_create(items):
    batch_id = str(uuid.uuid4())
    coordinator = enqueue_async_task(
        "myapp.api._coordinator",
        queue="short",
        job_name="Bulk Create",
        arguments={"items": items, "batch_id": batch_id},
    )
    return {"batch_id": batch_id, "coordinator_task_name": coordinator.name}

def _coordinator(items, batch_id):
    notify_task_status(message=f"Resolved {len(items)} items. Enqueueing…")
    tasks = [{"method": "myapp.api.process_item", "batch_id": batch_id, "arguments": {"item": i}} for i in items]
    bulk_enqueue_async_task(tasks, batch_id=batch_id)
    # bulk_enqueue_async_task writes Redis counters (total + done=0) with a 24 h TTL.
    # Batch task events will embed batch_done/batch_total from those keys automatically.

Cancellation

task = frappe.get_doc("Async Task Log", task_name)
task.cancel()  # works from Pending, Queued, or Started

Retry

Failed tasks can be retried manually or automatically.

Manual retry

task = frappe.get_doc("Async Task Log", task_name)
task.retry()        # reset to Pending and trigger dispatch
task.retry(now=True)  # enqueue for immediate execution (skip dispatcher)

retry() raises if called on a non-Failed/non-Canceled task.

Automatic retry

Set maxretries (and optionally retrydelay) when creating the task:

task = enqueue_async_task(
    "myapp.utils.sync_customer",
    max_retries=3,
    retry_delay=60,   # wait 60 seconds after failure before retrying
    customer_id="CUST-001",
)

The dispatcher calls retryfailedtasks() at the start of every dispatch pass. It queries all Failed tasks where maxretries > 0 and retrycount < maxretries, then resets those whose retrydelay has elapsed since their modified timestamp. Each retry increments retrycount. When retrycount reaches max_retries the task stays Failed and is no longer picked up automatically.

Key points:

• retry_count tracks how many automatic (or manual) retries have been triggered, not how many failures occurred.
• retry_delay=None (or 0) means retry on the very next dispatch cycle.
• Per-task errors in retryfailedtasks are logged and do not block retries of other tasks.

Observability

Each Async Task Log document stores:

• job_name — human-readable label used as the document title (defaults to method)
• startedat, endedat, time_taken
• peakmemoryusage (RSS, KB)
• error_message with full traceback on failure
• debuglog if frappe.debuglog is populated
• job_id linking to the underlying RQ Job
• documenttype, documentname, document_action — persisted when created via the document action shorthand; used at execution time to re-fetch the document and call the action
• maxretries, retrydelay, retry_count — retry configuration and current retry counter

Dispatcher Control

The dispatcher can be suspended site-wide, which prevents any new tasks from being promoted to Queued. Already-Queued workers continue running.

toggle_dispatcher (whitelisted)

Requires System Manager role. Persists the suspended/running state as a site default so it survives process restarts.

from tweaks.tweaks.doctype.async_task_log.async_task_log_dispatch import toggle_dispatcher

toggle_dispatcher(enable=False)  # suspend — no new tasks will be dispatched
toggle_dispatcher(enable=True)   # resume  — dispatcher runs normally again

Can also be called via the HTTP API (it is @frappe.whitelist()):

POST /api/method/tweaks.tweaks.doctype.async_task_log.async_task_log_dispatch.toggle_dispatcher
{ "enable": 1 }   // or 0

candispatchnow

Returns True when the dispatcher is running (not suspended). Use this guard before triggering dispatch manually:

from tweaks.tweaks.doctype.async_task_log.async_task_log_dispatch import can_dispatch_now

if can_dispatch_now():
    enqueue_dispatch_async_tasks()

Note: bulkenqueueasynctask internally suspends the dispatcher while inserting tasks and calls setdispatcherstate directly (bypassing the System Manager permission check that wraps the public toggledispatcher). Never call toggledispatcher from inside a background worker for internal use — use setdispatcher_state instead.

Dispatch & Recovery

Tasks are never dropped. The dispatcher runs:

1. After every new task insert (after_insert hook)
2. After every task completes (success or failure)
3. Via the scheduler (all event) as a recovery mechanism for missed triggers

See [references/implementation.md](references/implementation.md) for the full dispatch algorithm and concurrency internals. See [references/comparisonwithpreparedreport.md](references/comparisonwithpreparedreport.md) for a side-by-side schema and implementation comparison with Frappe's built-in Prepared Report. See [references/comparisonwithrqjob.md](references/comparisonwithrqjob.md) for a side-by-side schema and lifecycle comparison with Frappe's built-in RQ Job virtual DocType. See [references/comparisonwithscheduledjob.md](references/comparisonwithscheduledjob.md) for a side-by-side schema and implementation comparison with Frappe's Scheduled Job Type / Scheduled Job Log.

Source Code

• tweaks/tweaks/doctype/asynctasklog/asynctasklog.py — Public API + Document controller
• tweaks/tweaks/doctype/asynctasklog/asynctasklog_dispatch.py — Dispatch algorithm
• tweaks/tweaks/doctype/asynctasktype/asynctasktype.py — Type configuration