Hacktoberfest 2026 : les issues que les mainteneurs ont marquées pour octobre, ouvertes et accessibles aux débutants. Parcourir les issues Hacktoberfest

[Feature Request] Add cancellation API for streaming callbacks

Ouverte
  • Quelque chose ne va pas ?
#4,004 2 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 2 jours

Personne n'a encore pris cette issue.

Évaluation

Difficulté
5/5
Temps estimé
Plus d'une semaine
Accessibilité débutants
45/100
Type d'issue
Fonctionnalité
Clarté
Plutôt claire
Activité
Active
Stack technique
javascript, python
Domaine
backend-api-design, frontend

Piste de recherche

Read the streaming callback implementation introduced in #3931 and the existing SharedWorker/page-close cancellation path. Confirm cancellation behavior for standalone NDJSON, multiplexed transport, async-generator cleanup, running outputs, and concurrent invocations; done means cancel=[Input(...)] works without application polling.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

enhancement P2 size: 5

Thanks so much for your interest in Dash!

Before posting an issue here, please check the Dash community forum to see if the topic has already been discussed. The community forum is also great for implementation questions. When in doubt, please feel free to just post the issue here :)

Is your feature request related to a problem? Please describe.

Dash 4.5 streaming callbacks allow an async def generator callback to stream yielded values to the browser, including incremental Patch updates. However, an application cannot currently provide a reliable stop button for an ordinary streaming callback through the public callback API.

running can keep start/stop controls in the appropriate loading and disabled states, but it does not cancel the active generator. Triggering the callback again also does not reliably stop the previous invocation. The existing cancel=[Input(...)] callback argument is limited to background callbacks.

The transport already has cancellation primitives, but they are not exposed to applications:

  • With multiplexed streaming, the renderer tracks a per-stream request ID and can send streamCancel; this is already used when a tab is closed.
  • Without shared storage, each stream falls back to its own NDJSON HTTP response. Aborting that request can close the response iterator and cancel the async generator, but application code has no public handle for the request or its AbortController.

As a result, applications must build a separate cooperative cancellation protocol, often involving ctx.shared_storage, unique stream IDs, polling a cancellation flag before every yield, and manual cleanup. This is much more application code than should be necessary for a fundamental streaming operation.

Describe the solution you'd like

Please support cancel=[Input(...)] for ordinary streaming callbacks, with semantics similar to background callback cancellation:

from dash import Input, Output, Patch, callback

@callback(
    Output("out", "children"),
    Input("start", "n_clicks"),
    cancel=[Input("stop", "n_clicks")],
    running=[
        (Output("start", "loading"), True, False),
        (Output("start", "disabled"), True, False),
        (Output("stop", "disabled"), False, True),
    ],
    prevent_initial_call=True,
)
async def stream(_):
    async for token in llm.stream(prompt):
        patch = Patch()
        patch += token
        yield patch

Ideally, Dash would implement cancellation according to the active transport:

  • Standalone NDJSON: abort the callback fetch request with its AbortController.
  • Multiplexed SharedWorker transport: send streamCancel for the active request ID without closing the shared downlink.
  • Server: cancel/close the active async generator, run its finally cleanup, restore all running outputs, and treat cancellation as expected control flow rather than a callback error.

This should work without requiring application-level polling or direct use of shared storage. If multiple invocations are permitted concurrently, the API should define whether the cancel input cancels all active invocations for that callback or only the invocation associated with the current output grouping.

Describe alternatives you've considered

  • Store cancellation flags in ctx.shared_storage and check them before every yielded item. This works across workers but requires stream IDs, polling, cleanup, and storage configuration in application code.
  • Use an in-process dict[str, asyncio.Event]. This works only in a single-process deployment and fails when the start and stop requests reach different workers.
  • Build a custom clientside component that retains the request's AbortController. This depends on renderer internals and duplicates transport logic.
  • Use a background callback only to obtain cancel=[Input(...)]. Background callbacks do not provide the same incremental streaming behavior.
  • Replace the native callback with EventSource or Socket.IO and implement a separate cancellation endpoint. This gives up the simplicity of native Dash streaming callbacks.

Additional context

Streaming callbacks were introduced in #3931. The existing SharedWorker/page-close cancellation path demonstrates that Dash already has most of the transport and server-side cancellation machinery. Exposing it through the callback API would make native streaming callbacks suitable for common LLM interfaces with explicit Start generating and Stop generating controls, while preserving the single shared browser connection used by multiplexed streams.

Langage dominant
Python
Étoiles
24.4k
Forks
2.3k
Merge moyen
2 j 14 h
PR mergées (30 j)
23

Préparer son environnement

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de plotly/dash

Toutes les issues de plotly/dash

Issues similaires

Plus d'issues Python

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.