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

[Bug] Dropping any client stops log forwarding for every other client of the same Runtime

Ouverte
  • Quelque chose ne va pas ?
#1,988 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

@parbhatkapila4 y travaille déjà.

Depuis le 21/10/2026.

  • #1998 par @parbhatkapila4 — ouverte

Évaluation

Difficulté
3/5
Temps estimé
1-2 jours
Accessibilité débutants
57/100
Type d'issue
Bug
Clarté
Clairement spécifiée
Activité
Active
Stack technique
python, rust
Domaine
backend, observability

Piste de recherche

Start in temporalio/bridge/src/runtime.rs and trace how cloned Runtime values manage log_forwarder_handle during Drop; compare that ownership with the clone created in client.rs. Check the runtime and log forwarding tests for existing lifecycle coverage. Done means dropping one client no longer stops forwarded logs for clients still using the shared runtime, while forwarding stops when the runtime is no longer owned.

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

Description

What are you really trying to do?

Forward core logs to a Python logger with LogForwardingConfig, on a Runtime that several clients share. Some clients are short-lived (made for one call and dropped), and one is kept for a worker.

Describe the bug

When any client of a Runtime is dropped, log forwarding stops for every other client of that runtime, for the rest of the process. No error or warning is shown: the forwarded logger just receives nothing more.

The cause appears to be in temporalio/bridge/src/runtime.rs (same on main and in 1.34.0):

  • Runtime is #[derive(Clone)], and each client keeps a clone of it (client.rs: let runtime = runtime_ref.runtime.clone();).
  • log_forwarder_handle: Option<Arc<JoinHandle<()>>> is shared by every clone.
  • impl Drop for Runtime calls handle.abort() on that shared handle.

So the first clone to drop, which is the first client garbage-collected, aborts the forwarder that the RuntimeRef and every remaining client still rely on.

A possible fix: abort the forwarder only when the last owner goes away. For example, wrap the JoinHandle in a small type that aborts in its own Drop, and keep that type behind the Arc. Or implement Drop on RuntimeRef only.

Minimal Reproduction

Against a local dev server (temporal server start-dev):

"""Dropping one client of a Runtime stops log forwarding for every other client of that Runtime."""
import asyncio, gc, logging, sys, uuid
from datetime import timedelta
from temporalio import workflow
from temporalio.client import Client
from temporalio.runtime import LogForwardingConfig, LoggingConfig, Runtime, TelemetryConfig, TelemetryFilter
from temporalio.worker import Worker

@workflow.defn
class Noop:
    @workflow.run
    async def run(self) -> None:
        return None

class Count(logging.Handler):
    def __init__(self): super().__init__(); self.n = 0
    def emit(self, record): self.n += 1

async def main(drop_one: bool) -> int:
    handler = Count()
    logger = logging.getLogger(f"forwarded-{uuid.uuid4()}")
    logger.addHandler(handler); logger.setLevel(logging.DEBUG); logger.propagate = False
    runtime = Runtime(telemetry=TelemetryConfig(logging=LoggingConfig(
        filter=TelemetryFilter(core_level="INFO", other_level="WARN"),
        forwarding=LogForwardingConfig(logger=logger))))
    kept = await Client.connect("localhost:7233", runtime=runtime)
    if drop_one:
        other = await Client.connect("localhost:7233", runtime=runtime)
        del other
        gc.collect()
    async with Worker(kept, task_queue=f"q-{uuid.uuid4()}", workflows=[Noop]):
        await asyncio.sleep(3)
    await asyncio.sleep(1)
    return handler.n

if __name__ == "__main__":
    drop = sys.argv[1] == "drop"
    print(("drop one client" if drop else "keep every client"), "->", asyncio.run(main(drop)), "core records forwarded")

Output, run twice each (temporalio 1.32.0, CPython 3.13, macOS arm64):

keep every client -> 2 core records forwarded
drop one client -> 0 core records forwarded
keep every client -> 4 core records forwarded
drop one client -> 0 core records forwarded

The only difference between the two runs is a second client made on the same runtime and dropped before the worker starts. In a longer test, a worker kept alive for 25 s forwarded 18 core records. The same worker after one short-lived client on its runtime was dropped forwarded none. Eight clients kept alive forwarded 15; the same eight dropped forwarded 0.

The workaround is one Runtime per client. It costs a thread per CPU for each runtime.

Environment/Versions

  • OS and processor: macOS, Apple Silicon
  • Temporal version: Python SDK 1.32.0 reproduced; the same code is in runtime.rs in 1.34.0 and on main
  • Server: temporal server start-dev (Temporal CLI)
Langage dominant
Python
Étoiles
1.2k
Forks
269
Merge moyen
1 j 20 h
PR mergées (30 j)
60

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 temporalio/sdk-python

Toutes les issues de temporalio/sdk-python

Issues similaires

Plus d'issues Python

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.