Scheduler processes one batch per worker per tick: default throughput capped at <10 msg/s, backlog grows unbounded

Ouverte
  • Quelque chose ne va pas ?
#132 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

Personne n'a encore pris cette issue.

Évaluation

Difficulté
3/5
Temps estimé
1-2 jours
Accessibilité débutants
52/100
Type d'issue
Bug
Clarté
Clairement spécifiée
Activité
Active
Stack technique
kotlin
Domaine
backend

Piste de recherche

Start in the OutboxScheduler in okapi-core, where scheduleWithFixedDelay runs processNext(batchSize) once per worker per tick, and read processBatch to see how each tick's transaction is opened. Change it so the loop keeps draining while a batch comes back full, with one transaction per iteration and a time or iteration budget that still lets stop() return promptly. Done means a sustained high publish rate no longer backs up the queue under default settings, stop() stays responsive, and the existing tests pass; the issue also asks for a benchmark of sustained throughput.

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

Description

bug critical

Severity: High (performance / unbounded backlog) · Area: okapi-core OutboxScheduler · Found in: code review of main @ e5630a1

Problem

Each scheduler tick has every worker call processNext(batchSize) exactly once, using scheduleWithFixedDelay. That puts a hard ceiling on throughput:

max throughput ≈ concurrency × batchSize / (interval + batch processing time)

With the defaults (concurrency = 1, batchSize = 10, interval = 1s) that is under 10 messages/second, regardless of how fast the DB and transport are. Any sustained publish rate above that grows the backlog without bound, and latency grows linearly with it. A burst of 100k messages takes about 3 hours to drain with the defaults, even when each batch takes milliseconds.

Raising interval to cut idle polling makes this worse. Raising batchSize makes transactions longer (see the related issues on gap locks and delivery inside the transaction).

Suggested fix

In processBatch, keep draining while the previous batch was full. Each iteration gets its own transaction, and the loop is bounded by a time or iteration budget so stop() stays responsive:

do {
    val n = transactionRunner.runInTransaction { outboxProcessor.processNext(config.batchSize) }
} while (n == config.batchSize && running.get() && !budgetExceeded())

That way the poll interval only governs idle latency, not peak throughput. The benchmarks should cover a sustained-rate scenario as well as fixed-backlog drains.

Langage dominant
Kotlin
Étoiles
14
Forks
0
Merge moyen
14 h 16 min
PR mergées (30 j)
9

Préparer son environnement

Ce projet ne fournit ni conteneur de développement, ni Dockerfile, ni guide de contribution : l'installation est à votre charge. Commencez par son README, et consultez notre guide de la première contribution pour les étapes générales.

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 softwaremill/okapi

Toutes les issues de softwaremill/okapi

Issues similaires

Plus d'issues Kotlin

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.