Hacktoberfest 2026: as issues que os mantenedores marcaram para outubro, abertas e boas para iniciantes. Ver issues do Hacktoberfest

Logging: Reduce INFO log noise

Aberta
  • Algo errado?
#1,243 0 comentários 0 reações 1 responsável Ver no GitHub

Mantenedores costumam responder em até 1 dia

@vprashrex já está trabalhando nisso.

Desde 10/10/2026.

Avaliação

Esta issue ainda não foi avaliada.

Descrição

maintenance

Describe the current behavior
The backend ships every INFO log to Sentry Logs. Over 7 days the kaapi-production project received about 335k INFO logs. A large share is routine per-request or per-task chatter that is never read and adds cost and noise when debugging.

Unique logger.info call sites in the codebase (excluding tests and migrations), by module:

Module Files info warning error/exception
Knowledge base (collections, documents) 31 59 34 34
Assessment (API-client + submissions) 11 21 17 14
LLM call, chain, config & version 29 53 105 50
Evals 1.0 only (batch, Langfuse, embeddings) 8 42 31 22
Evals 2.0 only (judge, iteration loop) 14 4 9 6
Evals shared by 1.0 + 2.0 (core, cron, fast, dataset) 21 47 22 21
Onboarding (credentials, project, org, user-project) 12 25 11 9
Analytics (analytics + daily stats) 5 0 1 0

Whole app: 558 logger.info call sites.

Actual production INFO volume in Sentry, 7 days, by module:

Module INFO logs Main loggers
LLM call / chain ~77,800 services.llm.jobs 50.7k, crud.llm 16.9k, OpenAI provider 6.5k, Google AI Studio provider 3.7k
Evals (almost all shared code) ~15,000 crud.evaluations.cron_utils 13.4k, crud.evaluations.core 0.6k, services.evaluations.evaluation 0.5k
Assessment (API-client + submissions) ~6,500 crud.assessment.api 4.1k, services.assessment.api.batch 1.4k, callbacks / submission / submission_store 0.3k each
Knowledge base ~1,500 collection_job, crud.rag.open_ai, create_collection
Onboarding negligible
Analytics 0

Cross-cutting loggers outside these modules, also worth reviewing:

Logger INFO logs What it is
http_request_logger 86.7k one access line per request
app.celery.celery_app 18.5k [pool] status before and after every task
app.api.routes.cron 10.8k cron polling every 5 minutes
celery.worker.strategy 10.4k Celery "task received"
app.celery.tasks.job_execution 9.2k task wrapper
app.utils 8.6k mostly callback success lines
app.celery.utils 8.4k enqueue lines
app.core.langfuse.langfuse 8.2k tracing client

Describe the enhancement you'd like
Go module by module (one PR per module) and for each INFO call site:

  • Keep it if it marks a state change someone would search for (job created, job finished, provider called, callback sent).
  • Downgrade to DEBUG if it narrates internal steps of a single request or task.
  • Remove it if it duplicates another log, a span or a metric (for example per-task pool status, per-request access lines already covered by HTTP metrics and spans).

Suggested order by volume: LLM call/chain, cross-cutting Celery/cron loggers, Evals, Assessment, Knowledge base, Onboarding.

Why is this enhancement needed?
Lower Sentry Logs volume and cost, and faster debugging because the remaining INFO lines are the ones that matter.

Additional context

  • Assessment scope is the API-client BATCH path and submissions only (api/routes/assessment/api.py, services/assessment/api/, crud/assessment/api.py, crud/assessment/submission.py, services/assessment/submission.py). The legacy RUN pipeline (cron poller, UI routes, services/assessment/{service,stages,tasks}.py, crud/assessment/{cron,processing}.py) is outdated and out of scope.
  • Out of scope here but found while measuring: 46k uvicorn.access logs in kaapi-production come from kaapi-guardrails development containers using the production DSN (health-check lines). That is a guardrails configuration fix, not a log cleanup.
  • Counts are from 2026-10-10; re-measure in Sentry Logs (group by logger.name, filter severity:info) before and after each module.
Linguagem predominante
Python
Estrelas
18
Forks
11
Merge médio
4d 3h
PRs com merge (30d)
13

Preparar o ambiente

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de ProjectTech4DevAI/kaapi-backend

Todas as issues de ProjectTech4DevAI/kaapi-backend

Issues semelhantes

Mais issues de Python

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.