Memory usage spikes to ~3GB and never releases on coding-agent workload (restart required)

[ndaidong]

I run 2 GoModel instances with identical configuration (Postgres, Redis, and no semantic caching):

• Instance A (data processing)

  • ~40k requests/day
  • ~30M tokens/day
  • Memory remains stable and stays < ~150MB

• Instance B (coding agents)

  • ~1500 requests/day
  • ~20M tokens/day
  • When handling coding-agent workflows, GoModel memory grows very fast, sometimes reaching ~3GB
  • After the spike, memory does not decrease automatically; must restart the container to recover

I suspect the difference is related to how coding agents interact with AI providers. Coding agents typically run in loops and can send many requests in a short time, which may create sudden load. However, GoModel is mainly a middle layer that forwards parameters to providers and returns results, so it shouldn’t need to accumulate that much memory.

This behavior strongly suggests a potential memory leak or an object/lifecycle issue triggered by the coding-agent request pattern.

Please investigate and help identify the cause (memory leak / caching / request lifecycle / concurrency / streaming handling) and propose a fix so memory can be released or bounded under coding-agent workloads.

[SantiagoDePolonia]

@ndaidong what GoModel version do you use?

"When handling coding-agent workflows" - what endpoint and provider do you use? /responses ? /chat/completions?

Any non-standard settings I should be aware about?

[SantiagoDePolonia]

It might be caused by live audit logs preview. You can temporally try to disable it by setting: DASHBOARD_LIVE_LOGS_ENABLED=false in your .env

[ndaidong]

@SantiagoDePolonia I'm always using the last version of GoModel. All requests send to "/chat/completions".

General configs look like below:

LOGGING_ENABLED=true
LOGGING_LOG_BODIES=true
LOGGING_LOG_HEADERS=true
LOG_FORMAT=text
METRICS_ENABLED=false
BUDGETS_ENABLED=false
ADMIN_ENDPOINTS_ENABLED=true
ADMIN_UI_ENABLED=true
USAGE_ENABLED=true
USAGE_PRICING_RECALCULATION_ENABLED=true
ENFORCE_RETURNING_USAGE_DATA=true
BODY_SIZE_LIMIT=500KB
GOMODEL_MASTER_KEY={{__REDACTED__}}
CONFIGURED_PROVIDER_MODELS_MODE=allowlist
STORAGE_TYPE=postgresql
POSTGRES_URL=postgres://{{__REDACTED__}}
REDIS_URL=redis://:{{__REDACTED__}}
USAGE_RETENTION_DAYS=90
LOGGING_RETENTION_DAYS=15

Let me try to disable live audit logs as you suggested.

[SantiagoDePolonia]

Let me try to disable live audit logs as you suggested.

Let me know if it helped.

[ndaidong]

You may be right. The memory usage is quite stable with DASHBOARD_LIVE_LOGS_ENABLED=false