Skip to content

[Bug] [Infra]: doenet service sigkilled by OOM #2970

Description

@cqnykamp

Before submitting

Affected area

Other / not user-facing

Layer

Infrastructure / CI

Summary

Around 5-10 times per day, we're getting a 60 second downtime where no ECS tasks are live. Reported by https://community.doenet.org/t/503-and-504-errors/154.

Context

With our current load, our backend ECS tasks are getting sigkilled 5-10 times per day due to running out of memory. The memory issue isn't a Node/Javascript "Out of heap memory" issue, it's a container memory issues - something outside of Node is causing this.

I've verified this by looking through our load balancer logs, the ECS tasks events, and some ECS logs.

When we're running only one ECS task (which is most of the time), this causes a 60 second downtime of the backend.

This is probably a Prisma issue, since the version we're currently using runs a somewhat heavy Rust binary outside of the V8 engine. We should upgrade to a later version of Prisma which doesn't use Rust.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    Fields

    Priority

    Low

    Effort

    Medium

    Projects

    Status
    No status

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions