Capture aggregate lock contention — concurrent worker updates

Now I have the cluster details. This is a latency cluster for `Api::V1::CapturesController#process_o

Let me search Datadog logs and explore the source code in parallel.

The endpoint returns 200 consistently (no errors), but this is a latency cluster (2146ms). Let me lo

The `process_output_upload_url` method is from the `ProcessOutputableController` concern. Let me fin

The repository method calls `@model.change_process_output_state('uploading')` then `@model.save`. Le

Now I understand the flow. The controller calls `repository_instance.process_output_upload_url` whic

The serializer calls `model.process_output_upload_url` which generates a presigned S3 URL. Let me ch

Now let me look at the `set_capture` before_action and the `CaptureRepository#show` to understand th

Now I can see the issue clearly. The flow is:

Interesting! The metrics show typical values around 0.2-0.3 seconds (200-300ms). But our cluster sho

Let me get the max duration metrics properly and check for the specific trace around the incident ti

The trace ID search shows worker-related logs (capture 702529 processing), not the actual API reques

I see multiple calls around the incident window — including duplicate calls to capture 702529 at 17:

Now I have a comprehensive picture. Let me look at the `set_storage` method to understand if it does

Now I have a full understanding. Let me also check how the `presigned_url` is generated — is there a

Now I have a complete picture. Let me also look at the `before_action` chain and any middleware that

Excellent! Now I see a very important pattern. At 17:22:05-17:22:14 KST for capture 702529:

Now let me update the todo and confirm completion. Per the instructions, I should NOT modify the clu

RCA report written to `./content/docs/incidents/6da25540-b3eb-4608-98c7-e9be40d1dddc/rca.mdx`.

RCA complete. The report is at `./content/docs/incidents/6da25540-b3eb-4608-98c7-e9be40d1dddc/rca.md