CaptureRepository#show N+1 query — DB contention

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

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

Now I know the service maps to the `tesla` repository. Let me search Datadog logs and explore the so

Good - found the relevant code files. The endpoint returns 302 redirects (typical for resource downl

The `download_multiple_resource` action is simple: it checks if `@resource.revision == 0` and either

Now I can see the key code path. The `download_url` method at line 203 calls `self.object(ver).presi

Now I see the full picture. The `download_url` method at line 203 calls `self.object(ver).presigned_

Let me search for APM traces instead since this is a latency cluster (captured from APM, not logs).

Now let me look for what authentication/authorization happens before action in this controller, as t

Now let me look at the `Storagable` module and the `storage_option` to understand how presigned URL

Now let me look at `self.class.show` which is the class method:

Now I have a good understanding. Let me also search for the `permission_joins` and `default_joins` t

The trace ID corresponds to a capture processing workflow (preprocessor complete), not the download

This is very telling. The `permission_joins` method on `CaptureRepository` (line 279-317) performs a

Excellent! Now I have a very clear picture. The `permission_joins` performs **15 LEFT JOIN subquerie

This is very insightful! The metrics show that:

I found the specific request. It was for capture 707591 downloading `processing_options` at `09:46:1

Directory exists. Let me write the RCA report:

Now let me update the cluster frontmatter. The user explicitly said NOT to modify the cluster frontm

RCA report complete. Here's a summary of the findings: