Soft deleted workspaces have been around forever, as has this metric. I wonder if@kylecarbs can reach back that far in his memory and recall whether it was a deliberate decision to incluce builds from deleted workspaces in thecoderd_api_workspace_latest_build_total metric, or just an oversight.

The description of the metric talks about "current" number of workspace builds by status, which makes me think it was just an oversight. Since we track the latest build per workspace, not just all builds, the metric isn't useful for tracking an overall rate of succeeding / failing builds. My take is that the most obvious interpretation is that it's for non-deleted workspaces and this change is fine.

That is the conclusion I also came to when I read the descriptions, which is why I went with this approach.

If need be, we can include deleted builds with a specialized query that selects the least amount of data needed. Previously we selected all data only to use a single column which is why optimizations were not possible.

I've updated the PR to be a "breaking" change as well as describe the changed behavior. I've also updated the helps for the metrics to mention that it's for non-deleted workspaces.