Problem
fetch_ecs_clusters returns an empty Vec on any error from ListClusters or DescribeClusters, so a denied or throttled ECS call renders as "no clusters" — indistinguishable from a genuinely empty account. This is the exact bug class #17 fixed for every other list fetcher (which now surface ServiceStatus), but ECS was left behind.
Evidence
src/aws/ecs.rs:78 — Err(_) => return vec![] (the ListClusters pagination loop)src/aws/ecs.rs:98 — Err(_) => return vec![] (the DescribeClusters chunk loop)src/aws/account.rs:42 — the STS get_caller_identity call degrades to an unknown identity with .ok() for the same reason
Both files carry TODO(#16) comments saying "once the error surface exists":
src/aws/ecs.rs:71 — "surface throttle/denied errors in the UI instead of degrading to an empty cluster list once the in-UI error surface exists"src/aws/account.rs:41 — "surface a denied/throttled sts:GetCallerIdentity in the UI once the error surface exists"
Why this is untracked
#16 (the in-UI error/notification surface) is closed and was built. So the remaining work these TODOs describe — making ECS and the STS identity fetch actually use that surface — is no longer tracked by any open ticket. The TODOs also now technically violate the AGENTS.md rule that a TODO must reference a ticket: they point at a closed one.
Proposed fix
ECS list fetchers return (Vec<EcsClusterInfo>, ServiceStatus) like the other services, classifying the error via ServiceStatus::from_sdk_error, so the view can show access-denied / unavailable instead of an empty table. Do the same for the STS identity in the account overview. Remove the two stale TODO(#16) comments once done.
Acceptance
Context
Found during the #38 session while adding ECS test coverage and fixing the ECS capacity placeholders (#43). Unrelated to those changes; pre-existing.
Problem
fetch_ecs_clustersreturns an emptyVecon any error fromListClustersorDescribeClusters, so a denied or throttled ECS call renders as "no clusters" — indistinguishable from a genuinely empty account. This is the exact bug class #17 fixed for every other list fetcher (which now surfaceServiceStatus), but ECS was left behind.Evidence
src/aws/ecs.rs:78—Err(_) => return vecsrc/aws/ecs.rs:98—Err(_) => return vecsrc/aws/account.rs:42— the STSget_caller_identitycall degrades to an unknown identity with.ok()for the same reasonBoth files carry
TODO(#16)comments saying "once the error surface exists":src/aws/ecs.rs:71— "surface throttle/denied errors in the UI instead of degrading to an empty cluster list once the in-UI error surface exists"src/aws/account.rs:41— "surface a denied/throttled sts:GetCallerIdentity in the UI once the error surface exists"Why this is untracked
#16 (the in-UI error/notification surface) is closed and was built. So the remaining work these TODOs describe — making ECS and the STS identity fetch actually use that surface — is no longer tracked by any open ticket. The TODOs also now technically violate the AGENTS.md rule that a TODO must reference a ticket: they point at a closed one.
Proposed fix
ECS list fetchers return
(Vec<EcsClusterInfo>, ServiceStatus)like the other services, classifying the error viaServiceStatus::from_sdk_error, so the view can show access-denied / unavailable instead of an empty table. Do the same for the STS identity in the account overview. Remove the two staleTODO(#16)comments once done.Acceptance
TODO(#16)comments remain (the ticket is closed)Context
Found during the #38 session while adding ECS test coverage and fixing the ECS capacity placeholders (#43). Unrelated to those changes; pre-existing.