commitmentopts: confirm MemoryDB probe target db.t4g.small has reserved-node offerings

[cristim]

#54's MemoryDB prober in internal/commitmentopts/probe.go uses db.t4g.small as the probe target instance type. Historical MemoryDB reserved-node coverage has been skewed toward db.r6g.* tiers, so db.t4g.small may return an empty offerings list in some regions.

Risk

If the probe returns [] for MemoryDB without erroring, the cache persists the probe run but includes no MemoryDB combos, and the frontend silently falls back to hardcoded rules for that service only. Functionally safe, but surprising ("why is MemoryDB not dynamic?").

Action

Run the following from a permissioned account:

aws memorydb describe-reserved-nodes-offerings --region us-east-1 --node-type db.t4g.small

If empty or sparse, switch the probe target to db.r6g.large (or the narrowest type with consistent RI coverage).

Surfaced during review of #54. Low-confidence flag; real AWS data needed.

[cristim]

Heads up: PR #69 documents this risk in code (probe.go comment near probeTargetMemoryDB) but does not switch the probe target — the automated session that opened the PR had no AWS credentials (aws sts get-caller-identity returned NoCredentials).

AWS verification still needed. From a permissioned account, please run:

aws memorydb describe-reserved-nodes-offerings --region us-east-1 --node-type db.t4g.small

If the offerings list is empty or sparse, switch probeTargetMemoryDB in internal/commitmentopts/probe.go to db.r6g.large (the inline comment above the constant block has the rationale). Until then, MemoryDB silently falls back to the frontend's hardcoded rules in regions where db.t4g.small returns nothing.

[cristim]

PR #69 (chore: MemoryDB probe target verification + page-cap test coverage) was merged at 2026-04-25T13:31:28Z and addresses this issue.

Recommend closing once the deploy is verified.

[cristim]

Verification ran today (2026-04-25) against us-east-1 with a non-CUDly AWS profile (the offerings catalog is global to AWS, so the result is account-agnostic):

$ aws memorydb describe-reserved-nodes-offerings --region us-east-1 --node-type db.t4g.small
{ "ReservedNodesOfferings": [] }

$ aws memorydb describe-reserved-nodes-offerings --region us-east-1 --node-type db.r6g.large
# 6 offerings: 1yr × {all-upfront, partial-upfront, no-upfront}
#              3yr × {all-upfront, partial-upfront, no-upfront}

Confirms the suspicion: db.t4g.small returns zero offerings, so the production probe was silently producing an empty MemoryDB combo set and the frontend was always falling through to hardcoded MemoryDB rules — exactly the surprising-but-functionally-safe degradation the issue called out.

Fix in #90 (swaps probeTargetMemoryDB to db.r6g.large).