# AWS Public Endpoint Idle Timeout

<!-- auto-oncall-optimization-record
{
  "schemaVersion": "auto-oncall-optimization-record-v1",
  "optimizationId": "aws-public-endpoint-idle-timeout-2026-08-18",
  "problems": [
    {
      "problemId": "public-endpoint-timeout-misclassified-as-sql-performance",
      "summary": "TCID-13038 asked why DBeaver connections through an AWS Public Endpoint disconnect when SQL exceeds roughly five minutes, but auto-oncall classified it as SQL performance tuning and ran Clinic instead of answering the endpoint capability boundary.",
      "status": "optimized_verified",
      "applicability": "TiDB Cloud Starter or Essential on AWS where a SQL client uses a Public Endpoint and the connection is terminated around five minutes of idleness during a long-running statement.",
      "issueKeys": ["TCID-13038"],
      "sourceRefs": [
        "https://tidb.atlassian.net/browse/TCID-13038#problem:public-endpoint-timeout-misclassified-as-sql-performance"
      ],
      "signalTextIncludesAny": [
        "DBeaver",
        "Public Endpoint",
        "公网地址",
        "超过5分钟",
        "340 seconds",
        "AWS Global Accelerator idle timeout"
      ],
      "observedThrough": "2026-08-18",
      "effectiveAt": "2026-08-18T10:52:19Z",
      "verifiedAt": "2026-08-18T11:14:00Z",
      "attemptedAction": "Classify the bounded AWS Public Endpoint five-minute disconnect shape as a provider-free product-capability answer and supply scoped product knowledge for the 340-second Global Accelerator idle timeout, TCP keep-alive limitation, AWS PrivateLink alternative, and non-applicable fallback checks.",
      "evidenceUrl": "https://tidb.atlassian.net/browse/TCID-13038"
    }
  ]
}
-->

- Owner: Scenario Classifier and repository-owned product knowledge
- Source: `TCID-13038`
- Implementation status: merged in PR `#970`, deployed, and replay verified
- Effective status: `optimized_verified`

## Bounded problem

The customer asked how to handle DBeaver connections through an AWS Public
Endpoint that frequently disconnect when a statement exceeds about five
minutes. This entry problem is a documented connection-capability boundary for
TiDB Cloud Starter/Essential on AWS, not first a query-RCA problem. The previous
classification selected `sql_performance_tuning/clinic`, so the analysis spent
its budget on Slow Query evidence and did not answer the connection question.

## Intended change

- Normalize only the explicit Public Endpoint + SQL client + roughly five
  minute termination shape to `feature_usage_or_capability/direct_answer` with
  no diagnostic Provider.
- Preserve actual query slowdown RCA and real PrivateLink DNS/runtime incidents
  on their existing Provider-backed paths.
- Recall a scoped product topic containing the official 340-second idle
  timeout, TCP keep-alive limitation, AWS PrivateLink alternative, and the
  Starter/Essential-on-AWS applicability boundary.

## Verification status

Official product facts were checked against the current TiDB Cloud
Starter/Essential limitations and AWS PrivateLink documentation. Local
classifier, product-knowledge, build, and optimization-record tests are
pending. Case replay, deployment, and Jira internal-comment refresh have not
started.

## Unresolved gaps

The exact instance plan and endpoint type still control applicability. A
different cloud, tier, endpoint, or disconnect duration must not be forced into
the 340-second explanation.

## Next validation action

Run the focused local tests and one branch-local historical read replay. The
replay passes only when it uses the direct path, starts no Clinic diagnostic,
and answers the endpoint question with the scoped official facts.