adobe/outbound-call-timeouts

Beta Skill: This skill is in beta and under active development. Results should be reviewed carefully before use in production. Report issues at https://github.com/adobe/skills/issues

Outbound HTTP calls without a timeout — AEM as a Cloud Service

This pattern is executed by the code-assessment runbook — follow ../references/runbook.md for the full flow (preflight → plan → apply → verify, run log). This skill supplies the detection + recipe the runbook applies.

Overview

An HTTP client built without an explicit timeout inherits effectively-infinite connect/read defaults. When an upstream slows or hangs, every request thread blocks on that call; on AEM CS the shared Jetty request-thread pool saturates and the whole instance stops serving — this is the single highest-frequency cause in the CSO outage dataset. The fix sets a connect timeout and a read/socket timeout (and, for Apache, a connection-request timeout) so a stuck upstream fails fast instead of taking the instance down.

Classification — confirm this pattern applies

• An HTTP client is constructed in application code — Apache HttpClients.createDefault() / HttpClientBuilder.create()…build(), OkHttp new OkHttpClient() / …newBuilder()…build(), or JDK HttpClient.newHttpClient() / HttpClient.newBuilder()…build() — with no timeout configured in the same scope.
• The user asks to "add HTTP/outbound/external-call timeouts", or a scan flagged outbound-call-timeouts.
• Not this pattern: a client that already sets a timeout (any of RequestConfig, setConnectTimeout, connectTimeout, Timeout.of…, Duration.of… in scope); per-request timeouts only is a partial case the recipe handles.

Discovery

Detection is performed by the analyzer (../scripts/analyze.sh), run by the runbook:

bash ../scripts/analyze.sh <workspace-root> --pattern outbound-call-timeouts

Match criteria (what the detector flags):

• Client construction — text containing a known client type (HttpClientBuilder, HttpClients, OkHttpClient, HttpClient) and a construction marker (.createDefault, .newBuilder, .newHttpClient, .build(, new HttpClientBuilder, new OkHttpClient), emitted at the outermost fluent-chain link. Suppressed when the enclosing method (or, for a field initializer, the enclosing class) contains any client timeout token (setConnectTimeout, setSocketTimeout, setConnectionRequestTimeout, setResponseTimeout, connectTimeout, readTimeout, writeTimeout, callTimeout, RequestConfig, Timeout.of, Duration.of).
• JDK request builder — a HttpRequest.newBuilder()…build() chain (gated to java.net.http.HttpRequest by import/FQN so other libraries' HttpRequest is not matched). Suppressed when the chain itself sets .timeout(...). This is the per-request read timeout the JDK has no client-level setter for.

Parse-level only — no type resolution, so matching is textual.

Conservative posture (intentional false negatives). If a timeout is configured anywhere in the enclosing scope the site is skipped, even if it is the wrong knob — the detector prefers missing a real issue to flagging a safe client. The residual false-positive case (a client whose timeout is supplied from a different method — an injected RequestConfig or a shared HttpClientFactory) is caught at remediation by the recipe's skip policy, not at detection.

Resolution contract

self-evident — fixed default timeouts are applied; the developer supplies nothing up front (they review the resulting diff). Defaults: connect 3 s, read/socket 8 s, Apache connection-request 2 s — bounded against the infinite-hang outage and CSO-aligned; tighten in review if the upstream's SLA is stricter. The library and edit template are picked deterministically from the construction site's import package (org.apache.http → Apache 4.x, org.apache.hc → Apache 5.x, okhttp3 → OkHttp, java.net.http → JDK). All are fully mechanical — for the JDK client the connect timeout goes on the client, and JDK HttpRequest.newBuilder()…build() chains are detected and fixed separately with a per-request .timeout(...). Cases that cannot be completed deterministically (non-standard construction shapes) are skipped with a reason, never guessed.

Review checklist

• [ ] The timeout is applied to the flagged client construction, not a different client in the file
• [ ] Both a connect timeout and a read/socket timeout are set (connect-only still hangs on a slow response body)
• [ ] Apache: also set connectionRequestTimeout (pool-checkout) — a saturated connection pool otherwise blocks regardless of socket timeout
• [ ] Correct API for the library version — Apache 4.x RequestConfig.setSocketTimeout(int ms) vs 5.x Timeout / setResponseTimeout; OkHttp Duration overloads; JDK connectTimeout(Duration) + per-request .timeout(Duration)
• [ ] No value left at 0 / infinite; values are seconds-scale, not minutes
• [ ] Clients that already receive a timeout from a shared factory or injected config are skipped with a reason, not double-configured
• [ ] Surgical edit — no reformatting, imports added only as needed

Recipe

Read recipe.md in full before editing: input contract, locator, per-library remediation (Apache 4.x/5.x, OkHttp, JDK java.net.http), recommended values, skip policy, before/after, editing strategy.

Handoff

The skill never commits. See ../references/git-workflow.md for git vs in-place handoff and the suggested commit message.