Fleet Logging Setup (logback convention)

Cross-service log correlation needs two things: every service writing to one known directory, and trace keys (trace_id, span_id, sessionId, …) present in each line. This skill installs a logback convention that guarantees both. It is idempotent and opt-in per service.

What trace keys to use (2026)

• Primary: OpenTelemetry / Micrometer Observation. Add the spring-boot-starter-actuator + micrometer-tracing-bridge-otel (or, on Spring Boot 4, the dedicated spring-boot-starter-opentelemetry). The W3C traceparent header is auto-propagated across HTTP/gRPC, and the Observation API automatically populates MDC keys trace_id (32-char hex) and span_id (16-char hex). These are the modern, framework-default keys.
• Fallback / legacy: sessionId, requestId. Keep them in the pattern for fleets that pre-date the Observation API or that propagate an application- level session id (auth, X-Request-Id) the user wants to correlate on.

When to use

• /debug reports missing log files, or logs only go to the IDE console.
• Logs exist but lack a trace key, so correlation falls back to timestamps.

Procedure

1. Read the config for logDir and traceKeys.
2. Start from the template: ${CLAUDE_PLUGIN_ROOT}/logback/logback-spring.xml. For each service you want covered, place it at <service-repo>/src/main/resources/logback-spring.xml (or merge its appender
   • pattern into the existing one), then customize three things:
   • SERVICE_NAME → the service's name from the config (used as the log file base name, so /debug can find it).
   • The %X{...} keys in FLEET_PATTERN → one per entry in traceKeys.
   • SPRING_FLEET_LOG_DIR → the config logDir, or set that env var when launching the service.
3. Ensure the trace keys are in MDC. A pattern can only print %X{trace_id} if something put trace_id into the MDC.
   • trace_id / span_id: come for free once Micrometer Tracing or the OTel starter is on the classpath; Spring Boot wires the Observation API which populates MDC automatically. Verify by hitting any endpoint and tailing the log; the keys should already be there.
   • sessionId / requestId (legacy): usually require a filter/interceptor that reads the incoming header (or generates the id), calls MDC.put(...), and clears it in a finally. If propagation across services is missing, the same id must be forwarded as a header on outbound proxy-lib calls. (Modern fleets should rely on W3C traceparent instead and treat sessionId as a business id, not a correlation id.)
4. Verify. Run the service, exercise one request, then:

   python "${CLAUDE_PLUGIN_ROOT}/scripts/tail_logs.py" \
     --config ./spring-fleet.config.json --service <name> --lines 20

   Confirm lines land in <logDir>/<service>.log and carry the trace key.

Notes

• Keep the same pattern across services so timestamps sort cleanly when merged. The template's yyyy-MM-dd HH:mm:ss.SSS prefix is what the correlator parses.
• The template also keeps a CONSOLE appender, so IDE output is unchanged.
• For services you will not modify, prefer the (future) /run tee approach instead of editing their logback config.

Outside Claude Code

• ${CLAUDE_PLUGIN_ROOT} is the spring-fleet plugin root. Claude Code fills it in; in Codex it is the folder two directories above this SKILL.md.