Elastic Apm Symfony Bundle Laravel Package

Product Decisions This Supports

• Observability build-vs-buy: decides between (a) an in-app PHP bundle writing to Elastic APM Server, (b) the official elastic/apm-agent-php extension, or (c) OpenTelemetry PHP + Collector. 1.0.6 (listeners moved to the kernel.finish_request event) is a reliability fix that materially reduces the "traces arrive incomplete / spans missing at request end" failure mode — which is exactly the class of defect that makes an APM rollout look broken to stakeholders.
• Roadmap justification: funds SLO/error-budget work and latency regression detection by making per-route transaction and span data available without a platform-wide agent rollout.
• Incident-triage capability: gives on-call request-level traces correlated to errors in Kibana/APM, shortening MTTR for the services that adopt it.
• Adoption risk re-scoring: a 1.0.6 patch on a small, single-maintainer package is a positive maintenance signal but not a stability guarantee — treat it as "actively maintained, pre-2.0," suitable for incremental rollout behind a feature flag, not as a platform-wide standard.
• Instrumentation scope decision: because listeners now fire on finish_request (an HTTP-kernel event), this is scoped to web request lifecycles. If your roadmap includes Messenger consumers, queued jobs, or long-running workers (Octane, RoadRunner, Swoole), you must separately decide how those get traced — this package does not cover them by default.

When to Consider This Package

Adopt when:

• You are on Symfony (or a Laravel app comfortable pulling Symfony HttpKernel-based bundles) and already run, or have committed to, an Elastic APM Server (Elastic Cloud or self-hosted).
• You want code-level transactions/spans — per-route naming, custom spans, exception capture — without installing and operating a PHP extension across every host/container.
• You can accept a ~1.0.x package with a small maintainer base, and you pin the version and review config (sampling rate, capture_body, SQL bindings, header redaction) before production.
• APM is complementary: your product still functions correctly if telemetry drops.

Look elsewhere when:

• Your stack is Laravel-first: prefer the official elastic/apm-agent-php (zero-code auto-instrumentation, broader framework coverage) or OpenTelemetry PHP if you want vendor-neutral traces across non-PHP services.
• You need the same tracing standard across polyglot services, or you require vendor support/SLA and 24/7 maintainer responsiveness — neither is offered here.
• Most of your traffic is asynchronous (queues, workers). finish_request semantics won't help; validate coverage explicitly or pick an agent-based/OTel approach.
• You have strict data-governance constraints on what leaves the process — an extension-based agent often gives tighter, more auditable control than an in-app bundle.
• You need distributed tracing guarantees (context propagation across boundaries) as a hard requirement rather than a nice-to-have.

Note for this release: 1.0.6 is a behavioral fix, not a breaking change — listeners that previously ran earlier in the kernel lifecycle now run at finish_request. Verify any custom listener ordering or code that expected spans to be finalized before the response is sent; re-baseline dashboards after upgrade since trace completeness may visibly improve.

How to Pitch It (Stakeholders)

For executives: This is a low-cost, low-lock-in way to get real visibility into request latency and error rates, which is the prerequisite for setting SLOs and cutting mean time to recovery. The 1.0.6 release fixes a timing bug that caused traces to be incomplete at the end of a request — the exact symptom that makes an observability investment look like it isn't working. Recommendation: pilot on one or two high-value services, measure MTTR and latency-regression detection, then decide on broader rollout versus standardizing on the official Elastic agent or OpenTelemetry.

For engineering: A Composer-installable Symfony bundle that instruments transactions and spans and ships them to Elastic APM Server, no PHP extension to operate. 1.0.6 moves listener registration to kernel.finish_request so telemetry reliably flushes at request end — fewer dropped or partial traces. Before production: pin to an exact version, set the sampling rate deliberately, redact request bodies/headers/SQL bindings, and confirm whether finish_request-based capture is sufficient for your queue and long-running-worker workloads (it likely is not — plan a separate path for those).