Environment
- Platform: OpenShift
- App: Microservices on Quarkus with native OpenTelemetry
- Database: Oracle (the Oracle Database technology appears; when clicked we only see the listener and a note that the extension has been removed)
- Deployment: Dynatrace Managed, Operator‑based (ActiveGate + OneAgent)
What worked before (baseline OK)
- Initial rollout with Cloud Native Full‑Stack using a namespaceSelector (label app:pippo) to restrict deep monitoring and enrichment to specific namespaces.
- After ActiveGate completed registration and OneAgent pods were restarted, application pods in the app:pippo scope were restarted as well.
- Result: full PurePaths, including Oracle SQL SELECTs. Quarkus OTel data was collected and correctly correlated into the traces.
Operational incident
- Some workloads failed Readiness probes during startup and were terminated by OCP, preventing a proper start of certain microservices.
- In parallel, workloads that did start successfully continued to produce complete traces, including OTel.
Change performed
- To reduce non‑essential startup activities on application containers, we removed the Dynatrace deployment and switched to Classic Full‑Stack.
- We disabled enrichment and moved instrumentation switches from the DynaKube YAML into platform Settings.
Current situation (regression)
- Application traces / PurePaths still exist and OTel is still ingested and correlated.
- However, we no longer see Oracle SQL blocks (SELECT/…) in the PurePath.
- In Technologies & Processes → Oracle, we only see the listener and a message stating the extension was removed.
- Collection of OCP cluster metrics/events via the OpenShift integration was never interrupted.
Hypothesis (what we want confirmed)
- The switch from Cloud Native Full‑Stack to Classic Full‑Stack changed the OneAgent injection timing relative to the Quarkus JVM bootstrap.
- In CNFS, the agent was injected before the Oracle driver initialization, enabling JDBC hooking, hence SQL statements appeared in the PurePath.
- In Classic FS, the agent seems to attach too late (or not at the correct phase), so driver hooking doesn’t happen—PurePaths and OTel remain, but no SQL shows up.
Questions to the community
- Can you confirm that, with the same code/driver, the loss of SQL statements in PurePath can depend on OneAgent injection mode/timing (CNFS vs Classic FS) on OpenShift/Quarkus?
- Do you have guidance or a checklist to ensure, in Classic Full‑Stack, that injection happens before JDBC (Oracle) initializes—so that SQL hooking is reliably enabled?
- In similar cases, did you resolve it by returning to Cloud Native Full‑Stack, or by tuning Container monitoring rules / init containers / restart order to guarantee the proper startup sequence under Classic FS?
- Are there any OneAgent settings/flags (e.g., container env variables, preloading options) recommended specifically for Quarkus + Oracle in Classic FS to force early JVM attach?
