fix(ontoserver): expose podManagementPolicy and OTel exporters, restore empty serverPort default

[edeati]

Three issues found while migrating an install to the 0.4.0 chart and doing a throwaway scaled install on EKS alongside it.

1. podManagementPolicy was hardcoded to Parallel

statefulset.yaml set podManagementPolicy: Parallel as a literal. Under Parallel the controller does not wait for a replacement pod to be Ready before replacing the next one, so a rolling update on a multi-replica install can terminate every pod within seconds and leave the Service with no ready endpoint until the first replacement passes its readiness probe. Observed in the field: pods going ~27s apart, ~74s with no ready endpoint, some 503s. A PodDisruptionBudget does not prevent this — it only gates the Eviction API (node drains), not StatefulSet-controller-driven replacement.

Now ontoserver.deployment.podManagementPolicy, defaulting to Parallel so existing installs render identically and keep the fast parallel scale-up. Schema-constrained to Parallel | OrderedReady.

The field is immutable on a live StatefulSet, so the README documents the kubectl delete statefulset <release>-statefulset --cascade=orphan recreate needed to adopt it, and the trade-off that OrderedReady serializes startup and scale-up as well — which matters when pods need a long Lucene index preload under healthCheckOption: -s.

2. OTEL_METRICS_EXPORTER / OTEL_LOGS_EXPORTER were hardcoded to none

Both were literals in the Instrumentation resource with no values override, which puts metricsExporter: otlp — the usual way to get JVM heap and GC metrics off the Java agent — out of reach. Unlike most other agent settings it cannot be worked around from outside the chart.

Now ontoserver.opentelemetry.instrumentation.metricsExporter and .logsExporter, both defaulting to "none", so an existing install renders unchanged.

3. serverPort default went from "" to "8080" ⚠️ behaviour change

b37ce47 changed the default to "8080" so the local port-forward instructions would produce working URLs. The side effect: every install that never sets serverPort — including the aks/eks examples served on 80/443 — publishes host:8080 in its CapabilityStatement, ontoserver.fhir.base and canonical URLs. Nothing errors, so it is easy to miss.

Reverted to "", with serverPort: "8080" moved to where the intent belongs: explicit in examples/local/single-ro.yaml and single-rw.yaml, the two examples whose documented access path is a 8080:80 port-forward. The k3d examples already set it.

Reviewers note: installs on 0.4.0/0.4.1 that relied on the inherited default must now set serverPort: "8080" themselves. Called out in the README.

Verification

• helm lint clean, no warnings
• helm unittest: 271 tests pass, 5 new — podManagementPolicy override, exporter defaults, exporter overrides, and base URLs with and without serverPort
• helm template checked for the default render, OrderedReady, schema rejection of an invalid policy, and both edited examples

Chart.yaml version: deliberately untouched — Release Please owns it, and the feat commits should drive a minor bump. The hand-maintained artifacthub.io/changes list still describes the last released version and is worth updating when the release PR is cut.

🤖 Generated with Claude Code