[RUN-4802] Add opt-in parallel VM querying to Azure node resource model

[carlosrfranco]

Jira Ticket

RUN-4802

Summary

Adds an opt-in queryNodeInstancesInParallel plugin property (default false) to the Azure Resource Model source, mirroring the AWS EC2 fix (RUN-3100 / rundeck-ec2-nodes-plugin#176). When enabled, AzureManager.listVms() maps VMs to nodes concurrently on a fixed thread pool instead of sequentially, cutting refresh time for accounts with many VMs. Sequential behavior is unchanged by default.

Note: implemented on top of the current Track 1 Azure SDK (main). PR #84 (RUN-4692) is an in-flight Track 2 SDK migration that rewrites AzureManager/AzureManagerBuilder/AzureNode; once that merges, this logic will need to be re-applied on the new API shapes.

./gradlew compileGroovy and ./gradlew test both pass.

[Copilot]

Pull request overview

Adds an opt-in plugin configuration (queryNodeInstancesInParallel, default false) to allow Azure VM→node mapping to run concurrently during resource model refresh, intended to reduce refresh time for subscriptions with many VMs while keeping default behavior unchanged.

Changes:

• Adds a new resource model source property queryNodeInstancesInParallel and wires it through AzureResourceModelSource → AzureManagerBuilder.
• Updates AzureManager.listVms() to optionally map VMs to AzureNode concurrently using an ExecutorService.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
src/main/groovy/com/rundeck/plugins/azure/plugin/AzureResourceModelSourceFactory.groovy	Adds the new opt-in configuration property definition.
src/main/groovy/com/rundeck/plugins/azure/plugin/AzureResourceModelSource.groovy	Reads the new property from configuration and passes it into AzureManagerBuilder.
src/main/groovy/com/rundeck/plugins/azure/azure/AzureManagerBuilder.groovy	Adds builder support for queryNodeInstancesInParallel and sets it on AzureManager.
src/main/groovy/com/rundeck/plugins/azure/azure/AzureManager.groovy	Implements parallel VM→node mapping when the opt-in flag is enabled.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[jtobard]

Hey @carlosrfranco — reviewed this against RUN-4802. The approach looks right
and lines up well with the per-account latency described in the ticket. A
couple of things to address before merging, both raised by Copilot and
confirmed valid:

1. Bound the thread pool and add a timeout.
   Executors.newFixedThreadPool(list.size()) in AzureManager.listVms() sizes
   the pool to VM count with no cap, and future.get() has no timeout. Unlike
   the AWS precedent (PR #176), which bounds the pool by endpoint/region count
   (small and stable), this one scales with VM count — which is exactly the
   axis this flag is meant to help on large subscriptions. Worth capping it
   (e.g. Math.min(list.size(), N) with a small fixed max) and adding a timeout
   to invokeAll/future.get() so a hung SDK call can't block the refresh
   indefinitely.
2. Add test coverage for the new concurrency path. Right now there's no test
   exercising queryNodeInstancesInParallel end to end — not the config →
   builder wiring, and not the parallel branch in listVms() itself (executor
   creation, exception translation, shutdown/timeout handling). Since this is
   new concurrency logic, it'd be good to have at least one spec covering the
   parallel path before it ships.

Everything else checks out — pulled the branch and ran the tests locally,
build is green (./gradlew test passes), consistent with what you noted in the
PR description.

On the failing Snyk check: that's safe to ignore for this PR — it's a
pre-existing high-severity finding in spring-security-crypto pulled in
transitively via rundeck-core@6.1.0, unrelated to this change (would fail on
main today too, and there's no upstream fix yet).

[jtobard]

Re-reviewed after b9d4500. Both blockers from the previous pass are resolved:

• Thread pool is now bounded (min(list.size(), 10)) with a 120s timeout on invokeAll, and a timed-out future now raises a clear ResourceModelSourceException instead of hanging or leaking a bare CancellationException.
• queryNodeInstancesInParallel config wiring now has dedicated test coverage via the extracted createManager() method (true/false + default-unset cases).

Snyk is still red, but it's now flagging license-policy issues (EPL/CDDL on junit/spock/azure-sdk transitive deps) with no build.gradle changes in this PR — looks like a policy/DB drift between scans, unrelated to this change. Safe to ignore for merge purposes.

Approving.