fix(pdb): require minAvailable < replicas so PodDisruptionBudgets stay satisfiable

[monrax]

Summary

Both PodDisruptionBudgets could render as unsatisfiable on small clusters, silently producing disruptionsAllowed: 0 and blocking every voluntary eviction (kubectl drain, node cordon for maintenance, cluster-autoscaler scale-down). The templates rendered a PDB whenever the replica count met a fixed threshold (datanode >= 1, graylog >= 2) but never compared minAvailable to the actual replica count. This enforces the invariant minAvailable < replicas, matching the fail-fast idiom the MongoDB template already uses for arbiters.

Example of the broken case (DataNode, default minAvailable: 2):

datanode.replicas	minAvailable	disruptionsAllowed	Result
1	2	0	drains hang forever
2	2	0	drains hang forever
3	2	1	works

Details

• Replace the replica-threshold gate in both PDB templates with a fail when minAvailable >= replicas, mirroring custom/mongo-rs.yaml ("arbiters must be less than replicas"). The error names the misconfigured values.
• Gate the DataNode PDB on datanode.enabled (it previously rendered even when the DataNode was disabled).
• Align the DataNode minAvailable fallback (3 -> 2) with values.yaml, so nulling the value no longer silently raises the floor.
• Add helm-unittest coverage for both PDBs (no-render defaults, enabled/satisfiable case, and the fail-fast misconfig cases).

Behavior change: a PDB enabled with minAvailable >= replicas now fails at render/install time instead of installing a silently-broken budget — the same surface a bad arbiter count already has.

Linked issues

No linked issue — surfaced while auditing the chart against the 2025-12 third-party evaluation (which flagged the PDB gating direction; this fixes the underlying minAvailable-vs-replicas relationship the gate never validated).

PR Checklist

Please check the items that apply to your change.

• Tests added/updated
• Documentation updated
• This PR includes a new feature
• This PR includes a bugfix
• This PR includes a refactor

Testing Checklist

Static Validation

• Linter check passes: helm lint ./charts/graylog
• Helm renders local template sucessfully: helm template graylog ./charts/graylog --validate

Installation

• Fresh installation completes successfully: helm install graylog ./charts/graylog
• All pods reach Running state: kubectl rollout status statefulset/graylog
• Helm tests pass: helm test graylog

Functional (if applicable)

• Web UI accessible and login works
• DataNodes visible in System > Cluster Configuration
• Inputs can be created and receive data

Upgrade (if applicable)

• Upgrade from previous release succeeds
• Scaling up/down works correctly
• Configuration changes apply correctly

Specific to this PR

• helm unittest ./charts/graylog passes — 66 tests, incl. 10 new PDB cases
• PDBs render at defaults with satisfiable minAvailable (graylog 1/2 replicas, datanode 2/3 replicas)
• Enabling a PDB with minAvailable >= replicas fails with a clear message
• DataNode PDB no longer renders when datanode.enabled=false

Notes for reviewers

• Verify all applicable tests above pass
• Validate that the linked issues are no longer reproducible, if applicable
• Sync up with the author before merging
• The commit history should be preserved - use rebase-merge or standard merge options when applicable

🤖 Generated with Claude Code

[monrax]

Live validation on EKS

Validated on EKS (2× m6i.2xlarge, MongoDB Community operator), installed with --set provider=aws.

• Install with datanode.replicas=1 + datanode PDB enabled → fails fast: minAvailable (2) must be less than datanode.replicas (1)…
• Install with graylog.replicas=2, minAvailable=2 + graylog PDB enabled → fails fast: must be less than graylog.replicas (2)
• datanode.enabled=false + datanode PDB enabled → datanode PDB not rendered
• Default install, both PDBs enabled → both PDBs created with correct minAvailable and tightened selectors (app.kubernetes.io/name + /instance)
• Runtime status with pods ready → graylog-pdb and graylog-pdb-datanode both report ALLOWED DISRUPTIONS = 1
• Evict one datanode → permitted (201 Success)
• Evict a second datanode concurrently → blocked (429 TooManyRequests: … would violate the pod's disruption budget)
• Evicted pod reschedules and ALLOWED DISRUPTIONS returns to 1

[monrax]

Foot-gun live-validated on the main branch

Installed main with datanode.replicas=1 + datanode.podDisruptionBudget.enabled=true:

• Install succeeded silently with spec.minAvailable: 2 on a 1-replica DataNode
• Runtime: currentHealthy=1, desiredHealthy=2, disruptionsAllowed=0
• Evicting the DataNode (what kubectl drain issues) → 429 TooManyRequests — a node drain or autoscaler scale-down would hang indefinitely, with no warning at install