XS⚠️ ◾ fix: liveness probe tcpSocket default to prevent Raft catch-up restart loops

[dimoschi]

$(cat <<'EOF'

Problem

Typesense returns 503 from /health when queued_writes exceeds --healthy-write-lag (default: 500). This threshold is always exceeded during Raft log catch-up after a pod restart or new node join.

With the previous default livenessProbe using httpGet /health, this caused a death spiral:

1. Pod restarts and joins the cluster
2. Raft catch-up causes queued_writes > 500 → /health returns 503
3. Liveness probe kills the pod after failureThreshold failures
4. Pod restarts and must catch-up from scratch again
5. Queue pressure from the churning node causes liveness failures on other pods too
6. Multiple pods restart simultaneously → quorum lost

The previous default failureThreshold: 2 / periodSeconds: 10 made this worse — pods were killed after just 20 seconds of catch-up.

Additionally, users who tried to switch to tcpSocket via HelmRelease values overrides would hit a Kubernetes validation error ("may not specify more than 1 handler type") because Helm merges the override with the chart default, producing both httpGet and tcpSocket in the rendered probe.

Changes

• Default livenessProbe switched to tcpSocket — checks only that the process is alive and listening, not queue depth
• New typesense.probe helper renders only the handler selected by a .type field (httpGet or tcpSocket), avoiding Helm map-merge from producing multiple handlers when users override the type
• startupProbe and readinessProbe remain on httpGet /health — correct behaviour, takes pod out of rotation while catching up without killing it
• Default CPU request set to 2000m per Typesense minimum requirements
• Sane probe thresholds: liveness failureThreshold: 6 / periodSeconds: 20, startup failureThreshold: 60 / periodSeconds: 10, readiness failureThreshold: 12 / periodSeconds: 10

Switching probe type

Users can now switch probe type cleanly without merge conflicts:

# Switch liveness to httpGet
livenessProbe:
  type: httpGet

# Switch readiness to tcpSocket
readinessProbe:
  type: tcpSocket

Test plan

• All 91 helm-unittest tests pass (helm unittest .)
• Verify rendered StatefulSet has only tcpSocket in livenessProbe
• Verify overriding livenessProbe.type: httpGet renders only httpGet
• Deploy to a test cluster and confirm no liveness-triggered restarts during Raft catch-up

Closes #4
EOF
)

[github-actions]

PR Metrics

✔ Thanks for keeping your pull request small.
⚠️ Consider adding additional tests.

	Lines
Product Code	69
Test Code	9
Subtotal	78
Ignored Code	44
Total	122

Metrics computed by PR Metrics. Add it to your Azure DevOps and GitHub PRs!