A useful Node.js API health setup answers two separate questions: should this instance receive traffic, and is its process still able to run? For a deployment, monitor those probe results alongside API error rate and latency. Treat the four signals as evidence to pause or investigate a rollout—not as a universal automatic rollback formula. Choose observation windows and thresholds from your service’s baseline; Kubernetes and AWS do not prescribe universal limits for these signals.
How do I add health checks to a Node.js API?
Expose lightweight endpoints for the orchestration platform or load balancer to check, and configure each check for a distinct purpose. In Kubernetes, probes can use HTTP, TCP, or other supported mechanisms; the configured probe determines whether a check succeeds. An HTTP endpoint should return a success status only when its defined condition is met.
As an Amazon Associate I earn from qualifying purchases.
One possible application-level arrangement is /health/live for process responsiveness and /health/ready for traffic eligibility. Those paths are examples, not required Node.js or Kubernetes routes. Configure the actual paths in the workload manifest. Keep each handler fast and avoid making a health request do expensive work.
Readiness: can this instance serve traffic?
A readiness check indicates whether an instance should receive traffic. When it fails, Kubernetes can mark the pod unready so it is excluded from ready service endpoints while the problem persists. That is traffic removal, not a reason by itself to restart the process. Readiness is useful during startup, when the instance is overloaded, or when it cannot safely serve requests. See Kubernetes’ probe configuration guide and Amazon EKS application guidance.
#1 Best Overall
Liveness: should this process continue running?
A liveness check is intended to identify a process that is stuck or unable to make progress, so the orchestrator can restart it. It should generally verify that the application can respond, not that every downstream service is healthy. AWS warns against making liveness depend on an external factor such as a database: a database outage could otherwise trigger repeated restarts without fixing the cause. The distinction between liveness and readiness is described in AWS probe and load-balancer guidance.
Startup: has initialization finished?
For an application with slow or unpredictable initialization, a startup probe can give it time to start before liveness and readiness checks begin. This avoids treating normal startup as a failure. Kubernetes explains how startup probes work in its probe documentation.
Should a health check test the database?
It depends on what the endpoint is meant to decide. A dependency-aware readiness check can prevent an instance that cannot serve safely from receiving traffic. But if every replica fails readiness because they all depend on the same unavailable database, the service can lose all ready capacity at once. Choose deliberately whether a particular dependency failure should remove one instance from traffic; do not make liveness fail merely because a shared dependency is down. Consider reporting dependency status separately for diagnosis rather than turning every dependency check into a restart trigger.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What are the four deployment signals to monitor?
Use these signals together during a defined observation window after a release. Their role is to expose possible regressions and help decide whether to pause a rollout, investigate, or roll back. Error-rate and latency signals are operational recommendations; the probe sources define readiness and liveness behavior but do not supply universal rollback thresholds.
| Signal | What to watch | What it may indicate | Likely first response |
|---|---|---|---|
| API error rate | A sustained increase relative to the pre-deployment baseline and expected traffic profile. | The new revision may be returning more failed requests. | Compare by revision and investigate errors before deciding whether to pause or roll back. |
| API latency | A sustained regression against the service’s usual latency distribution or objective. | The release may have slowed user-facing requests. | Check the trend and affected revision; do not decide from one slow request. |
| Readiness or ready capacity | Readiness failures, fewer ready instances, or failures concentrated in the new revision. | Instances may not be able to serve traffic reliably. | Pause or investigate the rollout; readiness failure removes traffic eligibility rather than prescribing a restart. |
| Liveness failures or restarts | Increasing liveness failures or restarts, especially on the new revision. | Processes may be stuck or unable to make progress; a shared dependency or infrastructure problem may also be involved. | Inspect application and cluster context before attributing the failures to the release. |
No universal error-rate percentage, latency limit, ready-capacity minimum, or restart count applies to every API. Define thresholds using the service baseline, user-facing objective, rollout window, and acceptable impact. A combination of worsening signals that starts with the new revision is stronger evidence for rollback than one failed dependency check, but it is still a diagnosis—not a rule prescribed by Kubernetes or AWS.
When should I roll back a deployment?
Consider rollback when the new revision is associated with sustained user-facing degradation or loss of serving capacity, and the evidence points to the release rather than a shared dependency or cluster incident. If only one signal changes, first determine whether its scope and timing match the rollout. A database or network outage can cause readiness failures across healthy revisions; restarting all affected processes may make recovery harder.
- Set the comparison window. Compare the release period with a pre-deployment baseline, accounting for expected traffic changes.
- Break down results by revision. Check whether error rate, latency, readiness failures, or restarts are concentrated in the new version or affect all instances.
- Assess available serving capacity. Look at how many instances remain ready and whether the remaining capacity can safely handle traffic.
- Check for shared causes. Correlate probe failures with application logs, dependency health, and node or cluster conditions before treating every failing instance as a bad release.
- Pause or roll back according to policy. If the evidence indicates a release regression and impact is unacceptable, halt further rollout and use your established rollback procedure. If evidence points to a shared incident, address that incident rather than blindly restarting or reverting healthy application code.
Kubernetes node conditions and events provide useful context when separating application problems from infrastructure issues. Node status includes readiness and resource-pressure conditions; Amazon EKS documents node monitoring events and automatic repair behavior in its node monitoring agent guidance, node health and repair guidance, and Kubernetes’ Node Status reference. EKS automatic repair responds to specified node conditions, not every resource-pressure condition. Node repair is distinct from rolling back an application release.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteHow should probes be configured to avoid false alarms?
- Allow for normal startup. Use a startup probe where initialization is long or unpredictable, and avoid settings that classify expected startup time as failure.
- Do not make thresholds too aggressive. Kubernetes applies configured failure thresholds. Brief load spikes or transient delays should not automatically look like a dead process.
- Keep checks lightweight. Frequent probes and exec-based checks can add CPU overhead, particularly at high pod density. Kubernetes discusses probe configuration and its trade-offs in its probe reference.
- Understand the routing layer. Kubernetes readiness, a load balancer’s health check, and a process restart are distinct actions. Verify how the deployment platform routes traffic and how its health checks interact with pod readiness; AWS discusses both probes and load-balancer health checks in its EKS guidance.
Kubernetes API-server health endpoints are not the same as endpoints for your Node.js application. For the API server, /livez and /readyz have distinct meanings, while its older /healthz endpoint is deprecated; see the Kubernetes API health checks reference.
What can implement Node.js health endpoints?
You can implement the handlers directly in your application or use a library that supports the lifecycle behaviors you need. The npm package Lightship describes readiness, liveness, startup checks, and graceful shutdown for Node.js services running in Kubernetes. It is one implementation option; whichever approach you choose, map each check to the correct action and verify its behavior under startup, dependency failure, and shutdown conditions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
