xiangyanw opened a new issue, #1011:
URL: https://github.com/apache/apisix-helm-chart/issues/1011
The APISIX Helm chart hardcodes this lifecycle hook in the Deployment:
```yaml
lifecycle:
preStop:
exec:
command:
- /bin/sh
- -c
- sleep 30
```
However, the chart does not expose
spec.template.spec.terminationGracePeriodSeconds.
Kubernetes starts the termination grace-period timer before executing
preStop. When the Pod uses Kubernetes' default terminationGracePeriodSeconds:
30, the hardcoded sleep 30 consumes the entire grace period. APISIX then
receives SIGTERM with effectively no time left to complete its Nginx graceful
shutdown.
This conflicts with the APISIX worker_shutdown_timeout setting. For example,
the chart can render:
```yaml
apisix:
nginx:
workerShutdownTimeout: 300s
```
but the Pod may still be force-killed around 30 seconds after termination
starts.
This is especially problematic behind an ALB/NLB target group with a
300-second deregistration delay. Existing in-flight requests or long-lived
connections can be terminated before load balancer draining completes,
resulting in client-side errors.
Environment
- APISIX Helm chart: 2.17.0
- APISIX image: 3.18.0-ubuntu
- Kubernetes deployment mode
- Target group deregistration delay: 300s
Expected behavior
The chart should expose Pod termination settings through values, so users
can align:
1. load balancer deregistration delay;
2. preStop delay;
3. APISIX/Nginx worker_shutdown_timeout;
4. Kubernetes terminationGracePeriodSeconds.
For example:
```
terminationGracePeriodSeconds: 360
lifecycle:
preStop:
exec:
command:
- /bin/sh
- -c
- sleep 30
apisix:
nginx:
workerShutdownTimeout: 300s
```
A grace period of 360 seconds provides:
- 30 seconds for endpoint/LB deregistration propagation;
- up to 300 seconds for APISIX worker draining;
- 30 seconds of kubelet buffer.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
