express route that indicates whether a service is ready or not. Mostly created to make graceful restart of node express servers in a Kubernetes environment.
Graceful restart of a nodejs express server in Kubernetes
I intially thought that catching SIGTERM and waiting for server.close()` to finish would be enough to do a graceful restart of a nodejs service. I was wrong.
The reliable way of handling a graceful restart is to use the readynessProbe functionality either with a pre-stop hook or when catching the SIGTERM signal. The readynessProbes are used by Kubernetes to know if a service is ready and can receive traffic, in its http form Kubernetes checks for the responseCode, anything above 399 is considered not ready.
When a service receives a stop signal (SIGTERM), it should respond with a 500 responsecode when probed for readiness, this will ensure Kubernetes does not send load to it anymore. Once that's done,
server.close() can be called to ensure ongoing connections are terminated gracefully.
Note that by default Kubernetes will send a SIGKILL 30s after the SIGTERM if the service did not terminate, this timer is configurable in the manifest.
Example of a graceful node http server for Kubernetes
'use strict';var http = ;var express = ;var ready = ;var app = ;var server = http;app;/*** adds a `Connection: close` to all responses stopping.*/app;server;//add graceful shutdownprocess;
Another use of the readinessProbe can be to indicate if the server is too busy, connect-ready can use the toobusy-js module to indicate whether the server is too busy and deflect load to another pod.
The toobusy-js module should be installed to use this functionality, it is not bundled in connect-ready.
- npm install toobusy-js
- enable toobusy in connect-ready: