envoygrace

Graceful shutdown for Envoy running in a Sidecar model.

The problem today with Envoy running as a sidecar next to your application is that when Kubernetes goes and sends a SIGTERM to Envoy, it doesn't gracefully drain the active connections automatically, leading to errors and not a pleasant user experience. There are many Github issues and articles that say to be able to gracefully shutdown envoy you need to send a POST /healthcheck/fail request to it's admin interface - that way it'll stop accepting new connections and drain existing connections.

Some links to follow for more information:

• envoyproxy/envoy#2920 (comment)
• https://www.envoyproxy.io/docs/envoy/latest/operations/admin#operations-admin-interface-healthcheck-fail
• envoyproxy/envoy#1990 (comment)

This package is an alternative to doing the follow in your K8s prestop hook:

• curl -XPOST localhost:8001/healthcheck/fail
• sleep 5

Why use this package, when I can just use cURL?

• You don't want to install cURL in your final Docker image (but are willing to put this binary in it).
• Future versions of envoygrace can add additional functionality that you would otherwise need to write yourself.

Installation

Download a version stamped binary from the Releases page into your application's Dockerfile. For example:

ARG ENVOYGRACE_VERSION=v1.0.0

RUN wget -qO/bin/envoygrace https://github.com/tunein/envoygrace/releases/download/${ENVOYGRACE_VERSION}/envoygrace-linux-amd64 && \
    chmod +x /bin/envoygrace

Usage

Configure the binary to run on a preStop hook in your Kubernetes deployment.

lifecycle:
  preStop:
    exec:
      command:
        - "/bin/envoygrace -envoy-url localhost:8001 -eviction-period 5000 -envoy-timeout 500 -pre-sleep 5000"

MAKE SURE YOU SET A PRESTOP HOOK ON YOUR ENVOY CONTAINER WITH SLEEP THAT'S LONGER THAN THE ABOVE CONFIG

Command Line Options

  --help
      shows this help output
  -envoy-timeout int
      envoy request timeout in milliseconds (default 1000)
  -envoy-url string
      the base url where envoy lives (default "localhost:8001")
  -eviction-period int
      amount of milliseconds to sleep after calling envoy graceful commands (default 5000)
  -pre-sleep int
      how long to sleep before sending requests to envoy in milliseconds (default 5000)
  -hard-fail
      if this flag is specified and the call to envoy fails, the script will exit with code 1, otherwise code 0