preStop Kubernetes Lifecycle Resque Hook

kube-resque-pre-stop.sh

#!/bin/bash

# This script gracefully stops resque workers by issuing the USR2 
# signal to the resque-pool manager and then waits for the workers 
# to be paused before exiting. Resque-pool master process relays 
# the USR2 signal to the children.
# 
# For reference: 
# (1) https://github.com/nevans/resque-pool#signals
# (2) https://github.com/resque/resque/blob/master/lib/resque/worker.rb#L376-L378
#
# This script can serve as a preStop lifecycle hook in Kubernetes
#
# Written by: Anton Ouzounov <[email protected]>
#
# Version 1.0 - June 14th, 2017
# Version 1.1 - October 20, 2017

kill -USR2 1;
#if resque workers present, wait 10 seconds and check again
numWorkers=$(ps aux | grep -i '[p]rocessing' | wc -l);

while true; do
  if [ "$numWorkers" -eq "0" ]; then
    #exit when all resque workers are terminated
    echo "Time to terminate this pod";
    break;
  else
    echo "Not ready to exit, workers are processing, more sleepy sleepy.";
    echo "$numWorkers -- number of workers processing.";
    sleep 5;
    kill -USR2 1;
    numWorkers=$(ps aux | grep -i '[p]rocessing' | wc -l);
  fi

done;

echo "Exiting from the bash script";

Thanks!! I was looking a simple way to do this, found some much longer script that turned me off on the whole process. The "new" way of signal handling (using TERM_CHILD=1 in Resque > 1.22) was not acceptable to me ... it interrupts immediately and expects you to handle the exception and re-queue the job. They offer a grace period, but that is only to specify how long you're willing to wait for cleanup.

@sunild , for resque-pool master with workers in a k8s pod, you can specify pod.Spec.TerminationGracePeriodSeconds and tune this value in the spec section of the Kubernetes pod's manifest if you need a grace period.

Here is the documentation: https://kubernetes.io/docs/tasks/administer-cluster/out-of-resource/#soft-eviction-thresholds