Use tasks to stop a thundering herd in Puppet Enterprise

You can use a task to stop a thundering herd on Unix (including MacOS) or Windows nodes. On each targeted agent node, the task stops puppet agent service and restarts it, delaying check-in for anywhere from one second up to the configured runinterval.

The task might take up to the time of the runinterval to complete. While the task runs, each node will be offline for 10 to 15 seconds and the console will be unresponsive.

Version and installation information

PE version: All supported versions

Solution

Install the puppetlabs-support_tasks module and use a task to stop a thundering herd on Unix and Windows nodes.

1. Download and install the puppetlabs-support_tasks module which includes the task for this solution. Please use the most updated version of the module.

2. Run the task on all of your nodes or a subset of your nodes by following the instructions in our documentation for targeting and using tasks in the console. Make sure you’re using the right version of the documentation for your deployment.

   1. Select the task support_tasks::st0346_herd_resolver.

   2. Run the task on the targeted nodes.

Address future thundering herds

This is a short-term solution. To address future thundering herds, choose only one of the following solutions.

• Mitigate a thundering herd: Use max-queued-requests

• Spread out agent catalog requests using splay