Ansible forks, serial and Rolling Changes

Scale and Debugging / Guide 37 of 45 / Intermediate

• OSUbuntu 26.04 LTS
• ansible-core2.20.1
• Python3.14.4
• TimeAbout 15 min

serial accepts a number, a percentage (serial: 25%) or a list (serial: [1, 5, 10]) for a canary batch followed by progressively larger ones. Combined with max_fail_percentage it is how a large rollout is made safe - stop after the first batch if too much of it fails.

This guide includes

Use this when a change must not hit every host at once. This matters because forks and serial both control parallelism and are not the same thing - one is how many hosts run together, the other splits the play into batches.

• measuring forks, which is how many hosts run at once
• using serial to turn one play into batches
• using run_once and delegate_to for work that belongs on a single host
• throttling one task rather than the whole play

Before you start

• A control node configured as shown in the first step.
• The session runs several short playbooks and changes nothing on the managed nodes.

1. 1

   The configuration this guide assumes

2. 2

   forks is how many hosts at once

3. 3

   serial turns one play into batches

4. 4

   run_once and delegate_to

5. 5

   Throttling one task

Official sources

• Controlling playbook execution
• Rolling update batch size (serial)
• Delegation