Measured on real playbooks.

Two workloads, each run by a recipe in the repository, against ansible-core 2.19.12 with pipelining turned on. Every figure on this page comes from those recipes, measured on 2026-09-27.

50×faster on four galaxy roles
2.15×faster on k3s-ansible site.yml
Four Galaxy rolesgeerlingguy.security, nginx, git and pip on two converged Ubuntu 24.04 hosts, 56 tasks per host
0 s5 s10 s15 s20 s25 s30 s
ansible-core 26.435 sVolant 0.529 s

Move over the trace to read it.

k3s-ansible site.ymlthe official k3s-ansible playbook against an installed cluster, one server and one agent
0 s10 s20 s30 s40 s50 s
ansible-core 50.519 sVolant 23.477 s

Move over the trace to read it.

Median of three timed passes on 2026-09-27, against ansible-core 2.19.12 with pipelining = True. A line is high while that engine’s run is going.

The ratios are far apart for a reason. The roles change nothing on a converged host, so almost all of the reference’s time goes to the machinery around each module. The k3s playbook restarts services and runs install scripts on every pass, and that work takes as long under either engine.

Where Volant spends its time.

The median pass, split by step. A native step is answered in Rust by the agent on the host.

Four Galaxy roles0.529 s for two hosts
Opening the connections172.6 ms
file ×8, lineinfile ×8, stat ×5native52.8 ms
systemd, 3 tasksnative52.4 ms
setup, gathering factsnative47.6 ms
apt, 8 tasksnative38.4 ms
Frames and blobs on the wire27.1 ms
Preparing tasks on the controller25 ms
Building the Python module setcached15.3 ms
Waiting for the other host10.6 ms
k3s-ansible, the server23.477 s for two hosts
Restarting k3s, server and agentnative systemd13.03 s
service_factsnative2.48 s
apt with update_cachePython2.21 s
Opening the connections350 ms
Waiting for the other host253 ms
Preparing tasks on the controller185 ms
package_factsnative168 ms
Gathering factsnative55 ms
Building the Python module setcached30.6 ms

On k3s, most of the 23.5 seconds is work the playbook asks for: restarting the k3s service, running its install script and 13 command tasks. Both engines pay for those the same way, because they are processes on the host. What is left explains the rest.

The configuration.

Default settings on both sides, except pipelining, which makes the reference faster.

Managed hosts
two Ubuntu 24.04 hosts, 4 cores each
Controller
Ubuntu 26.04, 8 cores
Volant
release build, default settings: native modules on, one SSH connection per host
Reference
ansible-core 2.19.12 with pipelining = True, forks at its default of 5
Passes
one warm-up pass per engine, then three timed passes each; the median is reported
Check
every pass must end on the same PLAY RECAP under both engines, or the recipe fails

Measure your own run.

On your own playbook, --profile prints the split above for your run, step by step.

To reproduce the two workloads, the recipes build Volant in release mode, run against the hosts you name and print the medians and the ratio. They need ansible-core 2.19.12 installed with uv tool install, two hosts where the account has sudo without a password, and they change those hosts, so use machines you can wipe. The recipes

# Your own run, split by step 
volant playbook -i inventory.ini site.yml --profile

# The two workloads of this page 
VOLANT_TARGET_HOST=host1 VOLANT_SECOND_HOST=host2 just bench 3 
VOLANT_TARGET_HOST=host1 VOLANT_SECOND_HOST=host2 just bench-k3s 3

These figures describe two playbooks on two hosts. Yours depend on your network, on what your tasks do, on how many of them a native module covers and on how many hosts you drive at once. The full measurement