Safe experimentation workflow

Create pipeline
Set matcher for ONE test collector (e.g., collector.ID=“test-collector-01”)
Activate and verify (no errors, acceptable performance)
Expand matcher to small group (e.g., env=“staging”)
Verify at scale
Expand to full target (e.g., env=“production”)

Why this is safe

Isolation + Fallback + Instant rollback + Staged rollout = Safe to experiment

You’re not gambling with production. You’re making controlled changes with multiple safety nets. Start small, verify, expand. If something goes wrong at any step, deactivate and try again. Recovery is fast and doesn’t require deployments or restarts.

Coming up next

You’ll categorize your collector fleet in the path. You’ll assign attributes like team, service.name, and env, the foundation that makes precise configuration and safe rollouts possible.

0:00

Read transcript

Audio transcript

Now that you understand the safeguards, let’s talk about the workflow that takes advantage of them. This is how you should approach any new pipeline.

Start by creating your pipeline and setting a matcher for just ONE test collector. Maybe use the collector ID directly, or create a test attribute. Activate the pipeline and verify it works. Check that there are no errors and that performance is acceptable.

Once you’re confident, expand the matcher to a small group. Maybe all staging collectors, or a specific team’s collectors. Verify again at this slightly larger scale.

Only then do you expand to your full target: production, all regions, whatever your goal is.

Isolation, fallback, instant rollback, and staged rollout add up to safe experimentation. You’re not gambling with production. You’re making controlled changes with multiple safety nets.

If something goes wrong at any step, deactivate and try again. That’s the power of this approach. Recovery is fast and doesn’t require deployments or restarts.

Before you create your first pipeline, though, you need collectors organized with the right attributes. That’s what you’ll do next.

Safe experimentation workflow

Why this is safe

Coming up next

Still have questions?

Get every update