On a Tuesday, someone edits the env tag on an EC2 instance, from canary to prod. No deployment, no pull request, yet the machine’s role has just changed. Nobody checks that the service still meets its objectives. Pairing Site Reliability Guardian AWS events closes that gap: a tag change becomes the signal that starts a validation. The vendor describes the idea in a blog post; what follows is an operator’s reading, checked against the documentation on 10 November 2026.
I start from the decision, not the tool: which infrastructure changes deserve an automatic check, and which check? I did not replay the setup on an AWS account. Anything I could not confirm is marked « unverified ».
On the Dynatrace side, everything here applies to Dynatrace SaaS. I read no equivalent page for Managed, so do not assume it.
Site Reliability Guardian AWS: when should you validate?
Site Reliability Guardian is an app that compares indicators with thresholds. A « guardian » groups objectives, each computed by a DQL query. The documentation separates two families: Lifecycle guardians, built on software-lifecycle events, and Business guardians, built on business events. For an AWS tag, you pick the second: there is no deployment event, so you create a business event.
A naming check, as requested: the documentation page I read still says « Site Reliability Guardian » and « guardian ». I found no rename and no deprecation notice there. The source post does not mention one either. If your tenant shows something different, trust the interface.
To decide, ask three questions:
- Does the tag drive anything (routing, monitoring, autoscaling groups, backup policy)?
- Can a tag change degrade a service without any deployment tool noticing?
- Do health objectives already exist for that service? If not, start there, as in the release validation tutorial.
The event’s path from AWS to Dynatrace
The chain has four links. EventBridge receives the tag event, a rule filters it, an API destination posts it to Dynatrace, and Dynatrace stores it as a business event and triggers a workflow. Each link can break alone, so test each one alone.
Link 1: the EventBridge filter
AWS publishes tag changes under the source aws.tag, with the type Tag Change on Resource. The EventBridge documentation states best-effort delivery: expect that an event may be missed or arrive twice. The pattern below is limited to EC2 instances.
The source post filters on the same values. The changed-tag-keys and version fields, which would let you target a single tag key, come from my memory of the AWS documentation: unverified, so check them before use.
Link 2: dress the event as a CloudEvent
The business events ingest API accepts plain JSON or CloudEvents. For CloudEvents, four fields are mandatory: specversion, id, source and type. The type becomes event.type in Dynatrace, and the workflow will filter on it. An EventBridge input transformer builds this document.
This example is my own and I did not run it: unverified. The source post instead uses a transformer that passes detail, id, region, resources, source and time. Both approaches work as long as the mandatory fields are present. For the API destination’s header, the Dynatrace documentation lists application/cloudevent+json as an accepted type.
Link 3: authentication
The post creates an access token with the bizevents.ingest scope and stores it in the EventBridge connection (« API Key » mode), preferably through AWS Secrets Manager. The ingest page also mentions a platform token (openpipeline:bizevents:ingest) and an OAuth client (storage:events:write). It adds that the newer platform URL, on apps.dynatrace.com, accepts OAuth only. Which mode suits your tenant? Unverified: look at your environment URL before choosing.
Whatever the mode, apply least privilege: ingest scope only, a dedicated token, a planned rotation. For the rest of the SaaS API, see the API schemas.
How to check that it works
Change a tag on a test instance, then query business events in a Notebook. If the query returns nothing, the problem sits between AWS and Dynatrace; leave the guardian alone for now.
Once events arrive, move on to the guardian. For syntax, reread the DQL filter, summarize and timeseries commands.
Wire the guardian to this trigger
- Create a guardian for the service that depends on the tag. The post starts from the « Four Golden Signals » template; adapt the objectives to your service.
- Choose the Business type, then click Automate. Dynatrace generates a workflow whose trigger filters
bizevents. - Set the filter to
event.type == "ec2.tag.change". - Add a wait before the action. The post uses 600 seconds so the system can settle: an immediate check measures the old state, not the new one. Tune the delay to how fast your fleet converges.
- Change a tag and follow the workflow run, then the guardian result.
The validation action exposes guardian_id, validation_id, validation_url and validation_status. The last one takes info, pass, warning, fail or error, and the overall result follows the most severe objective. So chain a next step on fail: a notification, a ticket, or reverting the tag. For workflows in general, see the workflows module.
What can go wrong
- Bursts: ten instances retagged at once trigger ten validations. Plan a narrower filter or some grouping.
- Permissions: the workflow needs
automation:workflows:*rights, the guardian needsapp-settings:objects:*rights, and validation needs read rights on the queried data, plusstorage:events:writeto write its own business events. - Adaptive thresholds: they only take effect after at least five validations; until then the result stays « info ». A rarely changed tag will feed them slowly, so prefer static thresholds.
- Limits: 1,000 guardians and 50 objectives per guardian, according to the documentation.
- Best-effort delivery on the AWS side: do not make the guardian your only safeguard.
Habits worth adopting
- One event type per signal (
ec2.tag.change, not a catch-all), so filters stay readable. - A token dedicated to ingestion, kept in a vault.
- A trial on a disposable instance before production, one link at a time.
- Objectives tied to your SLOs; for alerting, burn-rate alerts complete the picture.
- A quarterly review: which tag changes actually failed a validation?
Frequently asked questions
Do I need an agent or a full AWS integration?
Not for this scenario. EventBridge only has to post the event to the business events ingest API. A CloudWatch integration remains useful for metrics; see the article on AWS and CloudWatch.
Does it work for resources other than EC2?
The post says the approach extends to other AWS resources. The exact list of types that emit a tag event is unverified: check the AWS documentation.
Is Site Reliability Guardian still current?
The documentation I read on 10 November 2026 uses it as the app name, with no mention of a rename or deprecation. Reread it before building a long project on it.
Conclusion
A tag is an operations decision, so it deserves a validation. Filter tightly in EventBridge, prove each link, let the system settle before judging, then hook an action to fail. Try it on a test instance first.
span.observer is an independent community blog, not affiliated with Dynatrace LLC (nor with Amazon Web Services). Trademarks belong to their owners. Tools change fast: check commands and options in the official documentation.
Originally published on span.observer. French version: https://span.observer/fr/site-reliability-guardian-aws-tags-ec2/.




