Yesterday gave developers their workstations; today gives them their environments. The recurring ticket in every engineering organization, “I need a test environment for this feature”, either takes days through a queue or happens as ungoverned sprawl in a shared subscription. Azure Deployment Environments makes it a button with guardrails, and it is the piece that completes the platform engineering story this series has circled.
The Model: Definitions, Types, and the Button
The economics of ephemeral environments die on the word ephemeral being aspirational, and this is where the service earns its place over a folder of pipelines: environments carry expiration dates, extension is deliberate, and deletion actually deletes, the whole resource footprint, because the environment is tracked as a unit. Pair the default expiry per type (days for dev, a sprint for test) with the FinOps tagging stamped automatically by the deployment identity, and the CFO question “what are all these resources” gets the answer “feature environments, median lifetime four days, cost per environment on this dashboard”, which is a sentence platform teams historically could not say. The CI integration closes the loop the October posts opened: the pull request pipeline creates an environment from the definition, runs the test suite against real infrastructure, and destroys it on merge, infrastructure following the same lifecycle as the branches that need it, with the Azure CLI extension making it a pipeline step rather than a project.
The Platform Engineering Pattern Completed
Assembled with the rest of the series, the shape is now complete: the landing zone governs subscriptions, the vending pattern issues them, modules and template specs encode the how, Dev Box issues the workstation, Deployment Environments issues the infrastructure, and the developer’s path from idea to running code crosses zero tickets while crossing every guardrail. The adoption advice is incremental and familiar: start with the two definitions teams request most (the web app plus database, the event pipeline), one project, dev type only, measure request to ready time against the old ticket queue, and let demand pull the catalog wider. The platform team’s role shifts from fulfilling requests to publishing products, definitions with sensible defaults, secure baselines from the module library, and documentation in the manifest, and the measure of success inverts pleasingly: the better this works, the less anyone talks about environments at all, which is the correct fate for infrastructure. Sprawl, meanwhile, gets its obituary written by expiry dates rather than quarterly cleanup campaigns, and the Resource Graph orphan queries from October start returning empty, the quietest victory in the series.
Cheers
Osama
Expiry Is the Feature
The economics of ephemeral environments die on the word ephemeral being aspirational, and this is where the service earns its place over a folder of pipelines: environments carry expiration dates, extension is deliberate, and deletion actually deletes, the whole resource footprint, because the environment is tracked as a unit. Pair the default expiry per type (days for dev, a sprint for test) with the FinOps tagging stamped automatically by the deployment identity, and the CFO question “what are all these resources” gets the answer “feature environments, median lifetime four days, cost per environment on this dashboard”, which is a sentence platform teams historically could not say. The CI integration closes the loop the October posts opened: the pull request pipeline creates an environment from the definition, runs the test suite against real infrastructure, and destroys it on merge, infrastructure following the same lifecycle as the branches that need it, with the Azure CLI extension making it a pipeline step rather than a project.
The Platform Engineering Pattern Completed
Assembled with the rest of the series, the shape is now complete: the landing zone governs subscriptions, the vending pattern issues them, modules and template specs encode the how, Dev Box issues the workstation, Deployment Environments issues the infrastructure, and the developer’s path from idea to running code crosses zero tickets while crossing every guardrail. The adoption advice is incremental and familiar: start with the two definitions teams request most (the web app plus database, the event pipeline), one project, dev type only, measure request to ready time against the old ticket queue, and let demand pull the catalog wider. The platform team’s role shifts from fulfilling requests to publishing products, definitions with sensible defaults, secure baselines from the module library, and documentation in the manifest, and the measure of success inverts pleasingly: the better this works, the less anyone talks about environments at all, which is the correct fate for infrastructure. Sprawl, meanwhile, gets its obituary written by expiry dates rather than quarterly cleanup campaigns, and the Resource Graph orphan queries from October start returning empty, the quietest victory in the series.
Cheers
Osama
Expiry Is the Feature
The economics of ephemeral environments die on the word ephemeral being aspirational, and this is where the service earns its place over a folder of pipelines: environments carry expiration dates, extension is deliberate, and deletion actually deletes, the whole resource footprint, because the environment is tracked as a unit. Pair the default expiry per type (days for dev, a sprint for test) with the FinOps tagging stamped automatically by the deployment identity, and the CFO question “what are all these resources” gets the answer “feature environments, median lifetime four days, cost per environment on this dashboard”, which is a sentence platform teams historically could not say. The CI integration closes the loop the October posts opened: the pull request pipeline creates an environment from the definition, runs the test suite against real infrastructure, and destroys it on merge, infrastructure following the same lifecycle as the branches that need it, with the Azure CLI extension making it a pipeline step rather than a project.
The Platform Engineering Pattern Completed
Assembled with the rest of the series, the shape is now complete: the landing zone governs subscriptions, the vending pattern issues them, modules and template specs encode the how, Dev Box issues the workstation, Deployment Environments issues the infrastructure, and the developer’s path from idea to running code crosses zero tickets while crossing every guardrail. The adoption advice is incremental and familiar: start with the two definitions teams request most (the web app plus database, the event pipeline), one project, dev type only, measure request to ready time against the old ticket queue, and let demand pull the catalog wider. The platform team’s role shifts from fulfilling requests to publishing products, definitions with sensible defaults, secure baselines from the module library, and documentation in the manifest, and the measure of success inverts pleasingly: the better this works, the less anyone talks about environments at all, which is the correct fate for infrastructure. Sprawl, meanwhile, gets its obituary written by expiry dates rather than quarterly cleanup campaigns, and the Resource Graph orphan queries from October start returning empty, the quietest victory in the series.
Cheers
Osama
It shares the dev center spine with Dev Box: catalogs (Git repos) hold environment definitions, IaC templates, ARM and Bicep natively, Terraform and Pulumi through the extensibility model, with a manifest describing parameters, and the October template specs philosophy applies, the catalog is the menu, curated by the platform team, versioned in Git, reviewed like everything. Projects grant teams access to environment types, dev, test, staging, each mapped to a subscription and, critically, to its own deployment identity and RBAC: the environment type’s managed identity does the deploying with the permissions the platform team granted it, while the developer who pressed the button holds only the rights the type gives them on the result, which is how a developer deploys a full environment into a governed subscription without holding contributor anywhere. Azure Policy applies per environment type through the normal management group machinery, so dev environments get the permissive policy set and anything named production does not exist here at all.
Expiry Is the Feature
The economics of ephemeral environments die on the word ephemeral being aspirational, and this is where the service earns its place over a folder of pipelines: environments carry expiration dates, extension is deliberate, and deletion actually deletes, the whole resource footprint, because the environment is tracked as a unit. Pair the default expiry per type (days for dev, a sprint for test) with the FinOps tagging stamped automatically by the deployment identity, and the CFO question “what are all these resources” gets the answer “feature environments, median lifetime four days, cost per environment on this dashboard”, which is a sentence platform teams historically could not say. The CI integration closes the loop the October posts opened: the pull request pipeline creates an environment from the definition, runs the test suite against real infrastructure, and destroys it on merge, infrastructure following the same lifecycle as the branches that need it, with the Azure CLI extension making it a pipeline step rather than a project.
The Platform Engineering Pattern Completed
Assembled with the rest of the series, the shape is now complete: the landing zone governs subscriptions, the vending pattern issues them, modules and template specs encode the how, Dev Box issues the workstation, Deployment Environments issues the infrastructure, and the developer’s path from idea to running code crosses zero tickets while crossing every guardrail. The adoption advice is incremental and familiar: start with the two definitions teams request most (the web app plus database, the event pipeline), one project, dev type only, measure request to ready time against the old ticket queue, and let demand pull the catalog wider. The platform team’s role shifts from fulfilling requests to publishing products, definitions with sensible defaults, secure baselines from the module library, and documentation in the manifest, and the measure of success inverts pleasingly: the better this works, the less anyone talks about environments at all, which is the correct fate for infrastructure. Sprawl, meanwhile, gets its obituary written by expiry dates rather than quarterly cleanup campaigns, and the Resource Graph orphan queries from October start returning empty, the quietest victory in the series.
Cheers
Osama
Leave a Reply