Rollout strategies

This document describes the rollout strategies available in App Lifecycle Manager for managing how releases are deployed across units in your SaaS offering.

When you manage distributed SaaS applications, deploying updates simultaneously across all customer environments can lead to widespread outages if an unexpected regression occurs. Rollout strategies protect your service availability by controlling the pace, sequence, and spatial impact of software releases across logical and physical failure domains. Configuring rollout strategies in App Lifecycle Manager lets you automate gradual rollouts and introduce soak times between deployment waves. You can also define error budgets that pause problematic releases before they affect your customers.

Overview of rollout strategies

App Lifecycle Manager provides configurable rollout strategies to manage how units in your SaaS offering are updated. These rollout strategies follow Google Cloud's approach to change by applying consistent, staged deployment practices across your SaaS offering.

Use rollout strategies to minimize negative customer impact and isolate issues to individual logical and physical failure domains. You define your rollout strategy by creating a rollout kind that specifies one of the following strategies:

  • One location at a time (simple) (Google.Cloud.Simple.OneLocationAtATime): One location is updated at a time, with no wait between locations. Units are selected arbitrarily within a location, with a maximum of 5% of the location's units updated at a given time.

    This strategy is recommended for development environments and emergency scenarios.

  • All at once (simple) (Google.Cloud.Simple.AllAtOnce): All locations begin updating at the same time. Within each location, units are selected arbitrarily, with a maximum of 5% of the location's units updated at a given time.

    This strategy is recommended for development environments and emergency scenarios.

  • Progressive (gradual) (Google.Cloud.Progressive.Gradual.v1): Within each location, units are updated in percentage batches (1%, 10%, 20%, 30%, and ~40%) with soak times between batches. Across locations, the rollout progresses in exponential waves (for example, one location, then two, then four). Each wave is separated by a soak time, such as 18 hours. Units within a location are selected randomly.

    This strategy is designed for safe and predictable rollouts across multiple locations. It starts with a small footprint and gradually expands as confidence grows. This strategy is recommended for production environments with more than one location.

  • Progressive (single location) (Google.Cloud.Progressive.SingleLocation.v1): Units are updated in static percentage batches (for example, 1%, 10%, 20%, 30%, and ~40%). Extended soak times between batches (for example, 18 hours) provide time for issue detection and limit the adverse impact of rollout changes.

    This strategy is tailored for SaaS offerings operating in a single location, prioritizing safety and caution. This strategy is recommended for production environments with a single location.

Rollout strategy comparison and recommendations

The following table summarizes the key characteristics, soak time behaviors, and recommended use cases for each rollout strategy:

Rollout strategy Value Location behavior Concurrency and batching Soak time Recommended use case
One location at a time (simple) Google.Cloud.Simple.OneLocationAtATime Single location sequentially Up to 5% of units concurrently per location None Development environments and emergency rollouts
All at once (simple) Google.Cloud.Simple.AllAtOnce All locations concurrently Up to 5% of units concurrently per location None Development environments, non-production testing, and emergency rollouts
Progressive (gradual) Google.Cloud.Progressive.Gradual.v1 Multi-location in exponential waves (1, 2, 4 locations) Percentage batches (1%, 10%, 20%, 30%, and ~40%) Soak times between batches; 18 hours between location waves Production environments with multiple locations
Progressive (single location) Google.Cloud.Progressive.SingleLocation.v1 Single location Percentage batches (1%, 10%, 20%, 30%, and ~40%) Extended soak times (18 hours) between batches Production environments with a single location

Configure rollout kinds and safeguards

When you create a rollout kind, you associate a rollout strategy with a target UnitKind and can configure additional safeguards:

  • Error budget (error_budget): Automatically pauses a rollout when failed unit updates in a location exceed your configured threshold (allowed_count or allowed_percentage). If you don't configure an error budget, App Lifecycle Manager attempts to update all units regardless of the number of failures.

  • Unit filter (unit_filter): Applies a Common Expression Language (CEL) filter string to restrict the eligible population of units targeted by the rollout.

  • Regional scope: Supported rollout locations are determined by the available regions configured on your SaaS offering. For details on how regions affect rollout propagation, see App Lifecycle Manager regionalization.

What's next