Managing implementation · Topic 1 of 3
Advising Teams on Deployment, Testing and Migration
Advising teams on deploying applications and infrastructure, deployment strategies, load, unit and integration testing, migration tools and Gemini Cloud Assist.
- 4 min
- Rolling, canary, blue-green, tagged revisions and feature flags each fit different needs.
- Infrastructure from code, previewed and applied by a pipeline or Infrastructure Manager.
- Test at the right level; load test in staging at multiples of the peak with synthetic data.
What it is
Architects advise the teams who implement the design. The advice covers how to deploy safely, how to test at the right levels, which tools move data and systems, and where AI assistance speeds the work. The exam presents a team's situation and asks for the method that meets its constraints.
How it works
Deployment strategies
| Strategy | How it works | Fits |
|---|---|---|
| Rolling | Replace instances gradually | Default for stateless apps with compatible versions |
| Canary | Send a small share of traffic to the new version, watch metrics, then expand | Risky changes where real traffic is the best test |
| Blue-green | Run the new version alongside the old, switch all traffic, keep the old for rollback | Fast rollback, extra capacity available |
| Tagged revision with no traffic (Cloud Run) | New revision gets its own URL; chosen testers use it; then move traffic | Letting specific users test against live data |
| Feature flags | Ship code dark, enable per user or group | Separating deployment from release |
Platform mechanics: Cloud Run traffic splits and revision tags; GKE rollouts, Gateway traffic splitting and Cloud Deploy canary; MIG canary rolling updates; Dataflow in-place update of a streaming job to keep window state.
Percentage traffic splits are random. When named people must see the new version and everyone else must not, use a tagged revision URL or a feature flag.
Infrastructure deployment
Deploy infrastructure from code (Terraform), reviewed in pull requests, previewed before apply and run by a pipeline or Infrastructure Manager. Keep state in a secure remote backend, and split it by environment and team so a change has a small blast radius.
Testing frameworks
| Level | Typical tools | Advice |
|---|---|---|
| Unit | The language's test framework | Fast, on every commit |
| Integration | Test containers, Google Cloud emulators, staging projects | Exercise real interfaces before release |
| Load | Distributed load generators (for example Locust or similar) on GKE or Compute Engine | Test at multiples of the expected peak, in a staging environment sized like production |
| Resilience | Fault injection, zone drains in staging | Prove recovery behaviour, not just speed |
Generate load from inside the required region when data rules apply, and use synthetic data rather than copying real customer data.
Data and system migration tooling
| Task | Tool |
|---|---|
| Discovery and cost assessment | Migration Center |
| VMs | Migrate to Virtual Machines |
| Databases | Database Migration Service |
| Continuous change capture | Datastream |
| Files and objects | Storage Transfer Service, Transfer Appliance |
| SaaS and cloud data into BigQuery | BigQuery Data Transfer Service |
| Mainframe data into BigQuery | Mainframe Connector |
Gemini Cloud Assist
Gemini Cloud Assist gives teams AI help across the Google Cloud lifecycle: designing application architectures, generating infrastructure configuration, explaining resources and errors, investigating incidents and suggesting cost and performance optimizations. Advise teams to treat its output as a draft: review generated configuration and changes through the normal pipeline.
A worked scenario
Three senior associates must try a new version of a Cloud Run review app against the live catalog while everyone else stays on the current version, with a one-step way back. Deploy the new revision with a tag and no traffic; the three use the tag URL; then move traffic. A 5 percent split would expose random users and not reliably reach the three.
How the exam asks about this
- "Open hourly windows must keep state, no lost or duplicate aggregates." Update the streaming Dataflow job in place.
- "Evidence that confirmations stay under two seconds at ten times peak, EU data rules." Synthetic load from workers in the EU against staging.
- "Mainframe copybook data into BigQuery without writing conversion code." Mainframe Connector.
What to remember for the exam
- Rolling, canary, blue-green, tagged revisions and feature flags each fit different needs.
- Infrastructure from code, previewed and applied by a pipeline or Infrastructure Manager.
- Test at the right level; load test in staging at multiples of the peak with synthetic data.
- Know the migration tool for each kind of data or system.
- Gemini Cloud Assist helps design, operate and optimize; review its output.
Load testing production that customers use around the clock risks the very outage you are trying to prevent.
Common mistakes
Draining a streaming Dataflow job to deploy a change. It closes open windows early, and the new job starts without their state.
Random traffic splits for targeted testers.
Copying real regulated data out of its region for testing.
Check yourself
How do chosen testers try a new Cloud Run revision while others stay on the current one?
Deploy it with a tag and no traffic, and give testers the tag URL.
Which tool moves VMs to Compute Engine?
Migrate to Virtual Machines.
How do you deploy a compatible change to a streaming Dataflow job without losing window state?
Use the update option to replace the running job with the same job name.
Topic 1 of 3 in Managing implementation
Test yourself on Managing implementation
The full Google Cloud Professional Cloud Architect question bank has 877 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Google Cloud. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.