Professional Cloud DevOps Engineer · Domain 1: Establishing and bootstrapping foundational infrastructure and organizational governance · 18% of the exam
Task 1.5: Managing multiple cloud development environments.
Cloud Workstations and Cloud Shell, bootstrapping them with custom images, IDEs and the Google Cloud SDK, AI help for development and operations, and configuring MCP servers.
Study it
Cloud development environments: Cloud Workstations, Cloud Shell, AI assistance and MCP servers
Try each one before opening the answer. Every option is explained, with the Google Cloud documentation page that proves it.
Question 1 · choose 1
Eighty developers use Cloud Workstations created from one configuration that uses a predefined Code OSS image. After every restart they reinstall the same Terraform version, kubectl plugins and in-house linters. The platform team wants every workstation to start with an identical, centrally maintained toolset that is checked for vulnerabilities and kept up to date as Google updates its images. What should it do?
AA custom container image built on top of the predefined image by a pipeline, used by the workstation configuration
BAsk each developer to install the tools once into the persistent home directory, where they survive restarts
CGive each developer a customize_environment script that the predefined image runs to install the tools at every start
DRaise the idle timeout in the workstation configuration so that workstations stop less often and keep their tools
Show the answer and why
AA custom container image built on top of the predefined image by a pipeline, used by the workstation configuration
Correct
Custom images extend the base images to pre-install tools; Google recommends a pipeline that rebuilds them when the base image is updated and a scanner such as Artifact Analysis for the added dependencies.
BAsk each developer to install the tools once into the persistent home directory, where they survive restarts
Incorrect
The home directory does persist, but each developer then owns a separate copy, so the toolset is neither identical nor centrally maintained and scanned.
CGive each developer a customize_environment script that the predefined image runs to install the tools at every start
Incorrect
The base image runs that per-user script at startup, so each start still downloads and installs the tools, and nothing scans or controls each user's copy centrally.
DRaise the idle timeout in the workstation configuration so that workstations stop less often and keep their tools
Incorrect
Workstations run on ephemeral VMs that are deleted when they stop, so a longer timeout only delays the loss and leaves the toolset unmanaged.
Predefined Cloud Workstations images hold only a minimal environment. A custom container image that extends the base image pre-installs the team's tools for every workstation created from the configuration; the team owns that image, so it rebuilds it when the base image changes and scans it.
An engineer installs a command-line tool in Cloud Shell with sudo apt-get install. The next morning the tool is gone, although the scripts in the home directory are still there. The engineer wants the tool in every new session without running a separate VM. What should the engineer do?
ASwitch Cloud Shell to Ephemeral mode, which keeps the session's installed packages and starts faster
BExpand the Cloud Shell persistent disk beyond 5 GB, so that the system directories have room to persist
CPut the tool's files under $HOME, such as in ~/bin, and add that path to PATH in the .bashrc file
DKeep the session open overnight with a loop, so that the VM is never terminated for inactivity
Show the answer and why
ASwitch Cloud Shell to Ephemeral mode, which keeps the session's installed packages and starts faster
Incorrect
Ephemeral mode runs Cloud Shell without persistent disk storage, so nothing survives the session, not even the home directory.
BExpand the Cloud Shell persistent disk beyond 5 GB, so that the system directories have room to persist
Incorrect
Cloud Shell persistent disk storage cannot be expanded, and only $HOME is persistent whatever its size.
CPut the tool's files under $HOME, such as in ~/bin, and add that path to PATH in the .bashrc file
Correct
Only the 5 GB $HOME directory persists between sessions; changes outside it are lost when the ephemeral VM is terminated.
DKeep the session open overnight with a loop, so that the VM is never terminated for inactivity
Incorrect
Sessions are capped at 12 hours and end after inactivity, so the VM and everything outside $HOME are discarded anyway.
Cloud Shell runs on an ephemeral VM per session, with 5 GB of persistent disk mounted as $HOME. Software installed outside $HOME does not survive the end of the session; install it in the home directory instead. If Cloud Shell is not used for 120 days, the home directory itself is deleted.
Security requires that when a developer leaves and the developer's Cloud Workstations workstation is deleted, administrators can still audit the content of that workstation's home directory before they dispose of it. Which setting in the workstation configuration meets this need?
AKeep the default reclaim policy, and re-create a workstation with the same name when an audit is needed
BSet the reclaim policy of the persistent home directory to Retain instead of the default, Delete
CCreate the persistent directory from a source disk snapshot taken from a template workstation
DEnable quick start workstations, so that a pool of pre-started VMs keeps the last session's disk
Show the answer and why
AKeep the default reclaim policy, and re-create a workstation with the same name when an audit is needed
Incorrect
With the default Delete policy the disk is removed with the workstation, and a re-created workstation gets a new, empty disk.
BSet the reclaim policy of the persistent home directory to Retain instead of the default, Delete
Correct
Retain keeps the persistent disk when a workstation is deleted, so administrators can attach it to a VM to audit it and then delete it by hand.
CCreate the persistent directory from a source disk snapshot taken from a template workstation
Incorrect
A source snapshot only sets the initial content of new home disks; it does not keep a disk after its workstation is deleted.
DEnable quick start workstations, so that a pool of pre-started VMs keeps the last session's disk
Incorrect
Quick start keeps pre-started VMs for faster startup; it has nothing to do with keeping a deleted workstation's home disk.
A workstation configuration sets the persistent directory for home directories, including its reclaim policy. The default, Delete, removes the disk with the workstation. Retain keeps it for auditing; administrators then attach the disk to a Compute Engine VM to read it and delete it manually.
An on-call engineer configures an AI coding assistant to use the Cloud Logging remote MCP server at https://logging.googleapis.com/mcp with Application Default Credentials. The Cloud Logging API is enabled and the engineer holds Logs Viewer on the project, yet every tool call is rejected with a permission error. What is missing?
AThe Service Usage Admin role, so that the assistant can enable the MCP server
BA broader logging role such as Logging Admin, so that the MCP tools can read every log bucket
CAn OAuth 2.0 client ID and secret created in the project, used in place of Application Default Credentials
DThe MCP Tool User role (roles/mcp.toolUser), granted on the project to the caller's identity
Show the answer and why
AThe Service Usage Admin role, so that the assistant can enable the MCP server
Incorrect
Service Usage Admin is needed to enable APIs, and the product API is already enabled; it does not grant permission to call MCP tools.
BA broader logging role such as Logging Admin, so that the MCP tools can read every log bucket
Incorrect
Product roles control what the tools may do on resources, but making any tool call to a Google Cloud MCP server also needs the mcp.tools.call permission.
CAn OAuth 2.0 client ID and secret created in the project, used in place of Application Default Credentials
Incorrect
ADC is a supported way to authenticate to Google Cloud MCP servers; changing the credential type does not add a missing permission.
DThe MCP Tool User role (roles/mcp.toolUser), granted on the project to the caller's identity
Correct
Making tool calls to Google Cloud MCP tools requires MCP Tool User, which contains the mcp.tools.call permission, in addition to the product roles.
To use a Google Cloud remote MCP server, enable the product's API in the project and give the caller the MCP Tool User role, whose mcp.tools.call permission allows tool calls. IAM roles for the product still decide what the tools can do on its resources. Non-Google Cloud MCP servers do not use this role.
A team's agent is connected to a Google Cloud remote MCP server that exposes dozens of tools, and since then the agent has become slow, makes confused tool choices and costs more to run. It only needs a handful of read-only tools that the server offers as one of its documented groups. What should the team change in the agent's MCP configuration?
APoint the configuration at that group's toolset endpoint instead of the server endpoint
BReplace the remote server with a local MCP server that talks to the agent over stdio
CAdd system instructions telling the agent to ignore every tool outside the read-only group it needs
DPoint the configuration at the server's regional endpoint instead of its global endpoint
Show the answer and why
APoint the configuration at that group's toolset endpoint instead of the server endpoint
Correct
A toolset is a logical group of tools with its own HTTP endpoint that acts as a virtual MCP server, so the agent loads only those tools.
BReplace the remote server with a local MCP server that talks to the agent over stdio
Incorrect
The transport changes, but the agent would still load every tool the server exposes into its context.
CAdd system instructions telling the agent to ignore every tool outside the read-only group it needs
Incorrect
The server's tools are still discovered and loaded into the agent's context, which is what makes it slow, confused and expensive.
DPoint the configuration at the server's regional endpoint instead of its global endpoint
Incorrect
Regional endpoints serve data residency needs for in-use data; they expose the same tools.
Loading too many tools makes agents slow, confused and expensive. Some Google and Google Cloud MCP servers offer toolsets, logical groups of tools that each have their own endpoint and work as a virtual MCP server. Configure the toolset endpoint the same way as a server endpoint.