Using the published Workload Identity Federation model, I added a viewer service account, log and monitoring viewer bindings, a pool whose AWS provider trusts only an account and role supplied later, and impersonation rights for that identity. The official attribute mapping needed extra care with quotes in the configuration language. Schema checks passed. I did not apply the change or exchange a token.
What worked
The pool, provider, attribute condition, and impersonation binding lined up with a read-only investigator and avoided creating a long-lived service-account key.
A Workload Identity Federation and IAM configuration was created with logging and monitoring viewer roles only, deliberately excluding deployment and data-plane write permissions. It validated locally but could not be applied without the external identity and production project values.
What worked
The role model supported a narrow, auditable separation between investigation access and code-remediation permissions.
What got in the way
Federation setup depends on vendor account identity details that were correctly absent from source control.
Selected predefined viewer roles for logging, monitoring, Cloud Run, Cloud Build and Pub/Sub to give an external investigator read-only access without database or Firestore reach, and chose to keep the service account key out of Terraform state.
What worked
Predefined viewer roles map cleanly onto a read-only investigation scope, so the least-privilege identity was a short list of bindings.
What got in the way
Key-based auth for a third party is still the documented path, which means a long-lived credential to rotate; workload identity federation would be preferable if the vendor supported it.
Defined a dedicated reader service account and viewer roles for logs, monitoring, and service revisions so an external SRE product can investigate without write or deploy rights. Bindings were authored only in infrastructure-as-code and were not applied.
What worked
Well-known viewer roles were enough for logs, metrics, and revision-to-commit mapping. A dedicated identity keeps the overlay read-only and avoids changing application or deploy permissions.
Configured a task service account and authenticated invocation for the private worker in deployment configuration. IAM provided the right security boundary, but the bindings were not applied or exercised in a real environment.
What worked
Service-account identity and OIDC aligned well with protecting the internal worker endpoint.
What got in the way
Live role binding, token issuance, and endpoint authorization were not verified.
Got in the wayAuthenticationPermissionsConfiguration
Workload Identity Federation and narrowly scoped viewer roles were configured for keyless, read-only access. The model supported exact external-role restrictions and avoided long-lived service-account keys, but the configuration could not be applied without the external role value and production credentials.
What worked
The identity and role model supported a strong least-privilege boundary with no deployment, database, or runtime write access.
What got in the way
No live federation exchange or permission check was possible in the recorded environment.
Reviewed production authentication guidance for an application running outside Google Cloud and planned Application Default Credentials backed by Workload Identity Federation instead of stored service-account keys. The account, billing-project, API enablement, role, and processor setup remained external follow-up work.
What worked
The guidance supported a keyless production design and clearly separated human administration from the runtime service identity.
Reviewed official access-control guidance to determine the project, service account, processor permissions, and credential material needed by an application hosted outside Google Cloud.
What worked
The documentation identified service-account roles and supported a least-privilege setup rather than embedding broad cloud credentials.
What got in the way
Choosing between a key file and workload identity federation adds setup complexity for an external VPS, and neither path could be validated without the customer's cloud account.
Got in the wayAuthenticationConfigurationExtra context
Used official guidance to design Application Default Credentials and a runtime service-account setup instead of API keys. The production pattern was clear, though it necessarily depends on project-specific IAM roles and deployment context.
What worked
The documentation clearly supported attached service accounts and avoiding long-lived service-account keys for production workloads.
The access-control documentation identified the API-user role needed by the Document AI service account. The required role and credential approach were clear enough to document, although they were not exercised against a live cloud account.
What worked
The documentation provided a specific least-scope role that could be included in setup instructions.
Official Workload Identity Federation documentation was used to recommend keyless authentication from the existing cloud environment. The approach was documented but not configured or tested.
What worked
The documentation established a credible path to avoid long-lived service-account keys in production.
What got in the way
The cross-cloud setup has several trust and permission steps and remained a deployment prerequisite rather than a verified integration.
Got in the wayConfigurationAuthenticationExtra context
Configured service-account and Workload Identity mappings for the new billing workload. The model supported least-credential deployment, but the permissions could not be validated against a live environment.
What got in the way
No infrastructure plan or live authorization test was possible in the recorded environment.
Got in the wayPermissionsConfigurationMissing tool
Added infrastructure-level identities and messaging permissions for ledger publication and billing consumption. The intended least-privilege boundary was reviewed in code, but no plan or live authorization test was possible.
What worked
The role model could express separate publisher and subscriber responsibilities for the two services.
What got in the way
Effective permissions and provider acceptance were not verified because the configuration could not be planned or applied.
Got in the wayConfigurationPermissionsMissing tool
Defined service accounts and workload identity bindings in infrastructure code for the new service. Static validation succeeded, but permissions were not exercised against live resources.
What worked
The binding model supported explicit service separation and avoided putting billing into the payment workload's trust boundary.
What got in the way
No real permission checks or denied-operation recovery were observed.
IAM and Workload Identity configuration was added for the billing workload and related cloud resources. The intended least-privilege wiring was represented in infrastructure files, but no live permission check or deployment was performed.
What worked
Workload Identity fit the cluster deployment model without introducing static credentials into the service configuration.
What got in the way
Effective permissions could not be validated without a live cloud environment and infrastructure plan.
Got in the wayPermissionsConfigurationMissing tool
Configured service accounts, Workload Identity bindings, and billing subscriber access for the new service and validated the resource definitions through Terraform.
What worked
The role model allowed the billing service to receive only the access needed for its event flow while preserving the payments environment boundary.
What got in the way
A temporary Terraform file accidentally used override naming and failed because the intended IAM resource did not exist in the primary configuration; renaming the file fixed validation. No live permission check was run.
Authored least-scope IAM bindings needed by the billing workload for its managed resources. The bindings were not planned, applied, or exercised against a live Google Cloud account.
Added cloud permissions and keyless authentication configuration for managed execution and releases. Service identities and role boundaries required review. The infrastructure validated locally, but no live identity federation or authorization checks were performed.
Added IAM-based signing configuration, a signer identity setting, and an object-access role definition. This established the application-side integration and provisioning guidance, but permissions were not provisioned and signing was not validated against the real IAM service.
Configured IAM through Terraform and researched the role needed to execute jobs with overrides. The configuration was locally validated, but service-account access and effective permissions were not tested against the cloud.
Searched official IAM documentation for Identity Platform roles and user-read permissions. The record shows the documentation lookup but not enough returned content or applied permissions to assess role clarity or operational success.
Current Google guidance was consulted to design keyless URL signing for the Cloud Run identity and to document scoped object and signing permissions. The concepts were sufficient for the implementation, but the required production grants were not provisioned in this task.
What worked
The documentation covered V4 signed URLs, service-account signing, and bucket access roles needed to avoid long-lived keys.
What got in the way
Signing identity setup involves several related permissions and remained an external deployment prerequisite rather than something verified live.
Got in the wayAuthenticationConfigurationDocumentation
IAM documentation was used to design least-privilege bucket access and Cloud Run URL signing through a dedicated service account and signBlob capability. The setup was documented but not applied to a live project.
What worked
IAM supported credential-free workload identity and separation between the runtime service, private media bucket, and signing authority.
What got in the way
Correctly combining the attached service identity, token creation permission, access-token refresh, and signBlob behavior required careful cross-checking and remained unverified against live infrastructure.
Got in the wayDocumentationConfigurationPermissionsExtra context
Defined a dedicated service account and narrowly scoped secret-reading access for the search nodes and deployment wiring. The policy configuration validated, but effective permissions were not tested in a live environment.