Skip to content
/ answer-app Public

Generic repo as a base for Vertex AI Agent Builder deployment

License

MIT license
0 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

doughayden/answer-app

BranchesTags

Repository files navigation

VERTEX AI AGENT BUILDER ANSWER APP

The Answer App uses Vertex AI Agent Builder and the Discovery Engine API to serve a conversational search experience with generative answers grounded on document data.

Why would I want to use this?

  • Fully-managed conversational search: Use the Answer method (REST reference) for a stateful multi-turn conversational search with generative answers.
  • Production-capable performance: Autoscaling, concurrency, and regional redundancy via multiple Cloud Run services.
  • Integration flexibility: Authenticated external HTTPS endpoint for 3rd party systems (using Google-managed TLS).
  • Resource observability: Single-pane-of-glass Cloud Monitoring dashboard with customizable metrics and alerts.
  • Explainable and debuggable results: Investigate generative answers using the full RAG pipeline results logged to BigQuery.
  • Data-driven LLM-ops: Tune the conversational search agent using question/answer pairs labelled with user feedback.
  • Identity-Aware Proxy: Secure access control.
  • Automated deployments: One-click install and uninstall with Terraform and Cloud Build.

Why would I not want to use this?

When all you want to do is test or demonstrate Vertex AI conversational search, create a search app and preview the widget in the Cloud Console or embed it in a web page.

 

TABLE OF CONTENTS

Architecture

Installation

Un-Installation

Tests

Known Issues

Next Steps

Reference Information

 

ARCHITECTURE

Diagram

(return to top)
Application Architecture

Description

(return to top)

 

INSTALLATION

Prerequisites

(return to top)

Complete the prerequisite steps before deploying the resources with Terraform:

  1. User Account and Local Development Environment
  2. Clone the Repo

1. User Account and Local Development Environment

(return to prerequisites)

  • Your Google user account must be a Project Owner in the target Google Cloud project.
  • Terraform will deploy resources to your gcloud CLI configuration default project.

OPTION 1: Deploying from Google Cloud Shell:

Use Google Cloud Shell for a convenient environment with gcloud and terraform pre-installed. Your user account (core.account) and default project (core.project) should already be set up in the Cloud Shell environment.

  1. Select your target deployment project in the Cloud Console.
  2. Open Cloud Shell and confirm your gcloud configuration.
gcloud config list --format=yaml

Example output:

accessibility:
  screen_reader: 'True'
component_manager:
  disable_update_check: 'True'
compute:
  gce_metadata_read_timeout_sec: '30'
core:
  account: project-owner@example.com
  disable_usage_reporting: 'False'
  project: my-project-id
metrics:
  environment: devshell
  1. Optionally, set the default compute region (compute.region). The helper script will default to us-central1 if your gcloud configuration does not specify a region.
gcloud config set compute/region 'region' # replace with your preferred region if it's not 'us-central1'
  1. Set any other configuration values as needed.

OPTION 2: Deploying outside of Google Cloud Shell:

  1. Install Terraform.
  2. Install the Google Cloud SDK.
  3. Authenticate.
gcloud auth login
  1. Set the default project.
gcloud config set project 'my-project-id' # replace with your project ID
  1. Configure Application Default Credentials (ADC) for local development.
gcloud auth application-default login
  1. Optionally, set the default compute region (compute.region). The helper script will default to us-central1 if your gcloud configuration does not specify a region.
gcloud config set compute/region 'region' # replace with your preferred region if it's not 'us-central1'

 

2. Clone the Repo

(return to prerequisites)

Clone the repository and open a terminal session in the answer-app directory.

  • If you're using Cloud Shell, consider using Personal Access Tokens (PATs) to clone the repo if you encounter issues with SSH keys.

 

One-Click Deployment

(return to top)

The install.sh script automates the steps required to prepare the project and deploy the resources.

Source the install.sh script to install the answer-app.

source scripts/install.sh # change the path if necessary

 

Add an A record to the DNS Managed Zone

(return to top)

NOTE: You do not need to configure DNS if you set loadbalancer_domain to null in config.yaml and instead used the default nip.io domain.

 

Enable Vertex AI Agent Builder

(return to top)

A project Owner must enable Vertex AI Agent Builder in the Cloud Console to use the Discovery Engine API and the Agent Builder console. It's a one-time setup to accept terms for the project for as long as the API remains enabled. (Checking the box to agree to model sampling is optional.) Enable Vertex AI Agent Builder

 

Test the endpoint

(return to top)

  • A newly-created managed TLS certificate may take anywhere from 10-15 minutes up to 24 hours for the CA to sign after DNS propagates.
  • The Certificate Managed status will change from PROVISIONING to ACTIVE when it's ready to use.
  • Navigate to Network Services > Load balancing > select the load balancer > Frontend: Certificate > Select the certificate and wait for the status to change to ACTIVE. Active Managed Certificate
  • Alternatively you can check the status using gcloud commands
gcloud compute ssl-certificates list --global # list all certificates and get the **CERTIFICATE_NAME**

gcloud compute ssl-certificates describe **CERTIFICATE_NAME** --global --format="get(name,managed.status, managed.domainStatus)"
scripts/test_endpoint.sh # change the path if necessary
  • The server responds with a 200 status code and {"status":"ok"} if the endpoint is reachable and the TLS certificate is active.
  • It may take some more time after the certificate reaches ACTIVE Managed status before the endpoint responds with success. It may throw an SSLError due to mismatched client and server protocols until changes propagate.
    • Example errors:
      • curl: (35) LibreSSL/3.3.6: error:1404B410:SSL routines:ST_CONNECT:sslv3 alert handshake failure
      • curl: (35) LibreSSL SSL_connect: SSL_ERROR_SYSCALL in connection to 34.117.145.180.nip.io:443

 

Import documents

(return to top)

 

Configure Identity-Aware Proxy

(return to top)

Steps

  1. Search for Identity-Aware Proxy (or "IAP") in the Console to navigate to it, then select "Enable API". Once the API is enabled, select "Go to Identity-Aware Proxy".
  2. You will be prompted to "Configure Consent Screen". A consent screen is what is shown to a user to display which elements of their information are requested by the app and to let them choose whether to proceed. Select "Configure Consent Screen".

Identity-Aware Proxy configuration

  1. Select a User Type of "External" to share the app with users outside of your organization's domain. Select "Continue".

Identity-Aware Proxy user type of External

  1. When configuring your consent screen, identify your app with a name (ex. "Answer App").
  2. Provide a user support email address (any).
  3. Under "Authorized domains" select "Add Domain" and then list the app top-level domain as Authorized domain 1.
    • i.e. if you used the hostname app.example.com then the top-level domain is example.com.
    • If you used the default domain using the nip.io service with a hostname like 35.244.148.105.nip.io, then the top-level domain is nip.io.
  4. Add your email address as the "Developer contact information" email address. Click "Save and continue". You can also click "Save and continue" on the following two screens, then on the Summary page click "Back to Dashboard."
  5. From the "OAuth consent screen" summary page, under Publishing Status, select "Publish app" and "Confirm". This will allow you to add any Google identity to which you grant the "IAP-secured Web App User" role to access the app, rather than additionally needing to add them as a test user on the OAuth consent screen.

OAuth Consent Screen configuration

  1. Navigate back to the "Load Balancing" dashboard, select your load balancer, and then the Certificate name. If this is not yet ACTIVE, we will need to wait until it reaches ACTIVE status. Take a break and refresh occasionally.
  2. When the certificate is ACTIVE, navigate back to Identity-Aware Proxy by searching "IAP" at the top of the Console.
  3. Toggle on IAP protection ONLY for the answer-app-client backend service (and not for the answer-app backend). You will be prompted to review configuration requirements, and then select the checkbox confirming your understanding and select "Turn On."
    • The service will show an Error status when IAP is not enabled. This is expected until the IAP configuration is complete.
    • Users don't directly access the answer-app backend service. Instead, it enforces authentication on the calling service (the client app in this case). The IAP Error status will not affect it's functionality (which does not rely on IAP for access).

OAuth Consent Screen configuration

  1. Add a Google Identity (i.e a user or group) with the "IAP-secured Web App User" role.
  2. You may see an "Error: Forbidden" message for about the first 5 minutes, but after that users with the "IAP-secured Web App User" role on the Project or IAP backend service should be able to access the app via the domain on the Load Balancer certificate.
    • i.e. https://app.example.com or https://35.244.148.105.nip.io

 

Use the app

(return to top)

Terraform deploys a streamlit web client to Cloud Run that's accessible via the load balancer domain.

  • Get the client app URL from Terraform output.(It's in the format https://<load_balancer_domain>)
cd terraform/main # change the path if necessary
terraform output client_app_uri

Example output:

"https://34.8.148.243.nip.io"
  • Open the URL in a web browser to access the client app. On Mac or Linux, you can use the open command.
cd terraform/main # change the path if necessary
/usr/bin/open -a "/Applications/Google Chrome.app" $(terraform output -raw client_app_uri)
  • Use the web client to send questions to the answer-app endpoint and view the generative answers.
  • Hover over the inline citations to view details about the source document chunks used to generate the answer.
  • Click on inline citations or the links in the Citations footer to view the source documents in a new tab. Web Client

 

UNINSTALL

(return to top)

The uninstall.sh script destroys all Terraform-provisioned infrastructure and removes project prerequisites by calling the un_bootstrap.sh script.

Source the uninstall.sh script to remove all answer-app resources from the project.

source scripts/uninstall.sh # change the path if necessary

 

TESTS

Unit Tests

(return to top)

Run pytest using poetry.

NOTE: The tests will fail if you've used the Helper Scripts to set the environment variables. Open a new shell session with a clean environment to run the tests.

  1. Install Poetry

  2. Install dependencies.

cd answer-app # the root of the repository where this README and the pyproject.toml file is located - change the path if necessary
poetry install
  1. Run the tests.
poetry run pytest
  1. Optionally run pytest with coverage and view the report.
poetry run coverage run -m pytest
poetry run coverage report -m

 

Run locally

(return to top)

Setup

source scripts/set_variables.sh # change the path if necessary
  • Install Poetry and the project dependencies (see Unit Tests).

Poetry

answer-app

  • The service will listen on local port 8888.
poetry run uvicorn main:app --app-dir src/answer_app --reload --host localhost --port 8888

client

  • With the environment variables set using the set_variables.sh script, the client app automatically gets and impersonated ID token for the Terraform service account on behalf of the user and sets the target audience for requests to localhost:8888.
  • To solve unexpected communication issues, restart a fresh shell session to clear the environment, then re-source the set_variables.sh script.
poetry run streamlit run src/client/streamlit_app.py

Docker

Build

answer-app

  • Uses the Dockerfile in the root of the repo
docker build -t local-answer-app:0.1.0 . # change image name and tag as needed

client

  • Uses a Dockerfile in the src/client directory
docker build -t local-answer-app-client:0.1.0 -f ./src/client/Dockerfile . # change image name tag as needed

Run

  • Use --rm to remove the container when it exits.
  • Use -v to mount the host's gcloud configuration directory to the container's /root/.config/gcloud directory to allow the app to use Application Default Credentials for authentication.
  • Use -e to set container environment variables for the Google Cloud project, log level, target service account, and audience.
  • Use -p to map the container's port 8080 to the host's port 8888 or 8080 (specify port mapping as -p <host-port>:<container-port>).
  • The $PROJECT and $TF_VAR_terraform_service_account environment variables will already be set after you ran the set_variables.sh script.

answer-app: map container port 8080 to localhost:8888

docker run --rm -v $HOME/.config/gcloud:/root/.config/gcloud \
-e GOOGLE_CLOUD_PROJECT=$PROJECT \
-e LOG_LEVEL=DEBUG \
-p 8888:8080 local-answer-app:0.1.2 # change image name and tag as needed

client: map container port 8080 to localhost:8080 and call the LOCAL answer-app service at localhost:8888

docker run --rm -v $HOME/.config/gcloud:/root/.config/gcloud \
-e GOOGLE_CLOUD_PROJECT=$PROJECT \
-e LOG_LEVEL=DEBUG \
-e "TF_VAR_terraform_service_account=$TF_VAR_terraform_service_account" \
-p 8080:8080 local-answer-app-client:0.1.0 # change env vars and image name and tag as needed
  • Open your browser to http://localhost:8080.
open -a "/Applications/Google Chrome.app" http://localhost:8080

client: map container port 8080 to localhost:8080 and call the DEPLOYED answer-app service at https://app.mydomain.com/answer-app

  • To target the deployed answer-app backend service, set the AUDIENCE environment variable to the custom audience for the answer-app service (see Use the Helper Scripts).
  • NOTE: You may want to unset the AUDIENCE environment variable after testing the deployed service if you want to continue local-only testing afterward.
source scripts/set_audience.sh # change the path if necessary
docker run --rm -v $HOME/.config/gcloud:/root/.config/gcloud \
-e GOOGLE_CLOUD_PROJECT=$PROJECT \
-e LOG_LEVEL=DEBUG \
-e "TF_VAR_terraform_service_account=$TF_VAR_terraform_service_account" \
-e "AUDIENCE=$AUDIENCE" \
-p 8080:8080 local-answer-app-client:0.1.0 # change env vars and image name and tag as needed
  • Open your browser to http://localhost:8080.
open -a "/Applications/Google Chrome.app" http://localhost:8080

Debug

  • Open a sh shell in the container image.
docker run --entrypoint /bin/sh --rm -it local-answer-app-client:0.1.0 # change image name and tag as needed

 

KNOWN ISSUES

Failure to create the Artifact Registry repository

(return to top)

Problem

When running the bootstrap module, Terraform fails to create the Artifact Registry repository.

Example:

╷
│ Error: Error creating Repository: googleapi: Error 403: Permission 'artifactregistry.repositories.create' denied on resource '//artifactregistry.googleapis.com/projects/my-project-id/locations/us-central1' (or it may not exist).
│ Details:
│ [
│   {
│     "@type": "type.googleapis.com/google.rpc.ErrorInfo",
│     "domain": "artifactregistry.googleapis.com",
│     "metadata": {
│       "permission": "artifactregistry.repositories.create",
│       "resource": "projects/my-project-id/locations/us-central1"
│     },
│     "reason": "IAM_PERMISSION_DENIED"
│   }
│ ]
│ 
│   with google_artifact_registry_repository.cloud_run,
│   on main.tf line 38, in resource "google_artifact_registry_repository" "cloud_run":
│   38: resource "google_artifact_registry_repository" "cloud_run" {
│ 
╵

Solution

The error occurs on the first run of the bootstrap module due to a race condition between the Artifact Registry API activation and applying the Terraform plan. The API activation can take a few minutes to complete. Rerun the bootstrap.sh script or manually re-apply the bootstrap module configuration.

 

Cloud Build fails with a Cloud Storage 403 permission denied error

(return to top)

Problem

Cloud Build fails with a storage.objects.get access error when trying to access the Google Cloud Storage object after the first run of the bootstrap module.

Example:

> gcloud builds submit . --config=cloudbuild.yaml --project=$PROJECT --region=$REGION --impersonate-service-account=$TF_VAR_terraform_service_account --verbosity=error --substitutions="_RUN_TYPE=apply"
Creating temporary archive of 25 file(s) totalling 650.8 KiB before compression.
Uploading tarball of [.] to [gs://my-project-id_cloudbuild/source/1732724661.794156-c07691d9db7449499595855914578017.tgz]
ERROR: (gcloud.builds.submit) INVALID_ARGUMENT: could not resolve source: googleapi: Error 403: run-app-cloudbuild@my-project-id.iam.gserviceaccount.com does not have storage.objects.get access to the Google Cloud Storage object. Permission 'storage.objects.get' denied on resource (or it may not exist)., forbidden

Solution

The error can occur shortly after setting up a new project with the bootstrap module for the first time. It's a race condition where the Cloud Build service account IAM role bindings have not yet propagated. Rerun the Cloud Build job to resolve the error.

 

Error creating a DataStore (named DataStore being deleted)

(return to top)

Problem

When re-running the Terraform configuration after deleting the DataStore, the following error occurs:

╷
│ Error: Error creating DataStore: googleapi: Error 400: DataStore projects/.../locations/global/collections/default_collection/dataStores/cymbal-bank-data-store is being deleted, please wait for deletion to complete before recreating with the same ID. The deletion could take a couple of hours.
│ 
│   with module.answer_app.google_discovery_engine_data_store.layout_parser_data_store["answer-app-default"],
│   on ../modules/answer-app/discoveryengine.tf line 1, in resource "google_discovery_engine_data_store" "layout_parser_data_store":
│    1: resource "google_discovery_engine_data_store" "layout_parser_data_store" {
│ 
╵

Solution

Choose one of the following solutions:

  • Wait for the DataStore deletion to complete before re-running the Terraform configuration.
  • Change the DataStore ID in the Terraform configuration to create a new DataStore. Re-apply Terraform.

 

Errors adding users to Identity-Aware Proxy

(return to top)

Problem

When adding members to the IAP-secured backend service, a Domain restricted sharing Org policy causes an error message like this:
Policy update failed\

Solution

  1. Edit the policy to temporarily disable it.
  2. Add the members to IAP-protected backend service IAM policy.
  3. Re-enable the policy.

 

Errors reading or editing Terraform resources

(return to top)

Problem

Intermittent connectivity issues (for example, while using a VPN) can cause unresponsiveness during plan or apply operations.

Example:

│ Error: Error when reading or editing RedisInstance "projects/my-project/locations/us-central1/instances/my-redis-instance": Get "https://redis.googleapis.com/v1/projects/my-project/locations/us-central1/instances/my-redis-instance?alt=json": write tcp [fe80::ca4b:d6ff:fec7:8a11%utun1]:59235->[2607:f8b0:4009:809::200a]:443: write: socket is not connected
│ 
│   with google_redis_instance.default,
│   on redis.tf line 79, in resource "google_redis_instance" "default":
│   79: resource "google_redis_instance" "default" {
│ 
╵

Solution

Retry the operation to clear the error. If the error persists, check your network or VPN connection and try again.

 

The Search Agent refuses to answer questions

(return to top)

Problem

  • The Search agent responds with "No results could be found. Try rephrasing the search query." when the query is valid and should be answerable from the data store documents.
  • Cloud Run logs do not include citations or references for the answer response.
  • The Big Query answer-app.conversations table includes null values for the answer.citations, answer.references, or answer.related_questions columns.
  • The Big Query answer-app.conversations table might list some but not all of the documents expected to be returned in the search results.
  • The search preview in the Vertex AI Agent Builder console also does not return any results.
  • The data store documents appear to be properly imported with no errors in the Vertex AI Agent Builder console.

Solution

Importing documents using the refresh options without first purging the data store in separate step may result in this issue. The import operation may actually not have imported and indexed the documents properly. Purge the data store and re-import the documents to resolve the issue.

 

Failure to remove regional backends

(return to top)

Problem

Terraform fails to remove the regional backend Network Endpoint Groups connected to the backend service. It will throw a 'resource in use by another resource' error.

Solution

Apply changes in multiple steps to remove the regional backends:

  1. Edit the backend service to include only the default regional backend group.
    • Change line 95 in terraform/modules/answer-app/cloudrun.tf (module.answer-app.google_compute_backend_service.run_app) from
    group = google_compute_region_network_endpoint_group.run_app[backend.key].id
    to 
    group = google_compute_region_network_endpoint_group.run_app[var.region].id
  2. Apply the changes to update the backend service while leaving the additional_regions variable list full with the backends you wish to later remove.
  3. Remove the regions you want to destroy from the additional_regions list variable.
  4. Apply the changes to remove the regional backend NEGs and their associated Cloud Run Services.
  5. Revert the change to Change line 95 in terraform/modules/answer-app/cloudrun.tf back to group = google_compute_region_network_endpoint_group.run_app[backend.key].id and apply the changes to confirm the configuration works.

 

NEXT STEPS

(return to top)

Features to add

Priority

  • [] stream answers NOT AVAILABLE YET
  • collect user feedback for each question and answer pair and log to BQ
  • implement an optional list of regions to support multiple cloud run backends
  • deploy a templated monitoring dashboard via Terraform
  • replace manual utils._response_to_dict() with native proto-plus class method conversion Message.to_dict()
    • update downstream features as needed (bq conversations table, client app attribute extraction) to use the native serialization format
    • use preserving_proto_field_name=True (and optionally always_print_fields_with_no_presence=True, reference) to maintain the original field names and minimize or eliminate downstream changes
  • [] add document processing utilities (use async)
  • add structured prompt preamble to the client app config.yaml
  • create a separate run invoker service account for the client app
  • [] add a cloud logging handler to the answer-app server
  • [] provide a walkthrough of connecting the search agent to spark/agentspace (via a DFCX steering bot)
  • replace architecture diagram with one consistent with google cloud style
  • [] add multiple data stores
    • create at least 2 when creating the search app (one can be empty) to support adding/removing data stores later (a search app created with only one data must always have only one)
    • [website data, structured data, structured content (media), structured content for third-party data sources (maybe omit this), unstructured data (exists), healthcare FHIR data (maybe omit this)]
    • Agent Builder ref, Terraform ref
  • [] demonstrate IAP for service account auth to call the answer-app service

Pipeline analysis

  • [] establish a 'golden Q&A' to evaluate the model's performance (consider using notebooklm to generate the golden Q&A)
    • alternatively use deep-eval RAG triad without needing golden Q&A
  • [] automate RAG evaluation (using golden Q&A) - run with each new doc import and save results to BQ for offline analysis
  • [] automate load testing (an extension of automated evaluation) to test the pipeline's autoscaling capabilities - save results to BQ for offline analysis
  • [] add example (saved) queries in BQ to demonstrate common debugging and evaluation reports
  • [] add a looker dashboard with question and answer pairs along with relevancy scores and user feedback - graph relevancy and user feedback over time

Client

  • [] provide example questions to ask the search agent
  • [] mimic the answer format returned by the console widget (expand inline citations on click)
  • [] implement unique user IDs to establish sessions
  • [] implement end user authentication with identity platform or firebase
  • [] resumable conversation sessions (lookup available active sessions per user ID)
  • [] demonstrate workload identity federation for client apps not within GCP

 

REFERENCE INFORMATION

Answer Method Configuration Options (Python)

(return to top)

The Discovery Engine API answer method accepts a variety of configuration options to customize the search and answer generation phases. The following Python code snippet demonstrates how to configure the answer method with the available options.

The Python client library does not currently implement all of the features available in the RPC API. (SafetySetting, for example)

Click to expand 
async def answer_query(
    self,
    query_text: str,
    session_id: str | None,
) -> AnswerQueryResponse:
    """Call the answer method and return a generated answer and a list of search results,
    with links to the sources.

    Args:
        query_text (str): The text of the query to be answered.
        session_id (str, optional): The session ID to continue a conversation.

    Returns:
        AnswerQueryResponse: The response from the Conversational Search Service,
        containing the generated answer and selected references.

    Refs:
    https://cloud.google.com/generative-ai-app-builder/docs/answer#search-answer-basic
    https://cloud.google.com/python/docs/reference/discoveryengine/latest/google.cloud.discoveryengine_v1.types.AnswerQueryRequest
    """
    # serving_config.
    engine = f"projects/{self._project_id}/locations/{self._location}/collections/default_collection/engines/{self._engine_id}"
    serving_config = f"{engine}/servingConfigs/default_serving_config"

    # Query: accepts one of `text` or `query_id`.
    query = discoveryengine.Query(text=query_text)
    # query = discoveryengine.Query(query_id="query_id")

    # session - construct the name using the engine as the serving config.
    session = f"{engine}/sessions/{session_id}" if session_id else None

    # SafetySpec - options for search phase.
    safety_spec = discoveryengine.AnswerQueryRequest.SafetySpec(enable=False)

    # RelatedQuestionsSpec.
    related_questions_spec = discoveryengine.AnswerQueryRequest.RelatedQuestionsSpec(enable=False)

    # AnswerGenerationSpec - options for answer phase.
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
            model_version="gemini-1.5-flash-001/answer_gen/v2",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble=self._preamble,
        ),
        include_citations=True,
        answer_language_code="en",
        ignore_adversarial_query=False,
        ignore_non_answer_seeking_query=False,
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        ignore_jail_breaking_query=False,
    )

    # SearchSpec - options for search phase.
    search_spec = discoveryengine.AnswerQueryRequest.SearchSpec(
        # one of `search_params` or `search_result_list` must be set.
        search_params=discoveryengine.AnswerQueryRequest.SearchSpec.SearchParams(
            max_return_results=10,
            filter="filter_expression",
            boost_spec=discoveryengine.SearchRequest.BoostSpec(
                condition_boost_specs=[
                    discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec(
                        condition="condition_1",
                        boost=0.0,
                        boost_control_spec=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec(
                            field_name="boost_field_1",
                            # select one attribute_type
                            attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.ATTRIBUTE_TYPE_UNSPECIFIED,
                            # attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.NUMERICAL,
                            # attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.FRESHNESS,
                            # select one interpolation_type
                            interpolation_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.InterpolationType.INTERPOLATION_TYPE_UNSPECIFIED,
                            # interpolation_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.InterpolationType.LINEAR,
                            control_points=[
                                discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.ControlPoint(
                                    attribute_value="value_1",
                                    boost_amount=0.0,
                                ),
                                discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.ControlPoint(
                                    attribute_value="value_2",
                                    boost_amount=0.0,
                                ),
                            ],
                        ),
                    ),
                    discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec(
                        condition="condition_2",
                        boost=0.0,
                        boost_control_spec=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec(
                            field_name="boost_field_2",
                            # select one attribute_type
                            attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.ATTRIBUTE_TYPE_UNSPECIFIED,
                            # attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.NUMERICAL,
                            # attribute_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.AttributeType.FRESHNESS,
                            # select one interpolation_type
                            interpolation_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.InterpolationType.INTERPOLATION_TYPE_UNSPECIFIED,
                            # interpolation_type=discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.InterpolationType.LINEAR,
                            control_points=[
                                discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.ControlPoint(
                                    attribute_value="value_1",
                                    boost_amount=0.0,
                                ),
                                discoveryengine.SearchRequest.BoostSpec.ConditionBoostSpec.BoostControlSpec.ControlPoint(
                                    attribute_value="value_2",
                                    boost_amount=0.0,
                                ),
                            ],
                        ),
                    ),
                ],
            ),
            order_by="field_name_for_ordering",
            # select one search_result_mode
            search_result_mode=discoveryengine.SearchRequest.ContentSearchSpec.SearchResultMode.DOCUMENTS,
            # search_result_mode=discoveryengine.SearchRequest.ContentSearchSpec.SearchResultMode.SEARCH_RESULT_MODE_UNSPECIFIED,
            # search_result_mode=discoveryengine.SearchRequest.ContentSearchSpec.SearchResultMode.CHUNKS,
            data_store_specs=[
                discoveryengine.SearchRequest.DataStoreSpec(
                    data_store="projects/{project}/locations/{location}/collections/{collection_id}/dataStores/{data_store_id}",
                    filter="filter_expression",
                )
            ],
        ),
        # search_result_list=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList(
        #     search_results=[
        #         discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult(
        #             unstructured_document_info=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo(
        #                 document="document_resource_name",
        #                 uri="document_uri",
        #                 title="document_title",
        #                 document_contexts=[
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.DocumentContext(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.DocumentContext(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #                 extractive_segments=[
        #                     # Guide <https://cloud.google.com/generative-ai-app-builder/docs/snippets#extractive-segments>
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveSegment(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveSegment(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #                 extractive_answers=[
        #                     # Guide <https://cloud.google.com/generative-ai-app-builder/docs/snippets#get-answers>
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveAnswer(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveAnswer(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #             ),
        #             chunk_info=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.ChunkInfo(
        #                 chunk="chunk_resource_name",
        #                 content="chunk_content",
        #                 document_metadata=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.ChunkInfo.DocumentMetadata(
        #                     uri="document_uri",
        #                     title="document_title",
        #                 ),
        #             ),
        #         ),
        #         discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult(
        #             unstructured_document_info=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo(
        #                 document="document_resource_name",
        #                 uri="document_uri",
        #                 title="document_title",
        #                 document_contexts=[
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.DocumentContext(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.DocumentContext(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #                 extractive_segments=[
        #                     # Guide <https://cloud.google.com/generative-ai-app-builder/docs/snippets#extractive-segments>
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveSegment(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveSegment(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #                 extractive_answers=[
        #                     # Guide <https://cloud.google.com/generative-ai-app-builder/docs/snippets#get-answers>
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveAnswer(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                     discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.UnstructuredDocumentInfo.ExtractiveAnswer(
        #                         page_identifier="page_identifier",
        #                         content="content",
        #                     ),
        #                 ],
        #             ),
        #             chunk_info=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.ChunkInfo(
        #                 chunk="chunk_resource_name",
        #                 content="chunk_content",
        #                 document_metadata=discoveryengine.AnswerQueryRequest.SearchSpec.SearchResultList.SearchResult.ChunkInfo.DocumentMetadata(
        #                     uri="document_uri",
        #                     title="document_title",
        #                 ),
        #             ),
        #         ),
        #     ]
        # ),
    )

    # QueryUnderstandingSpec - options for query phase.
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.TYPE_UNSPECIFIED,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.JAIL_BREAKING_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY_V2,
            ]
        ),
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,
            max_rephrase_steps=1, # max allowed is 5 steps
        ),
        
    )

    # user_pseudo_id.
    user_pseudo_id = "user_pseudo_id"

    # user_labels.
    user_labels = {"key": "value"}

    # Initialize request argument(s).
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=query,
        session=session,
        safety_spec=safety_spec,
        related_questions_spec=related_questions_spec,
        answer_generation_spec=answer_generation_spec,
        search_spec=search_spec,
        query_understanding_spec=query_understanding_spec,
        user_pseudo_id=user_pseudo_id,
        user_labels=user_labels,
    )

    # Make the request.
    response = await self._client.answer_query(request)

    # Handle the response.
    logger.debug(response)
    logger.info(f"Answer: {response.answer.answer_text}")

    return response

 

Helper Scripts

(return to top)

Shell scripts in the terraform/scripts directory automate common tasks.

  • install.sh: Deploy the resources for the answer-app service. (Combines other scripts to set up the project and deploy the resources.)
  • bootstrap.sh: Prepare the target deployment project. (Used by the install.sh script.)
  • set_audience.sh: Set the custom audience used to call the answer-app service. (Used by the test_endpoint.sh script.)
  • set_token.sh: Set the ID token used to call the answer-app service. (Used by the test_endpoint.sh script.)
  • set_variables.sh: Set the environment variables for the shell session. (Used by the bootstrap.sh, set_audience.sh, un_bootstrap.sh, and uninstall.sh scripts and by Cloud Build.). Variables:
    • PROJECT: The Google Cloud project ID.
    • REGION: The default compute region.
    • BUCKET: The staging bucket for Vertex AI Data Store documents.
    • TF_VAR_project_id: The Google Cloud project ID for Terraform. (Automatically read by Terraform.)
    • TF_VAR_terraform_service_account: The Terraform service account email address. (Automatically read by Terraform.)
  • terraform_service_account_roles.txt: The IAM roles granted to the Terraform service account. (Used by the bootstrap.sh script.)
  • test_endpoint.sh: Test the answer-app endpoint with a curl request.
  • un_bootstrap.sh: Remove the resources created by the bootstrap.sh script. (Used by the uninstall.sh script.)
  • uninstall.sh: Remove the answer-app resources from the project. (Combines other scripts to remove the resources and clean up the project.)

 

Bootstrap

(return to top)

The bootstrap.sh script automates the gcloud and terraform commands required to prepare the project.

  • Source the set_variables.sh script to configure the shell environment, including Terraform environment variables. (Used automatically by later Terraform commands.).
  • Enable the Service Usage, IAM, and Service Account Credentials APIs.
  • Create a service account for Terraform provisioning.
  • Grant the required IAM roles to the service account.
  • Prepare Service Account Impersonation for the caller's user account.
  • Create a Terraform remote state bucket.
  • Initialize the Terraform bootstrap module and apply to provision resources required for the main module:
    • Project APIs.
    • Cloud Build service account.
    • Artifact Registry repository.
    • IAM role bindings for the Cloud Build service account:
      • Project IAM policy: Cloud Build Service Account (roles/cloudbuild.builds.builder) role.
      • Terraform service account IAM policy: Service Account Token Creator (roles/iam.serviceAccountTokenCreator) role.

Source the bootstrap.sh script to set your shell variables and run the Terraform commands.

source scripts/bootstrap.sh # change the path if necessary

 

Automate Deployments with Cloud Build

(return to top)

Use gcloud builds submit with build config files to plan and deploy project resources.

1. Set configuration values in config.yaml.

Verify/Change parameters as needed:

2. Set environment variables.

Source the set_variables.sh script to configure the shell environment if you restarted your shell session or made changes to the environment variables.

  • The bootstrap.sh script sources this file to set the environment variables and it's not necessary to run it again in the same shell session.
source scripts/set_variables.sh # change the path if necessary

3. Build & push the docker images and apply the Terraform configuration

Use gcloud to submit the build from the answer-app root directory (the location of this README file) as the build context.

  • [OPTIONAL] Omit the _RUN_TYPE=apply substitution to run a plan-only build and review the Terraform changes before applying.
cd /path/to/answer-app # replace with the local system path to the answer-app root directory
gcloud builds submit . --config=cloudbuild.yaml --project=$PROJECT --region=$REGION --substitutions="_RUN_TYPE=apply"

 

Connect cloud run services to an existing load balancer

(return to top)

Edit and verify these files to connect the Cloud Run services to an existing load balancer.

config.yaml

  • Set create_loadbalancer = false AND set loadbalancer_domain to the value of the existing load balancer domain.

cloudrun.tf

  • Ensure the Terraform cloud-run module resource google_compute_backend_service.run_app argument load_balancing_scheme matches the existing load balancer type.
    • Use load_balancing_scheme = "EXTERNAL_MANAGED" for the Global external Application Load Balancer.
    • Use load_balancing_scheme = "EXTERNAL" for the Classic Application Load Balancer.
  • Cloud Run services will use the load balancer domain in the custom audience for authentication.
  • You must connect the backend services to the existing load balancer outside of this repo's Terraform configuration.

 

Rollbacks

(return to top)

Option 1: Use the Cloud Console to switch Cloud Run service traffic to a different revision

THIS WILL CHANGE STATE OUTSIDE OF TERRAFORM CONTROL

  • Navigate to the Cloud Run service in the Cloud Console.
  • Click the 'Revisions' tab.
  • Click 'MANAGE TRAFFIC'.
  • Select the target revision and traffic percentage (100% to rollback completely to another revision).
  • Click 'SAVE'.

Option 2: Rollback to a different Docker image using Terraform

  • Identify the rollback target Docker image.
  • Pass the target image name and tag to the docker_image input variable in the main root module.
  • Apply the Terraform configuration to update the Cloud Run service to the rollback target.

Example: select an image by digest or tag from Artifact Registry.

  1. Source the set_variables.sh script to configure the shell environment (provides variables including PROJECT and BUCKET). Refer to the Prerequisites section for details on configuring the gcloud CLI.
source scripts/set_variables.sh # change the path if necessary
  1. Initialize the Terraform main module configuration with the remote state by passing required arguments to the partial backend configuration. 
cd terraform/main # change the path if necessary
terraform init -backend-config="bucket=$BUCKET" -backend-config="impersonate_service_account=$TF_VAR_terraform_service_account" -reconfigure
  1. Set the Terraform input environment variable TF_VAR_docker_image to the target image names.
    • The variable value is a JSON object with key-value pairs for each service name and corresponding image name.
    • Don't include spaces or line breaks in the JSON string when setting this environment variable.
    • This example uses the tag Terraform adds to image names corresponding to the Cloud Build ID.
    • You can also use the image digest, e.g., us-central1-docker.pkg.dev/my-project-id/answer-app/answer-app@sha256:4f2092b926b7e9dc30813e819bb86cfa7d664eede575539460b4a68bbd4981e1.
export TF_VAR_docker_image='{"answer-app":"us-central1-docker.pkg.dev/my-project-id/answer-app/answer-app:50bde371-76f8-452a-8a17-9b63273f13e0"}'
  • Apply the Terraform configuration to update the Cloud Run service to the target image.
terraform apply

 

Execute Terraform to apply infrastructure-only changes to the bootstrap or main module.

(return to top)

Applying the resources from the bootstrap module is a one-time setup for a new project. You can re-run it, for example, to add or enable additional APIs to support future development. You can apply infrastructure-only changes to the main module to update cloud resources without rebuilding the Docker images (and without using Cloud Build). If you don't provide docker image names as input variables, the main retrieves the last-deployed Docker images from the module state and reuses them in the Cloud Run services.

  1. Source the set_variables.sh script to configure the shell environment (provides variables including PROJECT and BUCKET). Refer to the Prerequisites section for details on configuring the gcloud CLI.
source scripts/set_variables.sh # change the path if necessary
  1. Initialize the Terraform main or bootstrap module using a partial backend configuration. 
cd terraform/main # or cd terraform/bootstrap, change the path if necessary
terraform init -backend-config="bucket=$BUCKET" -backend-config="impersonate_service_account=$TF_VAR_terraform_service_account" -reconfigure
  1. Apply the Terraform configuration to provision the cloud resources.
terraform apply

 

Un-Bootstrap

(return to top)

WARNING: THIS WILL DELETE YOUR TERRAFORM STATE

The un_bootstrap.sh script automates gcloud commands to remove these project resources created by the bootstrap script. It does not disable any APIs.

  • Terraform state bucket
  • Terraform service account project role bindings
  • Terraform service account resource

Source the un_bootstrap.sh script to remove the resources.

source scripts/un_bootstrap.sh # change the path if necessary

 

Service account impersonation

(return to top)

Instead of creating and managing Service Account keys for authentication, this code uses an impersonation pattern for Terraform to fetch access tokens on behalf of a Google Cloud IAM Service Account.

  • Grant the caller (a Google user account or group address) permission to generate short-lived access tokens on behalf of the targeted service account.
    • The caller needs the Account Token Creator role (roles/iam.serviceAccountTokenCreator) or a custom role with the iam.serviceAccounts.getAccessToken permission that applies to the Terraform provisioning service account.
    • Create a role binding on the Service Account resource instead of the project IAM policy to minimize the scope of the permission.
    • Perhaps counterintuitively, the primitive Owner role (roles/owner) does NOT include this permission.
export MEMBER='user:{your-username@example.com}' # replace '{your-username@example.com}' from 'user:{your-username@example.com}' with your Google user account email address
# Example to add a group -> export MEMBER='group:devops-group@example.com'

gcloud iam service-accounts add-iam-policy-binding "terraform-service-account@${PROJECT}.iam.gserviceaccount.com" --member=$MEMBER --role="roles/iam.serviceAccountTokenCreator" --condition=None

 

Terraform Overview

(return to top) General instructions to initialize a Terraform workspace/environment and set up a backend configuration and bucket for storing Terraform state.

Initialize

(return to Terraform Overview)

The Terraform working directory must be initialized to set up configuration files and download provider plugins.

# Initialize the working directory.
terraform init

Workspaces

(return to Terraform Overview)

Terraform workspaces allow separation of environments so each is managed in a unique state file.

# View the active and available workspaces (Terraform starts with only the 'default' workspace).
terraform workspace list

# Set an environment variable for the deployment environment/workspace name.
export ENVIRONMENT='sandbox'

# Create an environment-specific workspace.
terraform workspace new $ENVIRONMENT

# Choose a workspace.
terraform workspace select default

# Select a workspace or create it if it doesn't exist.
terraform workspace select -or-create nonprod

Terraform Backends

(return to Terraform Overview)

Using the default (local) backend doesn't require additional configuration. A Cloud Storage backend requires these prerequisites:

  • The GCS backend bucket must already exist - Terraform will not create it at init. Example (edit with your actual project and bucket name):
gcloud storage buckets create gs://my-terraform-bucket --project=my-project --uniform-bucket-level-access
  • The caller or impersonated service account needs permission to read and write to the bucket.
  • Define a GCS backend in the terraform.backend block.
terraform {
  backend "gcs" {
    bucket                      = "my-terraform-bucket"
    impersonate_service_account = "terraform-service-account@my-project-id..iam.gserviceaccount.com"
    prefix                      = "terraform_state/"
  }
  required_providers {
    google = {
      source  = "hashicorp/google"
      version = "5.25.0"
    }
  }
  required_version = ">= 0.13"
}

Flexible Backends - Partial Configuration

(return to Terraform Overview)

  • Backend declaration can't accept input variables or use expansion/interpolation because Terraform loads the backend config before anything else.
  • A partial configuration in the terraform.backend block allows flexible backend definition for multiple environments.
  • Partial configurations allow you to include some attributes in the terraform.backend block and pass the rest from another source.
  • Options for supplying backend configuration arguments include a file, command-line key/value arguments, environment variables, or interactive prompts.
  • Define the remaining backend details in a dedicated *.gcs.tfbackend file, i.e. backend_sandbox.gcs.tfbackend and pass it's path as a command-line argument to separate backends per environment. (Hashicorp docs recommend a *.backendname.tfbackend naming convention, but Terraform will accept any correctly-formatted file. IDE syntax highlighting and linting might not pick up .tfbackend files.)

Partial backend configuration example:

backend.tf:

terraform {
  backend "gcs" {
    prefix = "bootstrap"
  }
  required_providers {
    google = {
      source  = "hashicorp/google"
      version = ">=5.25.0"
    }
  }
  required_version = ">= 0.13"
}
  • Example 1 - initialize a partial backend using an environment-specific backend configuration file:
# Create a workspace-specific backend configuration file for the Google Cloud Storage backend.
cat << EOF > backend_$ENVIRONMENT.gcs.tfbackend
bucket                      = "terraform-state-my-project-id"
impersonate_service_account = "terraform-service-account@my-project-id.iam.gserviceaccount.com"
EOF

# Initialize the remote state
terraform init -backend-config="backend_$ENVIRONMENT.gcs.tfbackend"
  • Example 2 - initialize a partial backend using command-line arguments:
terraform init -backend-config="bucket=terraform-state-my-project-id" -backend-config="impersonate_service_account=terraform-service-account@my-project-id.iam.gserviceaccount.com"

Reconfiguring a Backend

(return to Terraform Overview)

To force Terraform to use a new backend without migrating state data from an existing backend, initialize with the -reconfigure flag. The existing state in the old backend is left unchanged and not copied to the new backend.

terraform init -reconfigure -backend-config="backend_$ENVIRONMENT.gcs.tfbackend

Plan and Apply

(return to Terraform Overview)

Terraform requires declared or default values for input variables. For example, variables defined in .tfvars files to separate environments.

# Define environment-specific variables in a .tfvars file
cat << EOF > vars_$ENVIRONMENT.tfvars
project_id                = "my-project"
terraform_service_account = "terraform-sa@my-project.iam.gserviceaccount.com"
terraform_bucket_name     = "my-terraform-bucket"
region                    = "us-central1"
zone                      = "us-central1-a"
EOF

# View the Terraform plan.
terraform plan -var-file="vars_$ENVIRONMENT.tfvars"

# Apply changes.
terraform apply -var-file="vars_$ENVIRONMENT.tfvars"

About

Generic repo as a base for Vertex AI Agent Builder deployment

Resources

Readme

License

MIT license
Activity

Stars

0 stars

Watchers

2 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages