Add support for Google Cloud

[realchh]

[Major Changes] Added support for Google Cloud Platform (GCP) cloud provider. Key components have been refactored to handle both AWS and GCP resources, selectable via an environment variable "CARIBOU_DEFAULT_PROVIDER".

• Closes Add Support for Google Cloud Platform #356
• Closes Add tests for zone finder and gcp latency retriever #351
• Closes Google Maps doesn't know where US East (N. Virginia) is #358
• Opens GCP Latency only reports median figures #359
• Opens Check if thread local hops overwrite the hops count #360

Key Changes & Features:

• Multi-Cloud RemoteClient Architecture:

  • The RemoteClientFactory has been refactored to use runtime imports. This prevents cross-cloud dependency issues (e.g., trying to import boto3 in a GCP environment and importing Google Cloud modules in an AWS environment), resolving ModuleNotFoundErrors at startup.
  • A new GCPRemoteClient class has been implemented to provide the GCP equivalent for core framework operations.

• GCP Resource Management:

  • Compute: Deploys functions as Cloud Run services using Docker images stored in Artifact Registry.
  • IAM: Implements the "one service account per workflow" pattern for GCP. It creates a single, shared Service Account for all functions within a workflow to handle GCP's stricter naming and length constraints. GCP also has a default quota of 100 service accounts per user.
  • Messaging: Uses Pub/Sub topics and authenticated push subscriptions to connect workflow functions, mirroring the SNS functionality in AWS. Improved by Pub/Sub's max limit of 10MB, the messages are not compressed.
  • Scheduling: Uses Cloud Scheduler to trigger the remote CLI for automated tasks, providing an equivalent to AWS EventBridge rules.
  • Storage: Uses Firestore as the key-value store for state management (e.g., workflow plans, image URIs), equivalent to DynamoDB.

• Framework Tooling Updates:

  • Setup Scripts: GCP does not need a setup table script. The tables are generated automatically when the first values are stored there. The user only needs to create the default database.
  • Teardown Scripts: The teardown_tables.py scripts are now multi-cloud. They check the CARIBOU_DEFAULT_PROVIDER environment variable to run the appropriate logic for either DynamoDB or Firestore.
  • Remote CLI: The remote_cli.py and main cli.py have been updated with provider-aware wrapper functions (deploy_remote_framework, remove_timers, etc.) to manage the remote framework on both AWS and GCP.
  • Log Syncing: The LogSyncWorkflow has been refactored to parse both plain-text AWS CloudWatch logs and structured JSON logs from Google Cloud Logging. It now queries the Cloud Monitoring API to fetch performance metrics (CPU, memory) for Cloud Run, for an equivalent to Lambda Insights. The framework queries the logging API with a page size of 1000, and has an exponential backoff if the logging read quota is reached.
  • Migrator: Changed crane to gcrane for migrating images in GCP. Both the local and remote CLI can only use the Python solver for GCP.

• Data collection

  • Implemented a more fine-grained PUE values for both AWS and GCP. This value is gathered from the respective cloud providers' self-reported value.
  • Updated default values for PUE, Memory power consumption, and other metrics.

• Unit Tests:

  • Added test_gcp_remote_client.py
  • Extended existing test suites (e.g., test_log_sync_workflow.py, test_endpoints.py) to cover the new GCP code paths
  • Increased the test coverage to 83.35%

How to Use:

To target GCP for deployment or other commands, set the following environment variables before running the caribou CLI:

export CARIBOU_DEFAULT_PROVIDER=gcp

To switch the target back to AWS, use the following command:

export CARIBOU_DEFAULT_PROVIDER=aws

Further instructions are provided in the INSTALL.md file in the project's directory.

Remaining Work & Known Issues:
This is an initial implementation. Further testing on more workflows is needed.

[Danidite]

Overall it looks great, amazing work! However I have a few questions and comment that I think may be problematic or need to be addressed.

[Danidite]

Looks great, thanks!

Note: Pending demo, please do not squash and merge until the demo and any issues found are addressed.

[Danidite]

The new changes looks great, I do have a few very minor comments that may be a nice to have.

[engshahrad]

LGTM. Good job, @realchh , and thanks for the detailed review @Danidite . Merging now.