feat(phase3b3): GCP worker fleet — GcsStateStore + CloudRunLauncher#22
Open
dcvz wants to merge 12 commits into
Open
feat(phase3b3): GCP worker fleet — GcsStateStore + CloudRunLauncher#22dcvz wants to merge 12 commits into
dcvz wants to merge 12 commits into
Conversation
…cher) Mirror of the AWS pair behind the existing StateStore/Launcher traits, using the official google-cloud-storage + google-cloud-run-v2 SDKs. cica-side primitives only; the GCP IaC is a separate track. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add StoreKind::Gcs and ProviderKind::CloudRun enum variants, plus GcsConfig and CloudRunConfig structs wired into DeploymentConfig. Feature stubs (gcs, cloudrun) added to Cargo.toml; match arms in sandbox dispatch forward to feature-gated modules (Tasks 4/6). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Mirrors S3StateStore over GCS objects (identical <prefix>/<key>/<rel> key scheme). Uses google-cloud-storage's Storage (read/write bytes) + StorageControl (list/delete) clients, built lazily so default_store stays sync and network-free. Resumable uploads are automatic — no manual multipart. When an endpoint override is set (emulator/testing) the clients use anonymous credentials (google-cloud-auth, default-features off to dodge the aws-lc-rs ^1.17 vs rustls-locked 1.15 conflict); real GCS keeps ADC. The gated IT (CICA_GCS_IT) auto-skips without the env. NOTE: StorageControl is gRPC-only and fake-gcs-server serves HTTP/JSON only, so the emulator can't exercise list/delete — the IT is correct for real GCS; CI (Task 5) must use a real bucket, not fake-gcs-server. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…REST-only, SDK control-plane is gRPC) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds the SDK-agnostic Cloud Run launcher: CloudRunClient trait, RunJobRequest/ExecutionStatus types, CloudRunLauncher with launch/poll/cancel state machine, and 5 fake-driven unit tests. GcpRunClient is a temporary stub (Task 7 replaces it). Default build is unaffected. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replaces the "not yet implemented" placeholder with the real feature-gated arm (mirrors Fargate): requires a store + [deployment.cloudrun] section when built with --features cloudrun, errors with a clear message otherwise. Adds cloudrun_provider_requires_feature and cloudrun_provider_requires_a_store tests; removes all #[allow(dead_code)] from cloudrun.rs now that the launcher is constructed in production code. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Extend the `cloud` umbrella feature to include both `fargate` (AWS) and `cloudrun` (GCP), so release binaries tagged `-cloud` carry both clouds automatically. Add a `cloud-features` CI job that lints gcs/cloudrun/cloud and runs the full unit-test suite (gated GCS IT auto-skips without CICA_GCS_IT). Also apply `cargo fmt` fixes in cloudrun.rs and sandbox/mod.rs. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Extract the inline task-count predicate in GcpRunClient::get_execution into a pure free function `execution_succeeded` and cover all boundary cases with a unit test, so the decision has coverage without live GCP. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Phase 3b-3 of the distributed deployment: makes the cica worker fleet run on GCP as a self-contained second deployment, mirroring the existing AWS pair behind the unchanged
StateStore/Launchertraits. cica-side primitives only — the GCP IaC is a separate track.GcsStateStore(featuregcs,src/sandbox/state/gcs.rs) — GCS-backedStateStoreover the officialgoogle-cloud-storageSDK. Same<prefix>/<key>/<rel>key scheme, replace semantics, and..traversal guard asS3StateStore; ADC auth; automatic resumable uploads (no hand-rolled multipart). Uses bothStorage(bytes) andStorageControl(list/delete) clients.CloudRunLauncher+GcpRunClient(featurecloudrun,src/sandbox/cloudrun.rs) — runs a worker turn as a one-shot Cloud Run Job execution viagoogle-cloud-run-v2. AWS calls sit behind a 3-methodCloudRunClienttrait so the launch→poll→timeout→best-effort-cancel state machine is fake-tested.RunJobreturns the execution name from the poller's first poll without awaiting the LRO, then we pollGetExecutionon our own cadence (symmetric withFargateLauncher).StoreKind::Gcs+ProviderKind::CloudRun;GcsConfig(no region — GCS is globally addressed) andCloudRunConfig(no networking fields — Cloud Run pins VPC/egress on the Job resource;RunJobhas no per-run network knob). Env overlayCICA_STORE=gcs/CICA_GCS_BUCKET.prep_skill_deps_locallytreats CloudRun as a remote worker.cloud = ["fargate", "cloudrun"];release.ymlalready builds--features cloud, so the published worker image now carries AWS + GCP with no image change (entrypoint is alreadycica). New CIcloud-featureslane lints + unit-tests the GCP features.Notable decision
The live
GcsStateStoreintegration test runs against real GCS, deferred to the deploy/IaC track — exactly like Fargate's first realRunTask.fake-gcs-serverwas infeasible (it's REST-only; the SDK's list/delete control-plane is gRPC-only) andstorage-testbench's gRPC-in-Docker was too flaky for CI. The IT code is written and auto-skips withoutCICA_GCS_IT.Two behaviors are only confirmable against live GCP (inherent, deferred): RunJob's first poll carrying the execution name, and the count-based success decision against real executions.
Test plan
cargo test(default) — 105 passing;cargo test --features cloud— 128 passingcargo clippy --all-targets -- -D warningsand--features cloud— cleancargo fmt -- --check— cleancargo treeverified)RunJob— on the GCP-IaC/deploy track (real ADC + bucket + Cloud Run Job)Spec:
docs/superpowers/specs/2026-06-09-phase3b3-gcp-worker-fleet-design.md· Plan:docs/superpowers/plans/2026-06-09-phase3b3-gcp-worker-fleet.md🤖 Generated with Claude Code