Add RetryPolicy CRUD service and queue attachment validation

[dejanzele]

Implements the RetryPolicyService whose proto was defined in #4998: create, get, update, and delete for retry policies, backed by a Postgres repository and a lookout schema migration for the retry_policy table. Referential integrity is enforced in both directions. Deleting a policy is rejected while any queue still references it, and creating or updating a queue is rejected if it references a policy that does not exist. Adds create_retry_policy, update_retry_policy, and delete_retry_policy permissions.

Stacked on #5047, which moves a shared gRPC test helper into servertest. Its commit shows in this diff until it merges. The client package and armadactl commands are split into #5048.

[datadog-armadaproject]

⚠️ Warnings

🚦 1 Pipeline job failed

CI | test / Proto Up To Date     

Useful? React with 👍 / 👎

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: bd38227 | Docs | Give us feedback!}

[greptile-apps]

Greptile Summary

This PR implements the RetryPolicyService (CRUD) defined in #4998, backed by a new retry_policy Postgres table (migration 035) and a PostgresRetryPolicyRepository. It also adds validateRetryPolicy guards on CreateQueue / UpdateQueue so a queue cannot reference a non-existent policy, and consolidates the previously triply-duplicated requireGrpcCode test helper into the shared servertest package.

• New retrypolicy package: repository.go (Postgres CRUD + RetryPolicyExists), service.go (gRPC handlers with auth), validation.go (RFC-1123 name + rule checks), all covered by unit and integration tests.
• Queue service changes: RetryPolicyExistenceChecker interface injected into queue.Server; validateRetryPolicy called before the repository write on both create and update paths.
• pkg/client/queue: Queue.RetryPolicies []string added with correct NewQueue / ToAPI round-trip.

Confidence Score: 5/5

Safe to merge; the new service is well-structured, auth-gated, and has solid unit and integration test coverage.

The implementation follows established patterns in the codebase, the migration is consistent with existing schema, auth checks are in place for all mutating operations, and read paths are intentionally unauthenticated in line with GetQueue. The two findings are minor efficiency/consistency nits that do not affect correctness or security.

internal/server/retrypolicy/repository.go — nil vs empty slice on GetAllRetryPolicies and the full-fetch in RetryPolicyExists are worth a second look before broader rollout.

Important Files Changed

Filename	Overview
internal/server/retrypolicy/repository.go	New Postgres-backed CRUD repository for retry policies; mirrors the queue repository pattern but GetAllRetryPolicies returns nil (not empty slice) and RetryPolicyExists fetches the full definition rather than using a lightweight existence query.
internal/server/retrypolicy/service.go	New gRPC service wiring CRUD operations with auth and queue-reference checks; the acknowledged TOCTOU race on delete and the post-auth empty-name check are noted in prior threads.
internal/server/retrypolicy/validation.go	Clean RFC-1123 name validation and rule structural checks; well-covered by validation_test.go.
internal/server/queue/queue_service.go	Adds validateRetryPolicy guard on CreateQueue and UpdateQueue; wires the new RetryPolicyExistenceChecker interface correctly.
internal/lookout/schema/migrations/035_create_retry_policy.sql	New retry_policy table that exactly mirrors the existing queue table schema; format is consistent with all other migrations in this directory.
internal/server/server.go	Correctly wires retryPolicyRepo as both the RetryPolicyExistenceChecker for the queue server and the repository for the new retryPolicyServer; queueRepository satisfies QueueLister via GetAllQueues.
internal/server/servertest/grpc.go	Consolidates the previously duplicated requireGrpcCode helper into a shared package; uses require.Equal instead of assert.Equal for stricter test failure behaviour.
internal/server/queue/queue_service_test.go	Adds tests for unknown-policy rejection and multi-policy acceptance; PriorityFactor: 1 is correctly set to avoid early-exit from NewQueue; UpdateQueue unknown-policy path remains untested (noted in prior threads).
pkg/client/queue/queue.go	Adds RetryPolicies []string field to the Queue struct with correct round-trip through NewQueue and ToAPI.

Sequence Diagram

sequenceDiagram
    participant Client
    participant QueueServer
    participant RetryPolicyServer
    participant PostgresRetryPolicyRepo
    participant PostgresQueueRepo

    Note over Client,PostgresQueueRepo: CreateQueue / UpdateQueue flow
    Client->>QueueServer: CreateQueue(req)
    QueueServer->>QueueServer: authorize(CreateQueue)
    QueueServer->>QueueServer: queue.NewQueue(req)
    loop for each RetryPolicy name
        QueueServer->>PostgresRetryPolicyRepo: RetryPolicyExists(name)
        PostgresRetryPolicyRepo-->>QueueServer: (bool, err)
    end
    QueueServer->>PostgresQueueRepo: CreateQueue(queue)
    PostgresQueueRepo-->>QueueServer: ok
    QueueServer-->>Client: "Empty{}"

    Note over Client,PostgresQueueRepo: DeleteRetryPolicy flow
    Client->>RetryPolicyServer: DeleteRetryPolicy(req)
    RetryPolicyServer->>RetryPolicyServer: authorize(DeleteRetryPolicy)
    RetryPolicyServer->>PostgresQueueRepo: GetAllQueues()
    PostgresQueueRepo-->>RetryPolicyServer: []Queue
    RetryPolicyServer->>RetryPolicyServer: check referencing queues
    alt queues still reference policy
        RetryPolicyServer-->>Client: FailedPrecondition
    else no references
        RetryPolicyServer->>PostgresRetryPolicyRepo: DeleteRetryPolicy(name)
        PostgresRetryPolicyRepo-->>RetryPolicyServer: ok
        RetryPolicyServer-->>Client: "Empty{}"
    end

Loading

_{Reviews (32): Last reviewed commit: "Add RetryPolicy CRUD service and queue a..." | Re-trigger Greptile}

[dejanzele]

@greptileai

[dejanzele]

@greptileai

[dejanzele]

@greptileai

[dejanzele]

Closing in favour of the split: #5047 (shared gRPC test helper), #5049 (server-side service), and #5048 (client package and armadactl commands).