Add RetryPolicy CRUD API and armadactl support

[dejanzele]

What type of PR is this?

Feature (retry policy PR 2 of 4)

What this PR does / why we need it

Adds RetryPolicy as a first-class API resource with full CRUD operations, following the same patterns as Queue. Operators can create retry policies and assign them to queues by name.

• Adds RetryPolicy, RetryRule, RetryExitCodeMatcher proto messages and RetryAction enum to submit.proto
• Adds RetryPolicyService gRPC service with Create/Update/Delete/Get/List RPCs
• Adds REST gateway bindings for all operations (/v1/retry-policy/...)
• Adds retry_policy field (string, references policy by name) to the Queue proto message
• Adds internal/server/retrypolicy/ - repository (PostgreSQL) and gRPC handler, mirroring internal/server/queue/
• Adds pkg/client/retrypolicy/ - client library for all CRUD operations
• Adds cmd/armadactl/cmd/retrypolicy.go - CLI commands for managing retry policies
• Adds --retry-policy flag to queue create/update commands
• Adds RetryPolicy resource kind for file-based creation (armadactl create -f policy.yaml)

Usage:

armadactl create retry-policy -f policy.yaml
armadactl get retry-policy ml-training
armadactl get retry-policies
armadactl delete retry-policy ml-training
armadactl create queue ml-queue --retry-policy ml-training

Which issue(s) this PR fixes

Part of #4683 (Retry Policy)

Special notes for your reviewer

• This is retry policy PR 2 of 4: Engine + config (independent) and CRUD + armadactl (this) -> Scheduler wiring -> Backoff + pod naming
• No dependency on error categorization PRs - this is pure CRUD infrastructure
• Independent of the engine PR (PR 1) - these can be reviewed in parallel
• Follows the queue CRUD pattern exactly: proto -> server handler -> PostgreSQL repository -> client library -> armadactl
• Repository uses INSERT with unique constraint check (not read-then-write) for concurrent-safe creates
• Update uses single UPDATE with RowsAffected check instead of read-then-upsert
• The retry_policy table (name text PK, definition bytea) stores serialized proto, same as the queue table
• File-based create/update because rules are too complex for CLI flags
• No scheduling behavior change in this PR

[greptile-apps]

Greptile Summary

This PR adds a complete RetryPolicy CRUD API — proto messages, gRPC service, PostgreSQL repository, REST gateway bindings, client library, and armadactl commands — following the established Queue pattern throughout. All prior review concerns have been resolved: the 032_create_retry_policy.sql migration is present, permissions are distinct for create/update/delete, idempotent delete and unauthenticated reads are documented, and ExitCodeOperator is properly an enum.

Confidence Score: 5/5

This PR is safe to merge; all previously flagged issues have been addressed and the implementation follows established patterns.

All prior P0/P1 concerns (missing migration, wrong gRPC codes, permission asymmetry, free-form operator string, missing name validation) are fully resolved. The only remaining finding is a P2 cosmetic issue with null output for an empty policy list.

No files require special attention; internal/armadactl/retrypolicy.go has a minor P2 cosmetic issue with null output for an empty list.

Important Files Changed

Filename	Overview
internal/server/retrypolicy/service.go	New gRPC service handler; auth, validation, and error mapping are correct and consistent with queue service
internal/server/retrypolicy/repository.go	PostgreSQL CRUD repository; concurrent-safe create via ON CONFLICT DO NOTHING, intentionally idempotent delete is documented
internal/server/retrypolicy/service_test.go	Comprehensive unit tests covering all code paths: auth failures, empty name, not-found, already-exists, and success cases
internal/lookout/schema/migrations/032_create_retry_policy.sql	Correctly numbered migration (following 031) creating retry_policy table matching queue table pattern
pkg/api/submit.proto	Adds RetryPolicy messages, RetryPolicyService, and REST gateway bindings; ExitCodeOperator is properly modelled as an enum
internal/armadactl/retrypolicy.go	Clean CLI integration; GetAllRetryPolicies outputs 'null' for empty result instead of a user-friendly empty message
cmd/armadactl/cmd/retrypolicy.go	CLI command definitions follow established patterns; file-flag required for create/update, positional arg for get/delete
internal/server/submit/submit.go	Adds retryPolicyService field and five delegation methods, mirroring the existing queue service delegation pattern
internal/server/server.go	Correct wiring: PostgresRetryPolicyRepository → retrypolicy.Server → both Submit and RetryPolicyService gRPC registrations
internal/server/permissions/permissions.go	Three distinct permission constants added: CreateRetryPolicy, UpdateRetryPolicy, DeleteRetryPolicy
pkg/client/retrypolicy/create.go	Standard client library create implementation using common connection/context patterns
pkg/client/retrypolicy/get.go	Standard client library get/list implementation; both single and bulk fetch covered
pkg/client/retrypolicy/update.go	Standard client library update implementation
pkg/client/retrypolicy/delete.go	Standard client library delete implementation
pkg/client/resource.go	RetryPolicy added to NewResourceKind switch correctly alongside Queue
internal/armadactl/queue.go	CreateResource extended with ResourceKindRetryPolicy case, following the existing Queue pattern

Sequence Diagram

sequenceDiagram
    participant CLI as armadactl
    participant Client as pkg/client/retrypolicy
    participant Submit as Submit gRPC Server
    participant Service as retrypolicy.Server
    participant Repo as PostgresRetryPolicyRepository
    participant DB as PostgreSQL

    CLI->>Client: Create/Update/Delete/Get(policy)
    Client->>Submit: gRPC or REST HTTP
    Submit->>Service: delegate (e.g. CreateRetryPolicy)
    Service->>Service: AuthorizeAction(ctx, permission)
    Service->>Service: validate req.Name != ""
    Service->>Repo: CreateRetryPolicy(ctx, policy)
    Repo->>Repo: proto.Marshal(policy)
    Repo->>DB: INSERT INTO retry_policy ON CONFLICT DO NOTHING
    DB-->>Repo: RowsAffected
    Repo-->>Service: nil or ErrAlreadyExists
    Service-->>Submit: types.Empty{} or gRPC status error
    Submit-->>Client: response
    Client-->>CLI: result

Loading

_{Reviews (10): Last reviewed commit: "Add RetryPolicy CRUD API and armadactl s..." | Re-trigger Greptile}

[dejanzele]

@greptileai

[dejanzele]

@greptileai