Plan 01: Extract FetchFeedJob Retry Orchestrator Service

Objective

Extract ~60 lines of retry/circuit-breaker execution logic from FetchFeedJob into a dedicated Fetching::RetryOrchestrator service (S2). This makes retry logic independently testable and keeps the job shallow per engine conventions.

Context

• @.claude/skills/sm-job/SKILL.md -- Job conventions (shallow delegation, ID args, source_monitor_queue)
• @.claude/skills/sm-pipeline-stage/SKILL.md -- Pipeline architecture and error handling patterns
• @.claude/skills/tdd-cycle/SKILL.md -- TDD red-green-refactor workflow
• app/jobs/source_monitor/fetch_feed_job.rb (147 lines) contains handle_transient_error, enqueue_retry!, open_circuit!, reset_retry_state! methods that mix job concerns with domain logic
• lib/source_monitor/fetching/retry_policy.rb (90 lines) already makes retry decisions via Decision struct -- RetryOrchestrator will execute those decisions
• Existing Result pattern in FeedFetcher, ItemCreator, Enqueuer provides the model to follow

Tasks

Task 1: Write RetryOrchestrator tests (TDD red)

Create test/lib/source_monitor/fetching/retry_orchestrator_test.rb:

• Test call with retry decision: updates source state (fetch_retry_attempt, next_fetch_at, fetch_status), enqueues FetchFeedJob with wait
• Test call with circuit-open decision: updates source (fetch_circuit_opened_at, fetch_circuit_until, fetch_status to failed), does NOT enqueue retry
• Test call with exhausted decision (neither retry nor circuit): resets retry state on source, returns exhausted status
• Test atomic source updates use with_lock
• Use create_source! factory, mock RetryPolicy decision structs

Task 2: Implement RetryOrchestrator service

Create lib/source_monitor/fetching/retry_orchestrator.rb:

• Class SourceMonitor::Fetching::RetryOrchestrator
• Result = Struct.new(:status, :source, :error, :decision, keyword_init: true) with retry_enqueued?, circuit_opened?, exhausted? helpers
• .call(source:, error:, decision:, job_class: SourceMonitor::FetchFeedJob, now: Time.current) class method
• Extract enqueue_retry!, open_circuit!, reset_retry_state! logic from FetchFeedJob
• All source state updates wrapped in source.with_lock { source.reload; source.update!(...) }
• Add autoload declaration in lib/source_monitor.rb under the Fetching namespace

Task 3: Refactor FetchFeedJob to use RetryOrchestrator

Modify app/jobs/source_monitor/fetch_feed_job.rb:

• Remove handle_transient_error, enqueue_retry!, open_circuit!, reset_retry_state! methods
• In handle_transient_error call site, replace with:
  rubyCopy

  decision = RetryPolicy.new(source:, error:, now: Time.current).decision
  return raise error unless decision
  result = RetryOrchestrator.call(source:, error:, decision:)
  raise error if result.exhausted?
  

• Keep handle_concurrency_error in the job (it's concurrency-specific, not retry-policy)
• Target: job should be ~80 lines

Task 4: Update FetchFeedJob tests

Update test/jobs/source_monitor/fetch_feed_job_test.rb:

• Replace tests that directly tested retry state updates with tests that verify RetryOrchestrator is called
• Keep integration-level tests that verify end-to-end behavior (fetch failure -> source state updated)
• Ensure existing test coverage is maintained, not reduced

Task 5: Verify

• bin/rails test test/lib/source_monitor/fetching/retry_orchestrator_test.rb -- all pass
• bin/rails test test/jobs/source_monitor/fetch_feed_job_test.rb -- all pass
• bin/rails test -- full suite passes
• bin/rubocop -- zero offenses