Remove deprecated GetTaskLogs RPC

[rjpower]

🤖 Filed by agent.

Summary

The GetTaskLogs RPC on ControllerService is now a thin forwarding shim to FetchLogs. All internal callers have been migrated to use FetchLogs with LIKE patterns directly.

What to remove (after 2026-05-01)

1. Proto (lib/iris/src/iris/rpc/cluster.proto):

   • Controller.GetTaskLogsRequest message
   • Controller.TaskLogBatch message
   • Controller.GetTaskLogsResponse message
   • rpc GetTaskLogs(...) from ControllerService

2. Controller service (lib/iris/src/iris/cluster/controller/service.py):

   • get_task_logs forwarding method

3. Dashboard types (lib/iris/dashboard/src/types/rpc.ts):

   • Any remaining TaskLogBatch / GetTaskLogsResponse references

4. Regenerate proto/connect code after proto cleanup

Context

The FetchLogs RPC now supports SQL LIKE patterns in the source field (e.g. /user/job/%), making GetTaskLogs' job/task ID parsing redundant. The child_job_statuses field on GetTaskLogsResponse is no longer populated.

[github-actions]

🤖 I'm looking into this issue. I'll post a task list shortly with my plan and progress.

[claude]

🤖 Plan:

• Fetch issue information
• Research codebase for all relevant source files
• Run duplicate-work preflight (no overlapping PRs)
• Post research + proposed approach
• Implementation (deferred — see analysis below)

Research

The issue states that get_task_logs is "a thin forwarding shim to FetchLogs" and that "all internal callers have been migrated." Neither is accurate today. get_task_logs in service.py:1331-1422 is a ~90-line implementation with its own job/task ID parsing, TaskLogBatch grouping, child_job_statuses population, and include_children → shallow flag logic — none of which exists in fetch_logs (service.py:1549-1587). Active callers span 15+ files across Python, TypeScript, and proto layers.

Relevant code

• service.py#L1331-L1422 — full get_task_logs handler (not a shim)
• remote_client.py#L368-L406 — fetch_task_logs() wraps GetTaskLogs RPC
• LogViewer.vue#L41-L47 — dashboard uses GetTaskLogs for task logs
• protocol.py#L103-L114 — client protocol with GetTaskLogsResponse return type
• cluster.proto#L852-L882 — proto messages

Active callers (not yet migrated)

Layer	Files
Client	remote_client.py, protocol.py, client.py
Dashboard	LogViewer.vue, TaskDetail.vue, rpc.ts
CLI	bug_report.py
Tests	test_iris_integration.py, test_smoke.py, test_dashboard.py, conftest.py, test_local_client.py, _docker_cluster.py, cluster.py

Proposed Approach

This is not a simple deletion — it requires migrating all callers to FetchLogs first. The migration has three gaps to bridge:

1. Source key construction: GetTaskLogs accepts a job/task ID and builds the log store key internally. FetchLogs requires the caller to pass a source key (e.g., /job/name/task/0:0). Callers need a helper to construct source keys from JobName, or FetchLogs needs to accept LIKE patterns (which the issue mentions it supports).

2. child_job_statuses: GetTaskLogs returns child job statuses when include_children=True. The issue says this "is no longer populated" but the code still populates it (service.py:1352-1364). The dashboard LogViewer.vue doesn't appear to use this field, so removal may be safe, but this should be verified.

3. TaskLogBatch grouping: GetTaskLogs wraps entries in a TaskLogBatch with task_id. FetchLogs returns flat entries. Callers that iterate task_logs batches need updating.

Recommended migration order:

1. Add a fetch_task_logs method on remote_client.py that calls FetchLogs with LIKE source patterns
2. Update protocol.py return type to FetchLogsResponse
3. Migrate all Python callers (CLI, tests)
4. Update dashboard LogViewer.vue to use FetchLogs for task logs (similar to how process logs already work on line 50-55)
5. Remove proto messages, RPC, and service handler
6. Clean up rpc.ts types
7. Regenerate proto/connect code

Recommendation: The issue states removal should happen "after 2026-05-01." Given the scope (~15 files, Python + TypeScript, proto regeneration) and the fact that callers have NOT been migrated yet, this should be split into two PRs: (1) migrate all callers to FetchLogs, (2) delete dead proto/service code. Both can proceed now if desired, or wait until the stated date.

[rjpower]

@claude this is in now, go ahead and make a pr, tag me as a reviewer. The pr staleness warning will remind me to submit it.

[claude]

Claude finished @rjpower's task in 11m 11s —— View job • claude/issue-4201-20260328-1855 • [Create PR ➔](https://github.com/marin-community/marin/compare/main...claude/issue-4201-20260328-1855?quick_pull=1&title=Issue%20%234201%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%234201%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

Removing deprecated GetTaskLogs RPC

• Analyze current state of GetTaskLogs usage — all callers already migrated to FetchLogs
• Remove proto messages and RPC definition (GetTaskLogsRequest, TaskLogBatch, GetTaskLogsResponse, GetTaskLogs RPC)
• Remove get_task_logs from controller service + unused build_log_source import and DEFAULT_MAX_TOTAL_LINES
• Update docs references (vue-refactor.md)
• Run pre-commit and tests — all passing (1669 passed, 1 skipped)
• Create PR: Remove deprecated GetTaskLogs RPC #4231

Changes: 3 files changed, 7 insertions, 99 deletions. Dashboard types and test helpers were already clean — the test get_task_logs methods in _docker_cluster.py and conftest.py already use FetchLogs internally (just named get_task_logs as convenience wrappers).

Testing: ./infra/pre-commit.py --all-files --fix ✅ | test_dashboard.py 32 passed ✅ | Full iris unit tests 1669 passed ✅