feat(M018/S01): GateEvalContext persistence — resume-on-miss for gate_run #3

Merged

wollax merged 3 commits from kata/root/M018/S01 into main 2026-04-01 13:09:17 +00:00

Conversation 1 Commits 3 Files changed 2 +531 -10

wollax commented 2026-04-01 12:55:53 +00:00

Owner

Copy link

Completes the write-through cache pattern for GateEvalContext so gate evaluations survive MCP server restarts.

Changes

T01 (crates/assay-core/src/gate/session.rs):

• pub fn find_context_for_spec(assay_dir, spec_name) -> Result<Option<GateEvalContext>>
• Scans .assay/gate_sessions/*.json in reverse chronological order, returns most recent match
• Skips unreadable files with tracing::warn! (never propagates individual errors)
• 4 unit tests: match, no-match, empty-dir, most-recent-wins

T02 (crates/assay-mcp/src/server.rs):

• Wired find_context_for_spec into gate_run handler before create_session()
• On resume: replaces command_results with fresh results, re-persists, inserts into HashMap, spawns timeout task
• On disk scan failure: logs warn, falls through to create_session() defensively
• Contract test test_gate_run_resumes_session_from_disk proves the full restart-resume cycle

Verification

• cargo test -p assay-core -- find_context_for_spec: 4 ✅
• cargo test -p assay-core -- save_and_load: 3 ✅
• cargo test -p assay-mcp -- gate_run_resumes: 1 ✅
• cargo test -p assay-mcp -- gate_run: 16 ✅
• just ready (1579 tests): ✅

Closes R100 — GateEvalContext persistence to disk

Completes the write-through cache pattern for GateEvalContext so gate evaluations survive MCP server restarts. ## Changes **T01** (`crates/assay-core/src/gate/session.rs`): - `pub fn find_context_for_spec(assay_dir, spec_name) -> Result<Option<GateEvalContext>>` - Scans `.assay/gate_sessions/*.json` in reverse chronological order, returns most recent match - Skips unreadable files with `tracing::warn!` (never propagates individual errors) - 4 unit tests: match, no-match, empty-dir, most-recent-wins **T02** (`crates/assay-mcp/src/server.rs`): - Wired `find_context_for_spec` into `gate_run` handler before `create_session()` - On resume: replaces `command_results` with fresh results, re-persists, inserts into HashMap, spawns timeout task - On disk scan failure: logs warn, falls through to `create_session()` defensively - Contract test `test_gate_run_resumes_session_from_disk` proves the full restart-resume cycle ## Verification - `cargo test -p assay-core -- find_context_for_spec`: 4 ✅ - `cargo test -p assay-core -- save_and_load`: 3 ✅ - `cargo test -p assay-mcp -- gate_run_resumes`: 1 ✅ - `cargo test -p assay-mcp -- gate_run`: 16 ✅ - `just ready` (1579 tests): ✅ Closes R100 — GateEvalContext persistence to disk

wollax added 2 commits 2026-04-01 12:55:53 +00:00

feat(S01/T01): add find_context_for_spec to assay-core with unit tests 60dc6a0186

feat(S01/T02): wire resume-on-miss into gate_run handler with contract test 60b56c128b

wollax added 1 commit 2026-04-01 13:08:15 +00:00

fix(S01): address review findings — in-memory-first lookup, pending_criteria filtering, regression tests ... a2c32c47ca

- C2: Check in-memory HashMap before scanning disk. Disk scan is now only triggered
  on actual miss (cold start / post-restart), not on every gate_run. This prevents
  a live in-memory session from being silently replaced by its stale on-disk snapshot,
  which would discard accumulated agent_evaluations.

- I2: pending_criteria now filters out already-evaluated criteria on resume (same logic
  as gate_report). New sessions are unaffected (no evaluations yet).

- I3: Removed the redundant re-persist inside the disk-resume arm. A single write-through
  at the end of the handler is sufficient.

- Tracing: renamed 'spec' field to 'spec_name' for consistency with other handlers.

- Tests added:
  - test_gate_run_creates_new_session_after_finalize: gate_run after gate_finalize
    must create a fresh session, not resume the finalized one (disk file deleted).
  - test_gate_run_preserves_in_memory_session_evaluations: live in-memory session
    must not be bypassed by disk scan (agent_evaluations preserved across gate_run).
  - Updated existing resume test to also assert pending_criteria excludes
    already-evaluated criteria.

wollax commented 2026-04-01 13:09:02 +00:00

Author

Owner

Copy link

Review summary (multi-agent, 3 reviewers)

Reviewed by: correctness, API/observability, performance reviewers in parallel.

Critical findings addressed in this PR:

• C2 (critical): In-memory HashMap was bypassed on every gate_run — disk scanned unconditionally even for live sessions, which would discard accumulated agent_evaluations. Fixed: check HashMap first; disk scan only on actual miss (cold start / post-restart). See fix(S01) commit.
• I2 (important): pending_criteria returned stale full list even on resume. Fixed: filter out already-evaluated criteria, same logic as gate_report.
• I3 (important): Redundant double-write in resume path. Fixed: one write-through at end.
• I5 (important): No test for gate_run after gate_finalize. Fixed: test_gate_run_creates_new_session_after_finalize.
• Added test_gate_run_preserves_in_memory_session_evaluations proving the HashMap-first path.

False positive from review (C1): Reviewers flagged finalized sessions being re-opened. gate_finalize already deletes the disk file — confirmed in source. Not a bug.

Backlogged (minor, not merge blockers):

• WOL-164: test for corrupted session file being skipped
• WOL-165: gate_sessions unbounded growth — add eviction
• WOL-166: normalize spec_name tracing field across handlers

Final state: 1582 tests, just ready green, all critical review findings addressed.

## Review summary (multi-agent, 3 reviewers) **Reviewed by:** correctness, API/observability, performance reviewers in parallel. **Critical findings addressed in this PR:** - **C2 (critical):** In-memory HashMap was bypassed on every `gate_run` — disk scanned unconditionally even for live sessions, which would discard accumulated `agent_evaluations`. Fixed: check HashMap first; disk scan only on actual miss (cold start / post-restart). See `fix(S01)` commit. - **I2 (important):** `pending_criteria` returned stale full list even on resume. Fixed: filter out already-evaluated criteria, same logic as `gate_report`. - **I3 (important):** Redundant double-write in resume path. Fixed: one write-through at end. - **I5 (important):** No test for gate_run after gate_finalize. Fixed: `test_gate_run_creates_new_session_after_finalize`. - Added `test_gate_run_preserves_in_memory_session_evaluations` proving the HashMap-first path. **False positive from review (C1):** Reviewers flagged finalized sessions being re-opened. `gate_finalize` already deletes the disk file — confirmed in source. Not a bug. **Backlogged (minor, not merge blockers):** - WOL-164: test for corrupted session file being skipped - WOL-165: gate_sessions unbounded growth — add eviction - WOL-166: normalize `spec_name` tracing field across handlers **Final state:** 1582 tests, `just ready` green, all critical review findings addressed.

wollax merged commit d6247fe3a1 into main 2026-04-01 13:09:17 +00:00

wollax referenced this pull request from a commit 2026-04-01 13:09:18 +00:00

feat(M018/S01): GateEvalContext persistence — resume-on-miss for gate_run (#3)

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

wollax/assay!3

No description provided.