feat(memory): #1792 Phase A3 — drop agent_memories table + legacy v1 surface #1809

2026-05-24T21:52:10Z

hongming commented

2026-05-24 21:52:10 +00:00

Phase A3 — drop agent_memories table + legacy v1 surface

Closes #1792. The v1→v2 memory migration ends here.

Per the audit before this PR (Phase A2 completion verification 2026-05-24):

Tenant	v1 (frozen)	v2 (live)	Status
agents-team	1805	1805+live	parity
hongming	144	144	parity
chloe-dong	1	1	parity
reno-stars	102	102	parity

Total 2,052 rows mirrored, zero errors, v1 frozen, v2 receiving live writes.

What this drops

agent_memories table (DROP TABLE migration)
MemoriesHandler.Search/Update/Delete methods + their /memories HTTP routes
Helper functions only used by the removed methods (EmbeddingFunc, embed field, WithEmbedding, formatVector, nextArg, memoryFTSMinQueryLen, memoryRecallMaxLimit)
agent_memories UNION branch from activity.go's session-search query
agent_memories from workspace_crud.go's purge cleanup list
MEMORY_V2_CUTOVER deprecation shim from entrypoint-tenant.sh
All tests that exercised the deleted code

Net deletion: -1311 LOC across 8 files. Net add: 91 LOC (the migration + revised comments).

What stays

POST /workspaces/:id/memories — high-volume write surface, routes through v2 plugin via #1794
All /v2/memories endpoints — canvas already uses these
All MCP memory tools — they were v2-only since #1747
cmd/memory-backfill binary — bundled in image but now inert (its source table is gone); kept for one image cycle then removable in a cleanup PR
MEMORY_V2_CUTOVER='true' in controlplane ec2.go — no-op now; should be cleaned up in a separate molecule-controlplane PR

Risk

Hard 404 on any caller still hitting GET / PATCH / DELETE /workspaces/:id/memories. Production traffic audit (2026-05-24) showed 2 GETs vs 66 POSTs over 24h — runtime callers are POST-dominant; canvas reads from /v2/memories. Acceptable.
No DB rollback restores data — A2 was one-way. If a critical bug appears post-merge, recover via memory_plugin.memory_records direct SQL (data is preserved there).

SOP Checklist (RFC #351)

1. Comprehensive testing performed

go test -short -count=1 ./internal/handlers/ green (18s, 0 regressions)
go test -short -count=1 ./cmd/memory-backfill/ green (sqlmock-based, still passes)
go vet ./... clean

2. Local-postgres E2E run

N/A. Single-axis schema change verified against the well-tested migration tool shape.

3. Staging-smoke verified or pending

Pending merge + tenant recycle. Will verify by SSM-checking:

agent_memories table gone from each tenant's DB
POST /memories still returns 201
Rows continue to land in memory_plugin.memory_records
Workspace deletion still succeeds end-to-end

4. Root-cause not symptom

Yes. v1 existed only as a dual-write target during the A1+A2 transition. With A2 complete and parity verified, the table is dead weight + SSOT-violation surface. Dropping it removes the surface entirely.

5. Five-Axis review walked

Walked solo. Happy to dispatch a hostile reviewer pre-merge.

6. No backwards-compat shim / dead code added

Net deletion: -1311 LOC. The MEMORY_V2_CUTOVER shim was the last backwards-compat hook — it's gone too. Documented one follow-up (controlplane env-var cleanup) in PR body.

7. Memory/saved-feedback consulted

feedback_no_single_source_of_truth — A3 is the final step establishing v2 as the only memory backend.
feedback_check_for_parallel_work_before_fix_pr — grep'd recent PRs touching memories.go / activity.go / workspace_crud.go; no parallel in flight.

🤖 Generated with Claude Code

## Phase A3 — drop agent_memories table + legacy v1 surface Closes #1792. The v1→v2 memory migration ends here. Per the audit before this PR (Phase A2 completion verification 2026-05-24): | Tenant | v1 (frozen) | v2 (live) | Status | |---|---|---|---| | agents-team | 1805 | 1805+live | parity | | hongming | 144 | 144 | parity | | chloe-dong | 1 | 1 | parity | | reno-stars | 102 | 102 | parity | Total 2,052 rows mirrored, zero errors, v1 frozen, v2 receiving live writes. ## What this drops - `agent_memories` table (DROP TABLE migration) - `MemoriesHandler.Search/Update/Delete` methods + their `/memories` HTTP routes - Helper functions only used by the removed methods (`EmbeddingFunc`, `embed` field, `WithEmbedding`, `formatVector`, `nextArg`, `memoryFTSMinQueryLen`, `memoryRecallMaxLimit`) - `agent_memories` UNION branch from `activity.go`'s session-search query - `agent_memories` from `workspace_crud.go`'s purge cleanup list - `MEMORY_V2_CUTOVER` deprecation shim from `entrypoint-tenant.sh` - All tests that exercised the deleted code Net deletion: **-1311 LOC across 8 files**. Net add: 91 LOC (the migration + revised comments). ## What stays - `POST /workspaces/:id/memories` — high-volume write surface, routes through v2 plugin via #1794 - All `/v2/memories` endpoints — canvas already uses these - All MCP memory tools — they were v2-only since #1747 - `cmd/memory-backfill` binary — bundled in image but now inert (its source table is gone); kept for one image cycle then removable in a cleanup PR - `MEMORY_V2_CUTOVER='true'` in controlplane `ec2.go` — no-op now; should be cleaned up in a separate molecule-controlplane PR ## Risk - Hard 404 on any caller still hitting GET / PATCH / DELETE `/workspaces/:id/memories`. Production traffic audit (2026-05-24) showed 2 GETs vs 66 POSTs over 24h — runtime callers are POST-dominant; canvas reads from `/v2/memories`. Acceptable. - No DB rollback restores data — A2 was one-way. If a critical bug appears post-merge, recover via `memory_plugin.memory_records` direct SQL (data is preserved there). ## SOP Checklist (RFC #351) ### 1. Comprehensive testing performed - `go test -short -count=1 ./internal/handlers/` green (18s, 0 regressions) - `go test -short -count=1 ./cmd/memory-backfill/` green (sqlmock-based, still passes) - `go vet ./...` clean ### 2. Local-postgres E2E run N/A. Single-axis schema change verified against the well-tested migration tool shape. ### 3. Staging-smoke verified or pending Pending merge + tenant recycle. Will verify by SSM-checking: - `agent_memories` table gone from each tenant's DB - POST /memories still returns 201 - Rows continue to land in `memory_plugin.memory_records` - Workspace deletion still succeeds end-to-end ### 4. Root-cause not symptom Yes. v1 existed only as a dual-write target during the A1+A2 transition. With A2 complete and parity verified, the table is dead weight + SSOT-violation surface. Dropping it removes the surface entirely. ### 5. Five-Axis review walked Walked solo. Happy to dispatch a hostile reviewer pre-merge. ### 6. No backwards-compat shim / dead code added Net deletion: -1311 LOC. The MEMORY_V2_CUTOVER shim was the last backwards-compat hook — it's gone too. Documented one follow-up (controlplane env-var cleanup) in PR body. ### 7. Memory/saved-feedback consulted - `feedback_no_single_source_of_truth` — A3 is the final step establishing v2 as the only memory backend. - `feedback_check_for_parallel_work_before_fix_pr` — grep'd recent PRs touching memories.go / activity.go / workspace_crud.go; no parallel in flight. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

devops-engineer approved these changes 2026-05-24 21:52:24 +00:00

Dismissed

devops-engineer left a comment

Approving PR #1809: Phase A3 final cutover. Parity verified pre-PR (4 tenants exact match). Net -1311 LOC, clean dead-code removal. CTO-bypass 2026-05-24.

core-devops approved these changes 2026-05-24 21:52:25 +00:00

Dismissed

core-devops left a comment

Approving PR #1809: Phase A3 final cutover. Parity verified pre-PR (4 tenants exact match). Net -1311 LOC, clean dead-code removal. CTO-bypass 2026-05-24.

app-fe force-pushed feat/issue-1792-phase-a3-drop-agent-memories from 9417a7ad64 to 94bdd8ff35

2026-05-24 22:03:58 +00:00

Compare

app-fe dismissed devops-engineer's review 2026-05-24 22:03:59 +00:00

Reason:

New commits pushed, approval review dismissed automatically according to repository settings

app-fe dismissed core-devops's review 2026-05-24 22:03:59 +00:00

Reason:

New commits pushed, approval review dismissed automatically according to repository settings

devops-engineer approved these changes 2026-05-24 22:04:23 +00:00

devops-engineer left a comment

Re-approving PR #1809 after force-push fix (removed unused globalMemoryDelimiter const that linter caught). Same scope. CTO-bypass 2026-05-24.