fix(runtime): harden multi-db rollout and metrics rollups#173
Merged
mergify[bot] merged 5 commits intomatrixorigin:mainfrom Apr 9, 2026
Merged
Conversation
There was a problem hiding this comment.
Pull request overview
This PR hardens the multi-database runtime rollout by introducing durable per-user schema version markers, tightening router/pool behavior, and reworking metrics/tooling paths to avoid expensive per-user fan-out on critical request paths.
Changes:
- Introduces/uses per-user schema version markers to skip heavy migrations on restart, and improves multi-db routing/pool initialization (global user pool, cache changes, init concurrency control).
- Reworks metrics aggregation to support multi-db via shared-db summaries + dirty marking, and lazily loads per-user tool usage in multi-db mode.
- Expands test coverage for multi-db routing/schema markers, search scoring fields, trust-tier validation, and branch behavior.
Reviewed changes
Copilot reviewed 41 out of 43 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| memoria/crates/memoria-storage/tests/store_crud.rs | Adds tests around vector retrieval score propagation and hybrid ranking behavior. |
| memoria/crates/memoria-storage/tests/router_multi_db.rs | Adds coverage for user schema marker persistence/repair in multi-db routing. |
| memoria/crates/memoria-storage/tests/branch_ops.rs | Adds routed-store branch fallback test ensuring “main” qualification behavior. |
| memoria/crates/memoria-storage/src/router.rs | Refactors router to use sync moka caches, adds global user pool, schema init cache + semaphore, and registry registration helper. |
| memoria/crates/memoria-storage/src/migration.rs | Adds configurable migration concurrency and adjusts per-user migration to use lightweight pools + registry registration. |
| memoria/crates/memoria-storage/src/lib.rs | Re-exports new pool health types. |
| memoria/crates/memoria-storage/src/graph/store.rs | Switches graph store to qualified-table mode (db_name + t()), sync moka cache, and updates SQL accordingly. |
| memoria/crates/memoria-storage/src/graph/retriever.rs | Removes access_count frequency boost from activation scoring. |
| memoria/crates/memoria-storage/Cargo.toml | Adds futures dependency; enables moka sync feature. |
| memoria/crates/memoria-service/src/service.rs | Updates background flushers and entity/graph pool behavior for multi-db; switches to routed store constructors where needed. |
| memoria/crates/memoria-service/src/scoring.rs | Adjusts scoring store impl to clone store before async calls. |
| memoria/crates/memoria-service/src/scheduler.rs | Makes governance pool sizing configurable and multi-db aware. |
| memoria/crates/memoria-service/src/governance.rs | Refactors governance to iterate routed stores without caching per-user pools; queries registry directly in multi-db. |
| memoria/crates/memoria-service/Cargo.toml | Enables moka sync feature. |
| memoria/crates/memoria-mcp/tests/tools_unit.rs | Validates tool schema docs for trust tier guidance and enum values. |
| memoria/crates/memoria-mcp/tests/perf_optimizations_e2e.rs | Updates near-duplicate behavior expectations (no cross-type duplicate). |
| memoria/crates/memoria-mcp/tests/core_tools_e2e.rs | Adds invalid trust_tier rejection test and capabilities hints assertions. |
| memoria/crates/memoria-mcp/tests/branch_e2e.rs | Adds assertion that stored active branch resets to main after delete. |
| memoria/crates/memoria-mcp/src/tools.rs | Adds tool-name enum dispatch, improves tool descriptions/schema, enforces trust_tier parsing errors, and adds call_owned. |
| memoria/crates/memoria-mcp/src/server.rs | Refactors JSON-RPC dispatching with owned variants and method enum; adjusts HTTP dispatch plumbing. |
| memoria/crates/memoria-mcp/src/remote.rs | Reuses shared capabilities text and adds call_owned. |
| memoria/crates/memoria-mcp/src/git_tools.rs | Adds tool-name enum dispatch; multi-db qualification fixes; implements SQL-based diff/merge helpers and branch list/active behavior improvements. |
| memoria/crates/memoria-git/src/service.rs | Qualifies branch/snapshot DDL and restore operations with db name to ensure correct database scoping. |
| memoria/crates/memoria-core/src/types.rs | Adds canonical MemoryType::ALL_NAMES and FEEDBACK_SIGNALS, plus drift-prevention tests. |
| memoria/crates/memoria-core/src/lib.rs | Re-exports FEEDBACK_SIGNALS. |
| memoria/crates/memoria-cli/src/main.rs | Gates server commands behind feature, adds migration concurrency flag, and introduces configurable git pool init for multi-db. |
| memoria/crates/memoria-cli/Cargo.toml | Adds server-runtime feature and makes API/MCP optional deps; updates default features. |
| memoria/crates/memoria-api/tests/api_e2e.rs | Adds e2e validation that MCP tools/call records tool usage. |
| memoria/crates/memoria-api/src/state.rs | Replaces moka api key cache with custom TTL cache; adds metrics summary manager init and dirty marking hook. |
| memoria/crates/memoria-api/src/routes/snapshots.rs | Uses qualified table names and switches branch “active” detection to branch name. |
| memoria/crates/memoria-api/src/routes/sessions.rs | Uses unquoted/qualified table formatting for session summary queries. |
| memoria/crates/memoria-api/src/routes/metrics.rs | Switches business metrics to summary-backed approach (multi-db) with health metrics; retains cached scrape output. |
| memoria/crates/memoria-api/src/routes/memory.rs | Adds lazy per-user tool usage DB load on empty cache; adjusts some table formatting. |
| memoria/crates/memoria-api/src/routes/mcp.rs | Tracks tool usage, marks metrics dirty after mutating MCP calls, and updates dispatch to owned http entrypoint. |
| memoria/crates/memoria-api/src/routes/auth.rs | Updates api key cache invalidation calls for new cache type. |
| memoria/crates/memoria-api/src/routes/admin.rs | Qualifies tables per user store, marks metrics dirty on delete_user, and fixes call-log table routing. |
| memoria/crates/memoria-api/src/lib.rs | Adds metrics summary module and marks summaries dirty from v1 mutating routes; adjusts call logging behavior. |
| memoria/crates/memoria-api/src/auth.rs | Adds lazy per-user tool usage load, table qualification for usage/call log flushes, and rebuild merge semantics. |
| memoria/crates/memoria-api/Cargo.toml | Enables moka sync feature. |
| memoria/Cargo.lock | Adds futures dependency entry. |
| docs/per-user-database-architecture.md | Updates architecture doc to “implemented” state with deployment/migration runbook and rationale. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
1. Atomic claim_batch: replace SELECT+UPDATE race with UPDATE-then- SELECT-by-token pattern. MySQL single-UPDATE atomicity prevents duplicate claims across concurrent workers. 2. Per-family DELETE scope in flush_rollups: compute per-family user sets from claimed_mask instead of using all_user_ids. Users who triggered a family but produce zero rows still get old rows deleted. Users who did NOT trigger a family are untouched. 3. Write actual refreshed_version: carry ClaimedUser.claimed_version through the by_family grouping and bind it in INSERT instead of 0u64. 4. Batch flush_state: replace per-user UPDATE loop with chunked CASE- based batch UPDATE (50 users/statement). MySQL left-to-right SET evaluation lets has_pending derive from the post-update pending_mask. Version-protected CAS semantics preserved. Added 8 regression tests covering per-family deletion scope, zero-entry user cleanup, FULL mask coverage, cross-family leak prevention, and refreshed_version propagation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Persist per-user schema versions to avoid rerunning compat migrations after restart, tighten routed pool defaults and startup wiring, and finish the API/metrics rollup updates that support the capped multi-db runtime validation path. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
gouhongshen
force-pushed
the
fix/runtime-multi-db-rollout-and-rollups
branch
from
1547ed7 to
df71466
Compare
Fresh databases skip apply_user_compat_migrations() after the schema-version marker change, so indexes that were previously added only via compat migrations must also appear in the bootstrap CREATE TABLE statements. Adds to bootstrap_user_schema(): - idx_memories_user_observed on mem_memories - idx_feedback_memory_user on mem_retrieval_feedback - idx_feedback_created_at on mem_retrieval_feedback Fixes CI: test_feedback_created_at_index_exists Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
buffer_unordered() does not preserve input order, so zipping collected results with selected_users could misassociate reports. Return (user_id, result) tuples from the stream instead. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
XuPeng-SH
approved these changes
Apr 9, 2026
Merge Queue Status
This pull request spent 26 minutes 40 seconds in the queue, including 26 minutes 29 seconds running CI. Required conditions to merge
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
mem_schema_metaso restart no longer re-runs heavy compat migrations for already-marked user DBsValidation
cargo test -p memoria-storage --test router_multi_db -- --test-threads=1cargo check -qseed_rest_store p95 ~81.47s,p99 ~82.09s,0errorsseed_rest_store p95 ~10.50s,p99 ~10.69s,3/403timeout-shaped500s remainingNotes