dev/MultiforaDB
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

engine: decompose the global lock; cross-connection group commit (roadmap item 5)

Browse Source 
The single engine-wide reader/writer lock is replaced by a lock hierarchy,
so writes to different collections no longer serialize on one mutex:

- Collections are heap-allocated, so their addresses are stable while a
  command holds a collection lock (the maps only store pointers).
- A catalog rwlock guards the database/collection maps: shared for every
  command (so a concurrent DDL cannot mutate the maps underneath it),
  exclusive for create/drop/dropDatabase. Each collection has its own
  rwlock; the ordering is always catalog -> collection -> log lock, never
  two collection locks at once (TTL sweep and compaction take collections
  one at a time).
- Command dispatch acquires the catalog + target collection locks for the
  handler's duration, resolving the collection (creating it for writes)
  under the catalog lock; create/drop upgrade to the exclusive catalog lock.
- Appends never fsync. Each write command's epilogue releases the
  collection lock, then commits once (seal + fsync) with a leader/follower
  group commit: the leader waits for writers mid-append (a pending counter)
  so its seal covers them, and followers whose records the seal covered
  skip their own fsync. Every acknowledged write is fsynced before its
  reply (crash pair verified); an unacknowledged write may vanish and a
  reader may observe a write before its fsync — ordinary w:1 j:true
  semantics instead of 'the log describes >= memory'.
- Compaction snapshots collections without the log lock (so a concurrent
  writer holding one can always finish its append) and retries when a
  writer appended mid-snapshot (detected via the record seq), then swaps
  under the log lock — no deadlock. The compaction trigger moved to the
  command epilogue and the TTL monitor.
- Engine.dup_index moved to the collection (per-command error paths).

Also lands two B-tree edge-case fixes driven by tests that were in flight:
a churned leaf full of dead bytes no longer splits with an empty right
half (the leaf is repacked before splitting, and an emptied node's page is
fully free again), and a slot-count split with all large records on one
side shifts records between the halves until the new record fits. Plus a
randomised fuzz test over key sizes (src/fuzz_split.zig) and the two
regression tests.

Measured (tests/e2e/results/phase6.txt): no regression on the
single-connection benchmark; concurrent durable-insert throughput ~5.1k ->
12.5k docs/s from 1 -> 8 clients, ~14.8k at 32. Verified: unit suite in
all three modes, all e2e suites, the kill -9 crash pair.
This commit is contained in:
Alexey Shakhmatov
2026-08-02 23:26:24 +03:00
parent 570900a6ef
commit ecd28d9b26
9 changed files with 999 additions and 302 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
tests/e2e/results/phase6.txt Normal file
View File

@@ -0,0 +1,49 @@
# Phase 6 gate — mongo-lite vs MongoDB 8.3.7, 1g dataset / ~16k docs
# Ratio < 1.0 = mongo-lite faster. Reproduce: bash tests/e2e/compare-run.sh 1g 16k
# This run includes all five roadmap items (B+tree, _id index, compressed
# log, byte storage, decomposed locks). Compare: tests/e2e/results/phase1.txt.
benchmark mongo-lite mongodb ratio
insertOne (sequential) ×200 0.20 ms 4.7 ms 0.0x
bulk insert throughput 739.0 MB/s 694.5 MB/s 1.1x
docs loaded 65,536 65,536 1.0x
createIndex({k: 1}) 64.7 ms 82.1 ms 0.8x
countDocuments({}) 3.4 ms 10.9 ms 0.3x
findOne({_id: <ObjectId>}) 0.83 ms 0.72 ms 1.2x
findOne({k: 500}) (indexed) 0.62 ms 1.6 ms 0.4x
find({p: {$gte,$lt}}).count() (scan) 12.4 ms 12.0 ms 1.0x
find({}).sort({_id:-1}).limit(20) 2.3 ms 2.1 ms 1.1x
find({}, {proj}).limit(1000) 3.7 ms 4.3 ms 0.9x
aggregate $group by k 10.2 ms 12.8 ms 0.8x
updateOne({_id}) ×50 0.17 ms 0.20 ms 0.8x
updateMany({k: 7}, {$inc}) 1.9 ms 6.3 ms 0.3x
deleteOne({_id}) + insertOne 0.58 ms 5.1 ms 0.1x
node client RSS 152 MB 157 MB 1.0x
server RSS 547 MB 1274 MB
kill -9 reopen 0.8s 1.3s
db on disk 97MB 88MB
# Item 5 (decomposed locks) deltas vs phase5: none on this single-connection
# benchmark (all rows within run noise). The structure is the deliverable:
# - collections are heap-allocated (stable pointers), the docs/slab/index
# maps are guarded by a catalog rwlock (shared for commands, exclusive
# for create/drop) plus one rwlock per collection (catalog -> collection
# ordering, one collection at a time for TTL sweep and compaction).
# - appends never fsync; each write command's epilogue commits once
# (seal + fsync) with leader/follower group commit: the leader waits
# for writers mid-append, seals, and syncs once, and followers whose
# records the seal covered skip their own fsync.
# - compaction snapshots collections one at a time without the log lock
# and retries if a writer appended during the snapshot (no deadlock
# against a writer holding a collection lock), then swaps under the log
# lock.
# - durability semantics: acknowledged writes are fsynced before their
# reply (crash pair verified); an unacknowledged write may vanish, and
# a reader can observe a write before its fsync completes — ordinary
# w:1 j:true semantics, no longer "the log describes >= memory".
#
# Concurrent-write throughput (sequential insertOne per client, durable):
# 1 client ~5.1k docs/s | 8 clients ~12.5k docs/s | 32 clients ~14.8k
# docs/s. The fsync per commit still dominates sequential-per-client
# workloads; the group commit coalesces when appends overlap (different
# collections on different connections).