memdb: prevent iterator invalidation

[ekexium]

ref pingcap/tidb#59153

To prevent potential misuse and iterator invalidation, modify the iterators provided by ART memdb as follows:

1. Iter and IterReverse now comes with an extra check: it is invalidated immediately by any write operation to the memdb after the creation of the iterator. Attempting to use such an invalidated iterator will result in a panic.
2. SnapshotIter and SnapshotIterReverse will be replaced by BatchedSnapshotIter.
   2.1. SnapshotIter is different from Iter that it can be valid after write operations, but only becomes invalid if a write operation modifies the "snapshot".
   2.2. We need to introduce BatchedSnapshotIter instead of directly modifying SnapshotIter because SnapshotIter maintains internal states and pointers. Consider a situation where a write operation causes changes to the internal data structure making the pointers invalid, while the snapshot should remain valid.
   2.3. SnapshotIter and SnapshotIterReverse are not removed now for compatibility.

RBT is unchanged as it is no longer used.
Pipelined MemDB still doesn't support iterators as it was.

Performance

Iterator microbenchmark

go test -run=^$ -bench=BenchmarkSnapshotIter -benchtime=3s
goos: linux
goarch: amd64
pkg: github.com/tikv/client-go/v2/internal/unionstore
cpu: AMD Ryzen 9 9900X 12-Core Processor            
BenchmarkSnapshotIter/RBT-SnapshotIter-24                     1783           1826587 ns/op             144 B/op               2 allocs/op
BenchmarkSnapshotIter/ART-SnapshotIter-24                     1870           1718771 ns/op             496 B/op              11 allocs/op
BenchmarkSnapshotIter/ART-BatchedSnapshotIter-24              1120           2964170 ns/op          417461 B/op             370 allocs/op
BenchmarkSnapshotIter/ART-ForEachInSnapshot-24                1771           1832084 ns/op             496 B/op              11 allocs/op
PASS
ok          github.com/tikv/client-go/v2/internal/unionstore        14.598s

TiDB union scan executor
BatchedSnapshotIter

go test -run=^$ -bench=BenchmarkUnionScanRead -benchtime=10s
    3999           3063035 ns/op          793197 B/op           17578 allocs/op

SnapshotIter

go test -run=^$ -bench=BenchmarkUnionScanRead -benchtime=10s
    3862           2990516 ns/op          386942 B/op           17434 allocs/op

[you06]

rest LGTM

[you06]

Shall we move the RLock operation before seqno check?

[ekexium]

It's discussed above. SnapshotSeqNo is not supposed to be accessed concurrently. This is the property that SnapshotSeqNo must have. Should this case happen we'd better expose this race

[you06]

Oh I get you, SnapshotSeqNo is changed in very low frequency.

[cfzjywxk]

@ekexium
Better to add some comments to explain the usage of RLock here, others may not be familiar with the upper MemDB and underlying ART implementation and usages.

[cfzjywxk]

LGTM

[ti-chi-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: cfzjywxk, you06

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [cfzjywxk,you06]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ti-chi-bot]

[LGTM Timeline notifier]

Timeline:

• 2025-02-19 06:38:18.325182833 +0000 UTC m=+1029740.721404894: ☑️ agreed by cfzjywxk.
• 2025-02-19 07:59:29.789850576 +0000 UTC m=+1034612.186072637: ☑️ agreed by you06.