#35827 bench: add index benchmarks

• Measure realistic block index sync and query performance without requiring a full initial block download
• Add benchmarks for indexes using blocks with multi-input and multi-output transactions

Adds shared utilities and benchmarks for index synchronization and lookups across txindex, txospenderindex, and blockfilterindex. It builds chains with realistic multi-transaction blocks to test write and lookup paths, and adds documentation on measuring I/O and memory usage with bench_bitcoin.

Problem: Index performance and memory optimizations currently lack granular benchmarks; the existing block filter benchmark only syncs coinbase transactions, and txindex has no benchmarks at all, forcing developers to rely on hours-long initial block download runs to test changes.

P3 · test coverage

• P3 because index benchmarks only test coinbase-only blocks and fail to show scaling behavior
• Adds shared benchmark coverage for realistic transaction loads

Worthwhile addition to the benchmark suite. As stated in the description, there is currently 'no cheap way to tell whether a change actually improves what it set out to improve' for indexes without a full IBD, and existing benchmarks only cover coinbase transactions. It provides clear utility for evaluating index performance changes.

Membership: Adds new benchmark harnesses to src/bench/ for indexing subsystems.

Factors: security/stability 0, bug 0, performance 0, user value 0, leverage 1

• Ready for review
• All review feedback addressed and CI is clean

The author has addressed all initial review feedback from l0rinc with updated code and documentation.

Author status: Active; addressed reviewer comments and pushed fixes on 2026-09-06.

Resolved concerns:

• Locking inside the timed loop in index_sync_util.h could distort results (addressed by precomputing the expected tip outside the loop).
• Small benchmark sizes risk staying entirely in LevelDB write buffers or OS page cache without hitting disk (addressed by forcing chainstate flush and documenting -testdatadir).

• Concept approval of the overall structure (l0rinc)
• All inline reviewer questions and feedback resolved by the author

Positive; Concept ACK from l0rinc with inline review concerns resolved by the author

l0rinc provided a Concept ACK and several constructive review comments, all of which were resolved in the author's follow-up commits.

• l0rinc gave Concept ACK on 2026-07-27
• arejula27 updated the implementation on 2026-09-06 to address locking, disk flushing, and documentation

Objections:

Support:

• l0rinc: Concept ACK, welcomed taking care of index benchmarks

Participants: l0rinc (support)

State derived from the lists: substantive support, no open objection (l0rinc) (model's own read: Positive)

Review verdicts (DrahtBot): 0

• Concept ACK: l0rinc

414 lines under test/bench/ci.

• src/bench/index_sync_util.cpp +146/-0
• src/bench/index_sync_util.h +98/-0
• src/bench/index_txospender.cpp +68/-0
• src/bench/index_txindex.cpp +66/-0
• doc/benchmarking.md +38/-0
• src/bench/index_blockfilter.cpp +33/-0
• src/bench/CMakeLists.txt +3/-0

This PR adds benchmarks in src/bench for txindex, txospenderindex, and blockfilterindex sync and lookup operations over chains with realistic multi-transaction blocks. It solves the lack of granular performance testing for index modifications, which previously required full initial block downloads to evaluate. It has received a Concept ACK from l0rinc, whose initial technical comments were addressed in a September 2026 push. It is ready for further review.