#36244 validation, net: Process blocks asynchronously and reduce cs_main contention

Decouples block persistence and chain activation from the P2P message thread by introducing a dedicated single-worker queue in ChainstateManager that processes admitted blocks asynchronously via futures. Initial validation checks remain synchronous on the P2P thread to decide admission, and each peer is limited to one queued or processing block at a time. The PR also removes cs_main requirements from peer block download tracking and compact block reconstruction, while allowing SendMessages to defer work when cs_main is contended.

Problem: The P2P message thread validates and activates blocks synchronously while holding cs_main. During initial block download or heavy block processing at the tip, message handling for all other peers stalls, leading to severe latency spikes for pings, block serving, and compact block reconstruction.

Evidence of importance:

• PR description benchmarks across a 10,000-block IBD segment: PING/PONG latency dropped 98% from 41.8 ms to 0.77 ms, and block-serving latency decreased by 32% (single connection) and 45% (four connections).
• PR description: Total IBD download and validation time for the benchmark segment fell by ~5% (473.6 s to 450.3 s) with four download connections.

Scope: Roughly 67% of the diff consists of unit, fuzz, and integration test coverage. The remaining ~1,100 lines touch sensitive concurrency paths across validation, net_processing, and the kernel C-API.

Risk of a mistake: high (validation, p2p). Altering lock ordering around cs_main and shifting block validation to an asynchronous worker could introduce race conditions, peer misattribution, deadlocks, or state corruption during chain activation.

What the assessment lacked:

• rest_of_diff

Ready · Needs rebase

Conflicts with master; review the logic now, the rebase is mechanical.

Waiting on (from the thread read):

• reviewer: reviewers to evaluate the draft approach and author's reply 2026-09-15
• reviewer: feedback on the draft PR's approach following author's clarification 2026-09-15

Crickets

no substantive comment either way

Thread notes: Added andrewtoth's question from c:5672590962 regarding motivation and parallel validation, which w0xlt answered in c:5673294708. / Added andrewtoth's question regarding steady-state responsiveness, which author w0xlt directly answered.

Objections, suggestions, and questions:

Participants:

• andrewtoth (member, question, 1 statements 2026-09-15 to 2026-09-15): asked whether steady-state responsiveness and future parallel validation are the primary motivation rather than IBD speedup

2304 lines under test/bench/ci.

• src/test/peerman_tests.cpp +1249/-2
• src/net_processing.cpp +326/-144
• src/test/validation_queue_tests.cpp +319/-0
• src/test/validation_block_tests.cpp +272/-7
• src/validation.cpp +144/-60
• src/test/validationinterface_tests.cpp +167/-3
• src/validation_queue.cpp +120/-0
• src/test/util/validation.cpp +103/-0
• src/validation.h +79/-9
• src/validation_queue.h +88/-0
• src/test/util/validation.h +71/-6
• src/primitives/block.h +37/-7
• src/node/block_template_manager.cpp +17/-12
• src/kernel/bitcoinkernel.cpp +16/-7
• src/rpc/mining.cpp +14/-6
• src/test/util/mining.cpp +11/-8
• src/validationinterface.cpp +13/-3
• src/blockencodings.h +10/-4
• src/test/blockfilter_index_tests.cpp +10/-3
• src/test/fuzz/cmpctblock.cpp +11/-1
• src/test/fuzz/process_messages.cpp +11/-1
• src/node/miner.cpp +4/-3
• src/test/baseindex_tests.cpp +6/-1
• src/test/util/net.h +7/-0
• src/validationinterface.h +7/-0
• src/bench/blockencodings.cpp +4/-2
• src/init.cpp +6/-0
• src/net_processing.h +6/-0
• src/test/util/setup_common.cpp +5/-1
• src/test/validation_tests.cpp +3/-3
• src/test/txindex_tests.cpp +5/-0
• src/test/validation_chainstate_tests.cpp +3/-1
• src/net.h +3/-0
• src/net.cpp +2/-0
• src/test/coinstatsindex_tests.cpp +2/-0
• src/test/fuzz/util/net.h +2/-0
• src/test/txospenderindex_tests.cpp +2/-0
• src/CMakeLists.txt +1/-0
• src/kernel/CMakeLists.txt +1/-0
• src/kernel/chainstatemanager_opts.h +1/-0
• src/test/CMakeLists.txt +1/-0
• src/test/validation_chainstatemanager_tests.cpp +1/-0

The short text the ranking pass compares across PRs.

Moves block storage and chainstate activation off the P2P message loop to a single-threaded asynchronous validation worker, returning outcomes via futures and limiting each peer to one pending block. It also decouples block download tracking and compact-block reconstruction from cs_main. This solves stalls in P2P message handling where nodes become unresponsive to pings and serving requests while validating blocks. In author IBD benchmarks, PING latency dropped by 98% and block-serving latency decreased by 32-45%.

P2 · speedup

Decoupling block processing substantially resolves message processing stalls and latency spikes during block reception. PR benchmarks show ping/pong latency dropped 98% from 41.8 ms to 0.77 ms and block-serving latency decreased by up to 45%.

Membership: Changes src/net_processing.cpp and src/test/peerman_tests.cpp to decouple P2P message handling from block validation and reduce cs_main contention.

Factors: security/stability 1, bug 0, performance 2, user value 1, leverage 1

P2 · speedup (ranking pass: Moved up: user-visible responsiveness fix plus ~5% IBD, broader effect than the other speedups below it)

Restructures how block persistence and chain activation are queued and executed within ChainstateManager. Benchmarks demonstrate a ~5% total IBD time reduction alongside major concurrency improvements.

Membership: Introduces src/validation_queue.cpp and modifies src/validation.cpp to asynchronously process block persistence and chain activation in ChainstateManager.

Factors: security/stability 1, bug 0, performance 2, user value 1, leverage 1