Move DoStateCheckpoint off critical path

[msmouse]

Description

1. Add struct State to represent the speculative state, utilizing the LayeredMap / MapLayer
2. Wrap the smt in StateSummary, to represent the speculative state tree. StateStorageUsage is removed from the smt and put in State.
3. The LayeredMap / MapLayer stack can read diffs between versions by following in mem links, hence we were able to remove the hashmaps to track the updates between versions.
4. Proof read is not removed from the execution stage so we save some CPU by only reading them for the writes, not the reads.

Todo: move it further to a separate stage.

Throughput

1. All single-node-performance cases have seen improvements. See the last commit.

2. On a 100M account DB and t4d 60 cores machine, apt-fa-transfer (with native vm, redoing for aptos vm) saw gain from 32k TPS to 40k TPS:

$ for suffix in main main final
do RUST_BACKTRACE=1  PUSH_METRICS_NAMESPACE=alden-benchmark-100Mp2p ./aptos-executor-benchmark.$suffix \
--execution-threads 32 \
--block-executor-type native-vm-with-block-stm \
--transactions-per-sender 1 \
--generate-then-execute \
--block-size 8086 \
--enable-storage-sharding \
run-executor \
--enable-feature NEW_ACCOUNTS_DEFAULT_TO_FA_APT_STORE \
--enable-feature OPERATIONS_DEFAULT_TO_FA_APT_STORE \
--transaction-type apt-fa-transfer \
--transaction-weights 1 \
--module-working-set-size 1 \
--main-signer-accounts 20000 \
--additional-dst-pool-accounts 600000 \
--data-dir ~/work/data/100M \
--checkpoint-dir ~/work/data/cp \
--blocks 300
done

3. with AptosVM, the same thing saw 24k -> 26.8k TPS improvement.

4. creating 100 100M accounts DB takes 40 minutes vs 48 minutes from main branch code.

Latency

latency in forge didn't change much:

realistic_env_load_sweep
main: https://github.com/aptos-labs/aptos-core/actions/runs/12696454894
this: https://github.com/aptos-labs/aptos-core/actions/runs/12696477211

realistic_env_workload_sweep
main: https://github.com/aptos-labs/aptos-core/actions/runs/12696466689
this: https://github.com/aptos-labs/aptos-core/actions/runs/12696621130

In theory a block with both heavy execution (but not much reading) and writing to a lot of COLD state items will suffer latency-wise since we don't prefetch the proof anymore and as a result the disk IOs for proof fetching no longer overlap with the execution time, but at our current DB size a warm DB pretty much catches everything (through RocksDB and the filesystem). Explicit caching of hot items is coming so I'm not complicating things by adding pre-fetching back.

How Has This Been Tested?

existing coverage

Key Areas to Review

Type of Change

• New feature

Which Components or Systems Does This Change Impact?

• Validator

[trunk-io]

⏱️ 2h 56m total CI duration on this PR

Slowest 15 Jobs	Cumulative Duration	Recent Runs
rust-cargo-deny	20m	🟩 ⬜ 🟩 🟩 🟩 (+7 more)
check-dynamic-deps	17m	🟩 🟩 🟩 🟩 🟥 (+7 more)
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	13m	🟩
rust-move-tests	12m	🟩
rust-move-tests	9m	⬜
rust-move-tests	7m	⬜
general-lints	5m	🟩 ⬜ 🟩 🟩 🟩 (+7 more)
semgrep/ci	5m	🟩 🟩 🟩 🟩 🟩 (+7 more)
check	3m	🟩

_{settings ⋅ feedback ⋅ docs ⋅ learn more about trunk.io}

[grao1991]

remove?

[github-actions]

✅ Forge suite realistic_env_max_load success on d21c2a6909f603961c5d5d4045423265a8fa73ae

two traffics test: inner traffic : committed: 14975.83 txn/s, latency: 2654.34 ms, (p50: 2700 ms, p70: 2700, p90: 2900 ms, p99: 3300 ms), latency samples: 5694120
two traffics test : committed: 99.98 txn/s, latency: 1486.55 ms, (p50: 1300 ms, p70: 1400, p90: 2400 ms, p99: 3000 ms), latency samples: 1740
Latency breakdown for phase 0: ["MempoolToBlockCreation: max: 1.561, avg: 1.439", "ConsensusProposalToOrdered: max: 0.290, avg: 0.287", "ConsensusOrderedToCommit: max: 0.301, avg: 0.292", "ConsensusProposalToCommit: max: 0.586, avg: 0.579"]
Max non-epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 0.51s no progress at version 291 (avg 0.19s) [limit 15].
Max epoch-change gap was: 0 rounds at version 0 (avg 0.00) [limit 4], 0.52s no progress at version 2850515 (avg 0.52s) [limit 16].
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking

[github-actions]

✅ Forge suite compat success on 6593fb81261f25490ffddc2252a861c994234c2a ==> d21c2a6909f603961c5d5d4045423265a8fa73ae

Compatibility test results for 6593fb81261f25490ffddc2252a861c994234c2a ==> d21c2a6909f603961c5d5d4045423265a8fa73ae (PR)
1. Check liveness of validators at old version: 6593fb81261f25490ffddc2252a861c994234c2a
compatibility::simple-validator-upgrade::liveness-check : committed: 16263.42 txn/s, latency: 2119.71 ms, (p50: 1800 ms, p70: 2000, p90: 3000 ms, p99: 6900 ms), latency samples: 529240
2. Upgrading first Validator to new version: d21c2a6909f603961c5d5d4045423265a8fa73ae
compatibility::simple-validator-upgrade::single-validator-upgrading : committed: 7179.69 txn/s, latency: 4155.73 ms, (p50: 4800 ms, p70: 5100, p90: 5200 ms, p99: 5300 ms), latency samples: 133820
compatibility::simple-validator-upgrade::single-validator-upgrade : committed: 7215.76 txn/s, latency: 4667.96 ms, (p50: 5000 ms, p70: 5100, p90: 5200 ms, p99: 5700 ms), latency samples: 245340
3. Upgrading rest of first batch to new version: d21c2a6909f603961c5d5d4045423265a8fa73ae
compatibility::simple-validator-upgrade::half-validator-upgrading : committed: 6130.56 txn/s, latency: 4950.64 ms, (p50: 5800 ms, p70: 6100, p90: 6300 ms, p99: 6500 ms), latency samples: 115340
compatibility::simple-validator-upgrade::half-validator-upgrade : committed: 6174.02 txn/s, latency: 5462.99 ms, (p50: 6000 ms, p70: 6100, p90: 6300 ms, p99: 6500 ms), latency samples: 212400
4. upgrading second batch to new version: d21c2a6909f603961c5d5d4045423265a8fa73ae
compatibility::simple-validator-upgrade::rest-validator-upgrading : committed: 14012.63 txn/s, latency: 2061.96 ms, (p50: 2100 ms, p70: 2500, p90: 2700 ms, p99: 2700 ms), latency samples: 239720
compatibility::simple-validator-upgrade::rest-validator-upgrade : committed: 14319.12 txn/s, latency: 2243.18 ms, (p50: 2200 ms, p70: 2600, p90: 2800 ms, p99: 3100 ms), latency samples: 461540
5. check swarm health
Compatibility test for 6593fb81261f25490ffddc2252a861c994234c2a ==> d21c2a6909f603961c5d5d4045423265a8fa73ae passed
Test Ok

• Grafana dashboard
• Humio Logs
• Axiom Logs
• Validator CPU Profile
• Fullnode CPU Profile
• Test runner output
• Test run is land-blocking