Migrate RSV to volume-invariant sampling and harden skiplist economics#498
Merged
HudsonGraeme merged 2 commits intoMay 14, 2026
Merged
Conversation
The 14.8.4 throughput uplift exposed a dilution gap in the previous sample-budget mechanism: with 20 verifications budgeted per hotkey per tempo and our validator now driving ~7000 dispatches per miner per tempo, the realised per-validator sample rate collapsed from the design target (4%) to ~0.3%. Stake distribution analysis on subnet 2 shows top-2 validators hold 80% of stake (effective N via 1/HHI = 3.0), so our individual detection rate dominates aggregate consensus security rather than averaging it out. Replace the budget gate with a flat probabilistic roll (VERIFICATION_SAMPLES_PER_TEMPO / RSV_EXPECTED_SUBS_PER_TEMPO = 4%), restoring the design-intent sample rate at any throughput. Drop the sample_budget field and its persistence ratchets. Tighten strike threshold to 1 (closes the rate-of-attack loophole where a miner trickling cheats below 3-per-60-min escapes accumulation) and extend skiplist to 20 tempos (~24h zero-weight contribution from us, sized against the ~0.02 TAO/day top-miner emission at our stake share). Separately introduce --dispatch-ceiling Option<usize> so the validator no longer ties in-flight QUIC fan-out to verification CPU. Default None (uncapped) — adaptive per-miner caps and the pending_verifications buffer remain the real backpressure. The verification_concurrency * 8 formula was an artifact of the pre-RSV era when every proof entered the verifier and CPU saturation was the dominant constraint.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Repository UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
🚧 Files skipped from review as they are similar to previous changes (2)
WalkthroughVerification constants changed; RsvManager removes per-(hotkey,tempo) sampling budgets and uses volume-invariant random sampling; a new optional ChangesVerification and Dispatch Tuning
Possibly related PRs
Suggested labels
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (1)
crates/sn2-validator/src/cli.rs (1)
57-58: ⚡ Quick winAdd help text for the new CLI flag.
The
--dispatch-ceilingflag lacks documentation. Users won't understand its purpose, valid values, or when to set it without examining the code or external documentation.📝 Suggested help text
- #[arg(long)] + #[arg(long, help = "Maximum concurrent requests in flight (tasks + verifications + pending). Defaults to unbounded; backpressure from pending_verifications cap and per-miner limits still applies")] pub dispatch_ceiling: Option<usize>,🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/sn2-validator/src/cli.rs` around lines 57 - 58, Add user-facing help text for the CLI flag by annotating the dispatch_ceiling field (the struct field named dispatch_ceiling currently annotated with #[arg(long)]) with a descriptive help string explaining its purpose, acceptable values (e.g., positive integer or "none"), and when to use it; update the attribute to include help = "..." (or add a doc comment above the field) so the --dispatch-ceiling flag shows this guidance in the CLI help output.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@crates/sn2-validator/src/cli.rs`:
- Around line 57-58: Add user-facing help text for the CLI flag by annotating
the dispatch_ceiling field (the struct field named dispatch_ceiling currently
annotated with #[arg(long)]) with a descriptive help string explaining its
purpose, acceptable values (e.g., positive integer or "none"), and when to use
it; update the attribute to include help = "..." (or add a doc comment above the
field) so the --dispatch-ceiling flag shows this guidance in the CLI help
output.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Run ID: 9c524874-065b-4c6b-8741-86e226bdcfb8
📒 Files selected for processing (5)
crates/sn2-types/src/constants.rscrates/sn2-validator/src/cli.rscrates/sn2-validator/src/config.rscrates/sn2-validator/src/rsv.rscrates/sn2-validator/src/validator_loop/dispatch.rs
…flag Two rsv tests asserted behavior that is unreachable at the new STRIKES_REQUIRED=1 setting: record_strike_below_threshold_no_skiplist checked that the first strike was a no-op, and strike_aging_removes_old_strikes required accumulating multiple strikes for the in-window pruning loop to be observable. Both behaviors only manifest at threshold >= 2; deleting them rather than reshaping since the underlying aging code path remains intact for any future threshold tuning. Annotate --dispatch-ceiling with a help string covering default behavior (uncapped, governed by adaptive caps + pending buffer) and when an operator would set a hard cap.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The 14.8.4 throughput uplift exposed a sampling gap. RSV used a per-(hotkey, tempo) budget of 20 verifications — designed against ~500 expected submissions per miner per tempo. At our current throughput on this validator (~7,000 dispatches per miner per tempo), the realised sample rate collapsed from the design target of 4% to ~0.3%.
Stake-weighted consensus magnifies this. On subnet 2:
So aggregate-detection security is dominated by a handful of high-stake validators rather than the count of all validators. Our per-validator detection rate is the load-bearing factor.
Changes
should_samplewith a flatrng.range(0..RSV_EXPECTED_SUBS_PER_TEMPO) < VERIFICATION_SAMPLES_PER_TEMPOroll. Restores 4% sample rate at any throughput. Drop thesample_budgetHashMap and its retention sweep.VERIFICATION_STRIKES_REQUIRED: 3 → 1. Closes the rate-of-attack loophole where a miner trickling cheats <3 per 60-min window never accumulates the threshold. zk verification is deterministic; legitimate false positives are ~zero, so single-strike is the cleaner choice.VERIFICATION_SKIPLIST_TEMPOS: 5 → 20(~24h zero-weight from us). At 35.7% stake × 24h × ~0.02 TAO/day top-miner emission, per-cheat E[penalty] is ~3× E[gain] at restored 4% sample rate. EV-negative.--dispatch-ceiling Option<usize>, defaultNone(uncapped). Wasverification_concurrency * 8— an artifact of the pre-RSV era when every proof entered the verifier. With RSV skipping ~96% of verifications, dispatch CPU and verify CPU decouple. Adaptive per-miner caps (sum currently ~6,400 across the metagraph) andpending_verificationscap remain the real backpressure.Verification
cargo check --workspace,cargo clippy --workspace --tests -- -D warnings,cargo fmt --check,cargo test --workspace --liball clean. New unit test (should_sample_rate_is_volume_invariant) confirms sample rate stays within 0.5% of the 4% target over 100k trials.Rollout
Ship as 14.8.5. Operators on default config see:
Operators wanting to keep the conservative cap can pass
--dispatch-ceiling=N.Summary by CodeRabbit
New Features
Chores