gossipsub: gate flood-publish by message size — fix drive saturation + outbox overflow (coverage ceiling)#279
Merged
Conversation
…tion + outbox overflow After v0.2.52 (which fixed the dominant non-delivery path, lifting coverage to ~28), the residual +0/-N decay (late=none) traced to DRIVE-LOOP SATURATION, not churn (the churn is downstream). Live steady-state log: SLOW drive iter inbound_streams=120-134ms + persistent gossip priority(1024)/bulk(64) outbox caps hit (dropping attestation/block frames). ROOT: v0.2.51 flood_publish applied to EVERY topic with no gating. On the dense block topic (~31 subscribers, ~3 MiB blocks) the proposer fanned its block to all 31 peers (~4x mesh_n) on the 64-slot bulk lane → bulk cap hit (block drops) + ~4x inbound fan-in → advanceInboundStreams 134ms → 200ms iters → ACK starvation → no-ACK teardown churn. flood on the dense/large block topic is net-harmful (blocks propagate fine via mesh; coverage is about small attestations) — zero coverage benefit for the volume. Fix: gate flood_publish to messages <= flood_publish_max_message_bytes (128 KiB, configurable). Small attestations/aggregations still flood (the intended first-hop-coverage benefit); large blocks fall back to the proven mesh-only forward path (pre-flood behavior). Cuts the dominant volume → relieves BOTH the outbox overflow AND the 134ms saturation (confirmed dominant lever by two independent live investigations). Build clean; 504/506 tests. Pure zig-libp2p.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
After v0.2.52 lifted coverage to ~28 (no more total collapse), the residual
+0/-Ndecay (late=none) traced to drive-loop saturation, with the churn downstream. Live steady-state log:SLOW drive iter ... inbound_streams=134ms+ persistent gossippriority cap (1024)/bulk cap (64)outbox hits dropping frames.Root cause
v0.2.51
flood_publishapplied to every topic with no gating. On the dense block topic (~31 subscribers, ~3 MiB blocks) the proposer fanned its block to all 31 peers (≈4×mesh_n) on the 64-slot bulk lane → bulk cap hit (block drops) + ~4× inbound fan-in →advanceInboundStreams134ms → ~200ms iterations → ACK starvation → no-ACK teardown churn. Flooding the dense/large block topic is net-harmful: blocks propagate fine via the mesh, and coverage is about small attestations — so it is pure volume for zero coverage benefit.Fix
Gate
flood_publishto messages ≤flood_publish_max_message_bytes(128 KiB, configurable). Small attestations/aggregations still flood (the intended first-hop-coverage win that took coverage 8→28); large blocks fall back to the proven mesh-only forward path (pre-flood behavior). One condition selecting between two already-reviewed code paths → low risk. Cuts the dominant volume, relieving both the outbox overflow and the 134ms saturation — confirmed the dominant lever by two independent live investigations.Build clean; 504/506 tests. Pure zig-libp2p.