Skip to content

fix(routing): measured ETA outranks the synthetic bench score (v0.3.70) - #369

Merged
ralyodio merged 1 commit into
masterfrom
fix/smart-policy-eta-real-load
Nov 14, 2026
Merged

ralyodio merged 1 commit into
masterfrom
fix/smart-policy-eta-real-load

Conversation

@ralyodio

Copy link
Copy Markdown
Collaborator

Why

Smart routing ranked the benchmark bucket before ETA. For a chat job on 2026-11-14, vienna (bench 4156, eta 301 s, real load 32 on 16 cores) beat bonita (bench 2452, eta 60 s), and the job sat ~56 s on vienna before moving. A bid's eta_s is measured (recent average job time x queue depth), so it already reflects real speed under load; bench_score is a one-off synthetic score, and load only counts c0mpute's own slots (vienna reported 0.0).

What

  • Ranking is now: overload, then ETA bucket (within 1.5x of the fastest), then bench bucket, then ETA, price, publish time.
  • Tests: replay of the vienna/bonita bids; close ETAs (60 vs 80 s) still go to the more capable node; the existing tests are unchanged.
  • v0.3.70

🤖 Generated with Claude Code

Smart routing ranked the benchmark bucket before ETA, so vienna (bench 4156,
eta 301 s, real load 32 on 16 cores) beat bonita (bench 2452, eta 60 s) and a
chat job waited ~56 s on the slow node before it moved. A bid's ETA is
measured (recent job time x queue depth) and already reflects how fast the box
runs under its real load; the bench is a one-off synthetic CPU score.

Now: overload, then ETA bucket (within 1.5x of the fastest), then bench bucket,
then ETA, price, publish time. Tests replay the vienna/bonita bids and keep the
capable node winning when ETAs are close.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown

ThreatCrush Security Scan

14 finding(s)

HIGH/CRITICAL: 2 | MEDIUM: 8 | LOW: 4

Severity Rule Location
HIGH js-ssrf-outbound-request apps/web/public/sw.js:24
HIGH sh-remote-script-execution scripts/install.sh:606
MEDIUM js-unescaped-html-sink apps/web/src/app/blog/[slug]/page.tsx:53
MEDIUM js-unescaped-html-sink apps/web/src/app/layout.tsx:109
MEDIUM sql-template-interpolation apps/web/src/lib/chat/convo-store.ts:116
MEDIUM sql-template-interpolation apps/web/src/lib/chat/convo-store.ts:117
MEDIUM sql-template-interpolation apps/web/src/lib/chat/push.ts:476
MEDIUM js-dynamic-code-execution apps/web/src/lib/router/verify.ts:91
MEDIUM sh-remote-script-execution scripts/dev-setup.sh:25
MEDIUM sh-eval-expansion scripts/dev-setup.sh:35
LOW py-ssrf-outbound-request examples/business-strategist-agent/chat.py:37
LOW py-ssrf-outbound-request examples/business-strategist-agent/train.py:54
LOW secret-generic-credential node/crates/c0mpute-core/src/webhook.rs:1171
LOW py-path-traversal node/crates/c0mpute-core/tests/fixtures/fake-infernet-text:29

Snippets are redacted; ThreatCrush never prints matched credential material.

@ralyodio

Copy link
Copy Markdown
Collaborator Author

Merging with hosting_e2e::sites_replicate_to_three_heal_and_serve_from_any_gateway red twice: chunk fetch EOF between in-process test nodes (IO error on outbound stream: Eof). Hosting does not use routing::rank; all 718 c0mpute-core unit tests including the new routing ones pass. Master has a history of flaky hosting e2e tests (preferred_hosts_hold_pass_challenges_and_get_paid on d4d4274, 0d9e14d).

@ralyodio
ralyodio merged commit a60c488 into master Nov 14, 2026
17 of 19 checks passed
@ralyodio
ralyodio deleted the fix/smart-policy-eta-real-load branch November 14, 2026 14:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant