Skip to content

Pecan: weighted SSSP and landmark programs on Pregel - #32

Draft
alexy wants to merge 2 commits into
pecanfrom
work/sem-e0-pregel-20261003
Draft

alexy wants to merge 2 commits into
pecanfrom
work/sem-e0-pregel-20261003

Conversation

@alexy

@alexy alexy commented Oct 3, 2026

Copy link
Copy Markdown
Member

Implements Sem's remaining Pregel programs on the existing Pecan loop: weighted, active-frontier SSSP and graphframes-rs-compatible landmark shortest paths. The explicit SSSP method returns distance only; the existing frontier methods retain their contracts. Landmark distances use Int32 hops and the reference MAX sentinel, with forward and reversed directions. No input-validation jobs are added.

Verification at f2b297fc8443221891ce5f4afe88f955ed125b38:

  • Ruff and strict mypy passed on the package and benchmark helpers.
  • All 227 Pecan tests passed against the optimized native Sail runtime at 9f0aa7d2a.
  • Eleven native Parquet input/output cells passed full independent output comparisons across cit-Patents, kgs and wiki-Talk; engines exit before validation.
  • Weighted, undirected kgs SSSP also matches all 832,247 official reference rows, including 12,998 unreachable vertices. Maximum finite absolute difference is 8.88e-16, with a 1e-12 comparison tolerance.
  • The original Cit source-1 controls are retained; two additional cells use the earlier A2 source 5795784, reaching 126,298 vertices in 14 steps.

The benchmark helpers retain physical plans, source and input identities, raw Parquet outputs and separate oracles. The review report and preserved measurements live in querygraph/grust; shared-host clocks are diagnostic observations. This PR does not implement a server-side loop or change public defaults.

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