Skip to content

feat: add bgp_jan2024_workload SQL corpus test (200 queries, ASAPQuery PR #561) - #219

Merged
milindsrivastava1997 merged 1 commit into
mainfrom
bgp-jan2024-workload-corpus
Aug 21, 2026
Merged

feat: add bgp_jan2024_workload SQL corpus test (200 queries, ASAPQuery PR #561)#219
milindsrivastava1997 merged 1 commit into
mainfrom
bgp-jan2024-workload-corpus

Conversation

@milindsrivastava1997

@milindsrivastava1997 milindsrivastava1997 commented Aug 21, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Adds a new SQL corpus test sourced from ASAPQuery PR #561's 200-query bgp_jan2024_rrc00 analyst workload, lowered under SqlDialect::ClickhouseSQL.
  • The test catalog widens bgp.bgp_updates past the existing 15-query bgp_analytics corpus's 7-column schema with the 9 extra BGP-attribute columns this workload references directly (origin, next_hop, local_pref, med, communities, atomic, aggr_asn, aggr_ip, source_file), modeled as Utf8 since every query wraps the numeric-looking ones in toString/toInt64OrZero/toFloat64OrZero and treats '' as a missing-value sentinel.
  • Unlike bgp_analytics.rs (15 hand-picked queries, exact per-query pinning), this corpus pins an aggregate tally by error category rather than a per-query outcome: at 200 queries a per-query message-snippet needle would mean 200 hand-maintained entries, and DataFusion's Plan-error "Did you mean 'x'?" spelling suggestion is confirmed non-deterministic across process runs, so needling that text would be flaky. The category split (Lowered / Plan / Schema / Parse / NotImplemented / UnsupportedFeature / Other) is stable across repeated runs and still ratchets regressions.
  • Today's pinned tally: Lowered: 64, Plan: 127, NotImplemented: 3, UnsupportedFeature: 6 (0 in Schema, Parse, Other).

Run using cargo test -p asap-frontend-sql --test bgp_jan2024_workload -- --nocapture

…561)

Copies the 200-query bgp_jan2024_rrc00 analyst workload from ASAPQuery
PR #561 verbatim as YAML (byte-identical to the source blob, plus a
trailing newline added by this repo's end-of-file pre-commit hook) and
lowers each query under SqlDialect::ClickhouseSQL against a
bgp.bgp_updates catalog widened with the BGP-attribute columns the
workload references directly (origin, next_hop, local_pref, med,
communities, atomic, aggr_asn, aggr_ip, source_file).

Unlike bgp_analytics (15 hand-picked queries, pinned per-query), this
corpus pins an aggregate tally by error category rather than an exact
per-query outcome: at 200 queries a per-query needle would mean 200
hand-maintained entries, and DataFusion's "Did you mean 'x'?" spelling
suggestion is non-deterministic across process runs, so message-snippet
needles on Plan errors would be flaky. The category tally is stable and
still ratchets regressions.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
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