Published results snapshots

Freeze published study tables into static Markdown on getbent.io—one subdirectory per PostgreSQL major version (starting with pg18/).

PG18 published live on Streamlit first, then snapshotted here. PG19 skips hosted Streamlit (no sponsor / no paid hosting): the deliverable is study SQL plus static pages and graphs under docs/results-summary/pg<N>/. Explorer code may remain for local use; do not treat re-enabling Streamlit Cloud as part of this cycle.

Publication paths

Path Role
Leaderboards (docs/leaderboards.md → /leaderboards/) Public OSM + pgbench SELECT ranking URL; refresh when new cells land. OSM disk/power: /leaderboards/osm-performance-power/
Results summary (docs/results-summary/pg<N>/) Full study tables on getbent.io; primary archive for PG19
Study SQL (reports/*.sql and/or explorer/submission_studies.py) Query that defines each study
Streamlit (explorer/submission-explore.py) PG18 path / optional local; not hosted for PG19

Snapshot from the live site (no database)

The script loads the live app in headless Chromium, selects each study via the radio buttons, and pulls the full table via Streamlit’s Download data as CSV button (the on-screen dataframe is virtualized and only shows ~12 rows).

pip install -r explorer/requirements-snapshot.txt
playwright install chromium
python3 explorer/snapshot-streamlit.py --pg 18

Writes docs/results-summary/pg18/<study-id>.md for all seven sections:

Study id Streamlit label
osm-power OSM Power
osm-leaderboard OSM Leaderboard
osm-network OSM Network
osm-checkpoint OSM Checkpoint
osm-dirty-memory OSM Dirty Memory
pgbench-build pgbench Build Time
pgbench-select pgbench SELECT

Review tables, commit, and link from getbent.io /results-summary/pg18/.

Snapshot from results database (preferred for PG19)

With read access to the submission view:

python3 explorer/archive-results.py --pg 19

Same output layout when studies are registered in submission_studies.py. Otherwise export each study’s SQL result into docs/results-summary/pg<N>/<study>.md by hand (or from reports/*.sql).

Workflow (PG 18)

  1. Run snapshot-streamlit.py (or archive-results.py if DB available)
  2. Review and commit docs/results-summary/pg18/*.md
  3. Link from docs index / tuning pages
  4. Pause Streamlit Cloud hosting — explorer code kept; PG19 does not resume hosting

Workflow (PG 19)

  1. Starting hub + studies from published grids (OSM ranking, OSM power, pgbench SELECT) under docs/results-summary/pg19/
  2. Prefer archive-results.py or direct SQL export; do not require Streamlit Cloud
  3. Initial 19rc1 report — 19 took the top spots with a tiny delta over 18; siren durable OSM 4.49h matches beta4. Not cycle close.
  4. Refresh OSM Power with court 19rc1 NVIDIA (fast 8.30h / 521) and Abbey PG18 as the Studio stand-in.
  5. Siren OS wrap then cutover: four-face SELECT PG15→18 on 9600X; 19rc2 durable OSM (fill fast/durable grid); Geekbench 7 before and after OS upgrade (profile; no more GB6); then OS upgrades on siren and twilight and re-test main SELECT + OSM (newset by OS vintage) — pg-release-runs § OS upgrades

Milestones

  • explorer/submission_studies.py — SQL for live Streamlit app
  • explorer/snapshot-streamlit.py — crawl live site → Markdown
  • explorer/archive-results.py — optional DB export
  • submission-explore.py uses study_sql()
  • docs/results-summary/ hub + pg18/ index
  • PG 18 study .md files captured via snapshot script
  • PG 18 study .md files committed
  • Site links updated

This site uses Just the Docs, a documentation theme for Jekyll.