Repository navigation
Spec 001 T007: a single event's PDF, stage by stage, cold; SC-005 decided - #102
Merged
Merged
Conversation
Each event of the baseline's sample, at depth 1, in a render service of its own at the website settings with a fresh cache key and the release's files read locally (RENDER_FILES). Figures alone exceed event-pdf's whole time for every reaction (2.6 s against 0.5-1.7 s) and every pathway but Signal Transduction (32.9 s against 46 s); one figure with an analysis token takes 4.5 s, and 19.6 s for the last of eight at once. SC-005 is raised with the user. Script ci/measure/single.ts, outputs in ci/measure/results/. Also: `figures` requires a render service reading the release's files locally (WebsiteAngular #777) and refuses one that does not; R7 records that T005's pathway timings included fetches from the public download site; T005 and T006 carry their PR number, #95. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From the review: a pathway's figures take about 1.8 s each at two at once, not 1.5; the warm-up figure was a pathway whose figure no sampled event draws, not an unrelated reaction; R7 says how the outputs were assembled and commits the one-process run behind the layout figure; the script's header says what made the runs cold, counts pages with pdf.js and keeps figure times in request order; cli.md and the folder contract name the files: local refusal and that the render service's folder is tied to no release. SC-005, left to us by the user: a single-event PDF without an analysis reads its figures from the release's figures' folder and keeps event-pdf's times; one with an analysis draws them on request within 6 s plus 2 s per figure. serve takes --figures; T022 and T023 follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… SC-005 widened The first run's warm-up asked for a pathway as a reaction and was refused unchecked. Re-run with a reaction no sampled event draws, its 200 checked, and each output recording the warm-up and the service's hit count (0): totals within 1.3 s of the first run, which is kept as events-unwarmed.jsonl. The 6 s + 2 s per figure budget was fitted to one run and two events now exceed it; it is 8 s + 3 s per figure, with room for run-to-run variation and a token's slower figure (4.6 s against 3.4 s alone). Also from the review: SC-005's no-analysis clause covers the events the figures' folder holds (human); other species and token PDFs draw on request. A folder of another release, or incomplete, is not used (the service keeps running, /health says why); the graph's release is read again at most once a minute. T022 ships drawing everything until T024's checks exist, T023 follows T046 and has its failure clause back, with "cold" defined. Three merged list items are split. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t right, T047 split out - The two runs differ by up to 2.2 s (R-HSA-68875), not 1.3; by margin the tightest event is R-HSA-68881 (5.7 s to spare), by share R-HSA-73857. - The folder's use in the service is its own task, T047 after T024, so T022 ships drawing on request and can be ticked; T023 follows T046 and T047, adds a non-human reaction and pathway, and measures a plain diagram's overlay now. - SC-005 covers a human PDF while the folder is missing or stale, and says what "cold" is. The service runs the folder's full checks only at start and when the fingerprint or release changes. The token's cost against T005 is called not like for like; the small pathways' remainder overlaps event-pdf's range; routing waits for T023; the runs' wrapper is said to be outside the repo. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drawing on request is several times slower than event-pdf for all but the largest events (reactions about 2-10 times, small pathways about 2-16), not "2-5 times". T023 also runs the human sample with the folder taken away, for SC-005's missing-folder clause. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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
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.
Spec 001, T007 (with T005's (c)). Replaces #96, #97, #99 and #101, with their reviews' findings fixed.
Measured each event of the
event-pdfbaseline's sample at depth 1 on this host. Each ran in a render service of its own:RENDER_FILES(WebsiteAngular #777);event-pdf(s)event-pdf's whole time in every group but Signal Transduction.events-unwarmed.jsonl.SC-005 decided (the user left it to us):
event-pdf's times. T023 measures this.book's checks against the graph's current release; otherwise the service draws on request and/healthsays why.Also:
figuresrefuses a render service that isn't reading local files.From #99's review: run-to-run variation is up to 2.2 s, and the tightest event by margin is R-HSA-68881 (5.7 s to spare). T047 is split out of T022, removing the last ordering knot. "Cold" and the stale-folder case are in SC-005, and the full folder checks run only when the folder or release changes.
🤖 Generated with Claude Code