Repository navigation
UI redesign - #2
Merged
Merged
Conversation
Replaces the single-page form in anontestlab/dashboard/static/index.html with a full client-side app (app.js router plus core.js, builder.js, charts.js, analyses.js, views.js) covering the experiment builder, live run tracking, results, and the research tools (compare, sweep, paired, predict, fidelity) as their own pages instead of CLI-only commands. Charts (ROC, score distributions, latency CDF, paired slope and interval plots, a force-directed relay graph) are hand-rolled inline SVG with no new JS dependency. server.py gained a proper job model: a stop endpoint that signals a job's real relay subprocess PIDs (SIGTERM, escalating to SIGKILL), scoped to the specific job that requested it so a late-arriving retry can't kill a later, unrelated job's relays; sweep and paired jobs check a cooperative stop flag between points/seeds so a stop doesn't throw away already-finished work; a custom output directory per job (the dashboard's equivalent of the CLI's --out), with correct handling when that path is itself nested under results/; and a job type that runs two fresh, possibly never-executed configs and diffs them in one action, persisting both component runs as well as the diff. Added anontestlab/dashboard/store.py (a read-only view of the results directory, shared by the JSON API and the new PDF reports) and anontestlab/dashboard/spec.py (round-trips the flat ExperimentConfig dict the dashboard edits against the nested YAML spec ExperimentConfig. from_yaml reads, verified against every example config). anontestlab/dashboard/report.py adds PDF reports (matplotlib's PdfPages, no new dependency) for a single run, a stored comparison, and a paired analysis. Small additions to the core library in support of the above: session_ complete and session_failed events now carry the session's actual chosen relay ids (anontestlab/emulator/orchestrator.py), relays_ready carries their PIDs, and MetricsCollector/ExperimentResult/write_results persist every delivered packet's one-way delay to oneway_delays.npz (anontestlab/metrics/collector.py, anontestlab/experiment/runner.py) so a real empirical latency CDF can be charted instead of just p50/p95/p99. None of this touches the packet forwarding hot path. tests/test_dashboard.py is rewritten for the new API shape: the full run/sweep/paired/fidelity/compare job lifecycle, stop behavior (including the relay-process-kill race and the cross-job watcher bug found while testing it), path traversal guards, and static asset serving.
and clear a UI/UX pass over the dashboard PDF reports needed matplotlib but the dashboard extra never installed it; a fresh install hit a bare 500 and the browser saved that error page as a broken report.pdf. matplotlib is now part of the dashboard extra, the route returns a clear 503 naming the install command when it's still missing, and PDF links are fetched before saving so a failure shows a message instead of a corrupt file. Live run view: the single "most recent session" circuit diagram is replaced with a sender-to-receiver rendering of every leg of a multipath session, relay by relay, with a shared exit (common_exit) drawn once where legs converge and animated dots along each leg sized by its measured share of real packets (a new leg_real_sent field on session_complete, from a count the emulator already kept). The relay connectivity graph gets the same sender/receiver anchors, pulling each relay toward its average hop position so the layout reads left to right. Fixed a double-counted event bug in the job poller that showed twice the real session count on a fresh page load. The rest is a UI/UX pass, each item checked against a real run in a headless browser, not just read in the source: - Dashboard tests no longer write into the real results/ directory (they pointed at results/ directly before, so running them could list, and a same-named rerun could overwrite, an actual experiment). - Compare's outcome table showed a raw-seconds delta next to millisecond A/B values; it now converts through the same formatter, with "-0.0 ms"-style noise cleaned off deltas that round to zero. - Sweep and Paired charts: a flat or near-ceiling series on a probability metric padded its axis past 1.0 (or below 0); the pad is now clamped to a metric's own [0, 1] once its real values sit inside it. Auto-generated axis ticks no longer repeat an identical rounded label. - Paired analysis wrote its per-seed CSV with the configs' own names as column headers; reference and treatment with the same name collided into one CSV column, silently losing the reference values to the treatment ones. Fixed at the source (fixed "reference"/"treatment" column names) with matching fixes in the PDF report and the dashboard, which still show the real names, disambiguated when equal. - The global observer's TPR card now visibly flags (amber, not quiet grey text) when its threshold was chosen on the same pairs it's scored against; this also fixed the card reading the wrong run's calibration flag (the correlation suite's, instead of global observer's own), so a properly calibrated run was shown as optimistic. - Experiments' "Open builder" and "New experiment" buttons went to the same place; replaced with one button, plus a resume-draft option only when an unfinished draft actually exists. Experiments, Runs and Artifacts list pages now cross-link and describe what's different about each. - The Adversaries tab is only offered when a run actually used the correlation suite, instead of always being present and empty. - A long run name overflowed the stored-run picker on Sweep and Paired (native <select> sizes to its longest option) and pushed the action buttons beside it onto their own line; the picker is now capped and truncates instead. - Mobile nav wrapped into an overflowing single row that ran off screen; it now wraps onto as many rows as it needs. KPI tiles sit two per row instead of one per row, which was most of what made a phone-width Results page so long to scroll. - Chart SVGs scaled with whatever container they landed in, so the same chart's text could run oversized in a wide desktop card and shrink unreadable in a narrow mobile one; width is now bounded to a comfortable range, with horizontal scroll as the fallback below that. - The Paths card appeared twice on a multipath run, once embedded in Summary and again as the whole Paths tab; Summary now shows a compact version (just the measured per-leg traffic share) linking to the full diagram. - Compare's configuration diff was a flat, alphabetized list of raw YAML keys, with object/list values cut off at the edge of a <pre> block; it's now grouped into the same sections the experiment builder itself uses, labeled the same way, with long values wrapping inside their cell instead of overflowing. - The Predict page was a bare dropdown and button with no sign of what running it does; it now explains what the model predicts, what picking a stored run adds, and where to find the comparison. Left open, by design: a run doesn't store which relays a session actually used (only the configured design), so Results can't yet show the same sender-to-receiver diagram the live view now draws from real per-session events. That needs new data persisted per run, not a UI change.
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
1 participant
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.
No description provided.