fix(usage): harden statistics aggregation, chart geometry, and telemetry - #2353
Open
chaos-cl wants to merge 8 commits into
Open
fix(usage): harden statistics aggregation, chart geometry, and telemetry#2353chaos-cl wants to merge 8 commits into
chaos-cl wants to merge 8 commits into
Conversation
added 8 commits
August 17, 2026 12:33
Add a settings page showing historical model call statistics (requests, tokens, estimated cost) broken down by model, provider group, and endpoint, with a token usage trend chart. - services-core: pure aggregation module with unit tests - core: attribution resolver (config + models.dev catalog) and get_statistics on the token usage service - desktop: get_token_usage_statistics Tauri command - web-ui: UsageStatisticsConfig settings tab + i18n (en/zh-CN/zh-TW)
- services-core: remove price estimation; UsageStatisticsEntry now carries cacheHitRate (cached / reported input); UsageStatistics gains totalCacheReportedInputTokens for the overall hit rate - core: resolver no longer resolves prices - web-ui: summary cards show requests/tokens/cached tokens/hit rate; tables show model/group/endpoint, requests, tokens, cache hit rate; fix table header layout (min column widths + no-wrap) that squeezed Chinese labels; drop cost columns and footnote - i18n: update en/zh-CN/zh-TW copy
…splay - move cache hit rate out of the distribution tables into a dedicated per-model card with a ratio bar - display hit rates truncated to two decimals (never rounded up); only a true 100% hit (cached == reported input, exact in the backend) shows as 100% - trend chart now spans the full row; distribution tables revert to name / requests / tokens
The image understanding tool calls AIClient directly, bypassing the round executor, so its consumption never reached the token usage store and multimodal models were missing from the usage statistics page. Persist the response usage through a global TokenUsageService reference installed by the desktop runtime.
Dynamic var() string building (--bf-appearance-token-${name}) is invisible to
the theme color audit, which flagged an unresolved required CSS var and broke
the CI Frontend Build job. Replace every token reference with a literal var()
so the audit resolves them statically.
The appearance contract audit (part of build:web) requires every data-bf-component to match a registered descriptor id and every data-bf-part to be a string literal declared on the same node as its component. Align the usage statistics page: component id 'usage-statistics-config', literal parts on root/filters/summary/distributions/modelHitRate/trendPanel/empty, and drop appearance attributes from internal chart components.
Critical/Major/Minor/Nit risk remediation from the usage-statistics-page code review. All 16 findings fixed; scope is limited to the usage statistics feature. Critical (2) - DonutChart SVG geometry: full-ring (100%) slice no longer paints a visible gap; zero-token slices are skipped instead of rendering a full stroke; cumulative arc length avoids floating-point drift. - Aggregation now validates each record for plausibility (non-saturating token counts, non-empty model/session identifiers) and skips garbage. Major (5) - analyze_image persists cache_write_tokens via token_details, and its record_token_usage is now fire-and-forget (tokio::spawn) so the I/O no longer blocks the user-visible response. - token_usage_api enforces a maximum custom range span to guard against pathological queries. - provider_endpoint_path matches the provider key exactly rather than by substring, preventing a 'claude' provider from matching 'anthropic-claude'. - Trend chart tooltip geometry clamps coordinates inside the plot rect. - services-core no longer relies on a tracing dependency for warn logs; uses log::warn! instead. Minor (6) - endpoint_from_base_url refactored into join_endpoint + a known-suffix table; avoids duplicated /v1 segments and duplicated chat-completions. - truncateName uses Intl.Segmenter (grapheme) with Array.from fallback so Chinese/emoji are never sliced mid-codepoint. - formatBucketLabel parses the ISO bucket string directly via regex instead of constructing a Date and re-formatting it. - aggregate_statistics receives &[&TokenUsageRecord] so the trend builder can share the already-filtered list without re-scanning. - Empty session_id/turn_id in analyze_image now synthesise a UUID-based key per call so per-session aggregates remain meaningful. - formatHitRate only reports 100% when Number.isInteger(value) && value===1, so 0.9999... no longer rounds up to 100%. Nit (5) - DonutChart no longer re-runs entries.filter inside the render loop; visible slices are pre-computed once in useMemo. - niceMax uses a NICE_MAX_TABLE loop instead of nested ternaries and adds a Number.isFinite guard. - TREND_SERIES key type extracted to TrendSeriesKey for readability. - cache_write_tokens JSON parsing extracted into extract_cache_write_tokens. - Frontend load() catch block now logs the error via console.error. Verification - cargo check -p bitfun-services-core --features local-storage - cargo check -p bitfun-core --features 'local-storage,workspace-runtime,process-runtime,agent-runtime' - cargo check -p bitfun-desktop - cargo test -p bitfun-services-core --features local-storage token_usage::statistics (8 passed) - cargo test -p bitfun-core --features '...agent-runtime' service::token_usage (8 passed) - pnpm exec tsc --noEmit -p tsconfig.json - pnpm run test:run (3507 passed, 485 files)
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.
Critical/Major/Minor/Nit risk remediation from the usage-statistics-page code review. All 16 findings fixed; scope is limited to the usage statistics feature. See commit message for the full breakdown. Verification: cargo check + cargo test (16 tests passed), pnpm tsc --noEmit + vitest (3507 passed).