Skip to content
AI usage dashboard

M5 telemetry and historical remaining work: Second checks, 2026-09-29

This page retains that day’s methods/results; the remaining-work table reflects that stage. Later native files/field rules/rate limits/background/performance results: latest acceptance. Current work: Plan. Receiver settings below describe the historical implementation; current authentication rules are in receiver authentication.

Scope: address previous inventory’s outstanding items: M5 local Copilot CLI/OTel spans/OTLP receiver, M1 pre-migration backups, M6 per-source schedules, V20 initial performance, M7 idle memory recheck. No commit/push/deployment.

Local data changes: Recheck, 2026-09-29

  • Copilot CLI data restored: ~/.copilot/session-store.db schema_version=8, assistant_usage_events 36 native rows dated 2026-08-03, claude-opus-4.8. Entire tree absent on 2026-09-25 during upgrade migration, now written back.
  • Claude ~/.claude only empty sessions/backups, without projects transcripts; native samples unavailable.
  • CodeBuddy ~/.codebuddy diagnostics/logs/memwatch only, no usage files. Official monitoring documents CLI OTLP export only, without file exporter.
  • APPDATA WorkBuddy/WorkBuddy AI directories reappeared but empty. Still no local format to verify; M4 paused state unchanged. Directory reappearance recorded separately.
  • VS Code Copilot Chat globalStorage session-store.db contains only sessions/turns/session_files/ session_refs, no usage table, matching the then-current A06 conclusion about telemetry requirements.

Later source investigation narrows the CodeBuddy conclusion to this local directory state and OTel file exporter. Official directory docs separately establish ~/.codebuddy/projects session JSONL; third-party source identifies usage fields and WorkBuddy .workbuddy/projects. Both implemented from local-session references, initially without native samples.

M5 implementation

1. Copilot CLI adapter: Native data check PASS

adapters/copilot/ common/detect/versions reads per-turn assistant_usage_events. map_copilot moved from root usage_map.rs into its product directory, adding reasoning, per V30. schema_version=8 KnownVersion; others latest-compatible.

  • Native redacted tests/fixtures/copilot/assistant-usage-events-v8.sanitized.json: 36 rows of permitted columns, anonymous session_id/agent_id; expectations in _expectations.md.
  • copilot_contract.rs rebuilds SQLite from the sample, calls run_adapter_scan, and matches manual input 4,649,981/output 34,157/cache read 4,416,791/cache write 233,118/reasoning 17,678/uncached 72/total 4,684,138. Repeat scanning adds nothing.
  • examples/real_verify_copilot.rs, desktop/src-tauri cwd: cargo run -p llm-usage-core –example real_verify_copilot – “$USERPROFILE/.copilot” build/m5-copilot/verify. 36 events, fields match independent SQL, stable repeat, VERDICT: PASS.
  • All 36 rows satisfy input≥cache_read+cache_write, verifying the M0 rule on actual data. request_multiplier=27.0 premium multiplier and total_nano_aiu excluded from tokens.

2. OTel spans JSONL adapter: Document-based

adapters/otel/ reads three formats: VS Code Copilot Chat file exporter github.copilot.chat.otel., official bdc5ebe NDJSON rather than OTLP, startTime seconds/nanoseconds pair; Copilot CLI file exporter, undocumented per-line schema with related-format tolerance awaiting samples; and this app receiver’s normalized exports, CodeBuddy agentlens unprefixed usage.. Skip aggregate invoke_agent/codebuddy_code.interaction spans and model_request per official double-count warning. Permitted attributes; accept both trace/span ID spellings since docs do not specify exact keys.

3. Local OTLP/HTTP receiver: src-tauri/src/otel_receiver.rs

  • 127.0.0.1 loopback only; otel_receiver_enabled off by default, otel_receiver_port=4318. At this stage, enabling represented explicit local-instance authorization, V25.
  • OTLP/JSON and OTLP/protobuf; CodeBuddy protobuf-only. Explicit wire parsing ResourceSpans→ ScopeSpans→Span→KeyValue→AnyValue. Bounded gzip decompression 64 MiB, headers 64 KiB/body 64 MiB, V22 compressed-input protection.
  • Save permitted gen_ai./usage./model* prefixes; explicitly exclude message-text keys such as gen_ai.input.messages; strings at most 256 B. Output APPDATA/llm-usage-desktop/otel/spans.jsonl, automatically discovered by OTel adapter.
  • JSON/protobuf unit tests and actual E2E: receiver→HTTP POST /v1/traces→saved output→OTel reads codebuddy model_stream event, HTTP 200.
  • Debugging fixed two defects: accepted connections from nonblocking listener need explicit blocking restoration; HTTP header separator now byte array [0x0d,0x0a,0x0d,0x0a].

M1 remaining work: Consistent backup and space checks before migration/rebuild/clear

src-tauri/src/db_backup.rs plus fs4 dependency:

  • consistent_backup uses VACUUM INTO for a WAL-consistent snapshot. Check free space at least 1.1×database including -wal, keep latest three backups, ordered by timestamp strings.
  • app_state.rs schema-mismatch rebuild first calls read-only consistent_backup_legacy; failure aborts with original intact. clear_all_data uses shared helper with space check, replacing inline backup_before_clear.
  • Pure space-check tests, independently reopenable backup round trip and retention of three. Windows open SQLite connections prevent deletion; test scopes now close them. fs4 ^0.13.

M6 remaining work: Per-source schedules, extraction_schedules

  • core/src/schedules.rs SourceScheduleRule: interval 15 seconds–24 hours/daily HH:MM/ weekly ISO weekday+HH:MM; initial release without arbitrary cron. Pure next_due with jiff timezone and then-current DST-ambiguity +24h fallback. upsert/delete/read/due_instances/ custom_scheduled_instances/mark_source_run with schedule_state. Five boundary/daily/weekly crossing/disabled/storage tests.
  • Framework run_adapter_scan_filtered uses InstanceFilter include/exclude; original run_adapter_scan signature retained as delegation.
  • scanner.rs global refresh excludes custom schedules. Poll due instances every 500 ms, trigger individually. FixedTime shares refresh with global/manual requests and merges same-source jobs without concurrent execution. Advance next_due after running; missed times on wake cause one scan.
  • set_source_schedule registered in invoke_handler; list_sources returns schedule/nextDueMs. SourceList.svelte per-source editing: inherit global, interval presets 15 seconds–24 hours, daily/weekly time-day selectors and next extraction. Weekday names use locale Intl. Six i18n keys across ten languages, positional locales.ts insertion plus zh-CN/en.
  • scheduling.md status updated to implemented.

V20 initial performance: M1/M7 baseline measurements

examples/bench_v20.rs, release, configurable size, desktop/src-tauri cwd:

Scale Insert throughput Database + WAL Daily-query p50/p95/p99 Detail pagination, 200 rows, p50/p95/p99
One million events, 366 days/4 models/6 agents/4 sources 30,453/s, 32.8 s 676.8 MiB + 34.9 MiB 0.01/0.01/0.01 ms 0.67/1.24/1.52 ms
Ten million events 23,560/s, 424.4 s 6,788.3 MiB + 35.6 MiB 0.01/0.01/0.02 ms 2.28/4.65/5.41 ms

No unbounded whole-file loading; indexed pagination. Ten-million temporary database cleaned.

M7 resource recheck: Partial V21 measurements

Release LLMUsage.exe GUI startup, headless scan, then idle: primary working set 32.5 MB plus WebView2 children 138.5 MB, about 171 MB, within 180 MiB target. Earlier M0 minimal-shell 206.5 MB exceeded target; this measurement meets it. First-screen P95 and actual install/ uninstall remained M7 outstanding at this stage.

Tests and commands

  • Repository-root npm run verify passes all checks: Markdown/Svelte zero errors or warnings, Cargo fmt/Clippy -D warnings/Vite build.
  • desktop/src-tauri cargo test: 588+ pass. New Copilot format test 1/schedules 5/db_backup 2/otel_receiver 4 including E2E, OTel adapter units and migrated mapping_v01 pass. V30 covers Copilot/OTel directories and moving map_copilot from the root.
  • git diff –check passes; temporary files under ignored build/.

Remaining work at this stage

Item Status Missing requirement
M5 native VS Code/Copilot CLI file exports Await user enablement Export must be enabled; per-line schema needs samples
M5 actual CodeBuddy CLI to receiver E2E Await receiver enablement/CodeBuddy use; local sessions in record Receiver implemented/E2E with synthetic payload; overlap with sessions unresolved
M2 native Claude/Gemini/Qwen samples No local data Claude recheck finds empty directories only
M4 WorkBuddy Local-session adapter registered; later native eight-file/420-event check passed Trace/session relationship and other versions unchecked
M3/M4/M8 native data acceptance Deferred Products not installed/no data locally
F2 cost engine Main implementation complete, record Optional online refresh/usage-cost alerts/native E2E remain
M6 actual desktop V13–V18/V23–V25 and V24 system tasks Await native operations Manual/automated GUI records needed
M7 actual install/upgrade/uninstall V19, first-screen P95 V21/offline Await actual environment checks —
V26 first three-platform CI Record after push Push not authorized at this stage
File-change triggers Not implemented; optional optimization, polling already allowed Scheduling rules permit polling first