feat(visits): classify browser and OS families in the worker #9

Merged

eleith merged 1 commits from feat/visit-classification into main 2026-11-04 23:55:30 +00:00

Conversation 0 Commits 1 Files Changed 12

+205 -17

eleith-gitea-agent Bot commented 2026-11-04 23:48:48 +00:00

Copy link

Copy source

Summary

• Add a small dependency-free user-agent classifier for broad browser and OS families. Edge variants share one label; unsupported or absent input becomes Other. No versions, device/bot detection, or growing parser framework.
• Classify inside the existing worker's recordVisit task immediately before the single insert. HTTP capture and the raw queued event shape remain unchanged.
• Add versioned migration 0002_visit_classification for stored browser and os labels. Existing visits and raw headers are preserved; existing labels start as Other, with no automatic backfill job.
• Distinguish raw NewVisit input from a prepared ClassifiedVisit database insert.
• Make the native SQLite dialect advertise transactional DDL through Kysely's public adapter API. This keeps migration schema changes and bookkeeping atomic rather than leaving a partial multi-statement upgrade after failure.

Scope

Classification only. No parser dependency, second insert, enrichment queue/marker, service, batching, retention-policy changes, invalid/base-visit collection, dashboard, visit-row UI, or export feature. Those analytics requirements remain separate work.

Classification is intentionally approximate: browsers may conceal their identity, and unsupported cases may fall into another broad family or Other.

Verification

• Separate final readability and correctness re-reviews are clean after the approved migration-atomicity correction.
• pnpm verify: schema drift, all workspace builds/types, formatting/ESLint, 420 unit tests and 7 integration tests passed; Svelte reports zero errors/warnings.
• Focused cases cover overlapping browser/OS tokens, Edge-family collapse, missing/unsupported input, legacy-row preservation, and failure → atomic rollback → successful migration retry.
• The existing real-worker HTTP/lock/drain story now checks that submitted Firefox/Linux user-agent data produces stored family labels; no additional HTTP story was added.
• Offline pruned production artifact tested outside the repository: the worker resolves inside the deployment, persists Firefox/Linux labels, and exits naturally on SIGTERM with code 0.
• DB package lint, Changesets status, and Git whitespace checks passed.

Agent-run Docker/Compose verification was not performed; container checks remain owner-only. Remote CI is separate from the passing local checks. Transactional migrations do not claim coordination between independently started processes.

## Summary - Add a small dependency-free user-agent classifier for broad browser and OS families. Edge variants share one label; unsupported or absent input becomes `Other`. No versions, device/bot detection, or growing parser framework. - Classify inside the existing worker's `recordVisit` task immediately before the single insert. HTTP capture and the raw queued event shape remain unchanged. - Add versioned migration `0002_visit_classification` for stored `browser` and `os` labels. Existing visits and raw headers are preserved; existing labels start as `Other`, with no automatic backfill job. - Distinguish raw `NewVisit` input from a prepared `ClassifiedVisit` database insert. - Make the native SQLite dialect advertise transactional DDL through Kysely's public adapter API. This keeps migration schema changes and bookkeeping atomic rather than leaving a partial multi-statement upgrade after failure. ## Scope Classification only. No parser dependency, second insert, enrichment queue/marker, service, batching, retention-policy changes, invalid/base-visit collection, dashboard, visit-row UI, or export feature. Those analytics requirements remain separate work. Classification is intentionally approximate: browsers may conceal their identity, and unsupported cases may fall into another broad family or `Other`. ## Verification - Separate final readability and correctness re-reviews are clean after the approved migration-atomicity correction. - `pnpm verify`: schema drift, all workspace builds/types, formatting/ESLint, **420 unit tests and 7 integration tests passed**; Svelte reports zero errors/warnings. - Focused cases cover overlapping browser/OS tokens, Edge-family collapse, missing/unsupported input, legacy-row preservation, and failure → atomic rollback → successful migration retry. - The existing real-worker HTTP/lock/drain story now checks that submitted Firefox/Linux user-agent data produces stored family labels; no additional HTTP story was added. - Offline pruned production artifact tested outside the repository: the worker resolves inside the deployment, persists `Firefox`/`Linux` labels, and exits naturally on SIGTERM with code 0. - DB package lint, Changesets status, and Git whitespace checks passed. Agent-run Docker/Compose verification was not performed; container checks remain owner-only. Remote CI is separate from the passing local checks. Transactional migrations do not claim coordination between independently started processes.

eleith-gitea-agent Bot added 1 commit 2026-11-04 23:48:48 +00:00

feat(visits): classify browser and OS families in the worker 0ef4ffc93e

eleith merged commit 0ef4ffc93e into main 2026-11-04 23:55:30 +00:00

eleith deleted branch feat/visit-classification 2026-11-04 23:55:30 +00:00

Sign in to join this conversation.

Reviewers

No Reviewers

Labels

No items

No labels

Milestone

No items

No Milestone

Assignees

Clear assignees

eleith-gitea-agent Bot eleith (eleith)

No Assignees

1 Participants

Notifications

Subscribe

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: eleith/goto#9