Vendor GeoLite2-Country.mmdb in the repository so agent builds still
work when the remote download is unavailable. Update the fetch script
and agent Dockerfile to prefer a fresh download and fall back to the
committed database file.
Agent embeds GeoLite2-Country.mmdb but the file is not checked into the
repository. Download it in CI, Docker, and Makefile build paths via
scripts/fetch-agent-geoip-mmdb.sh, and correct the gitignore exception
path for the geoipdata package.
- Concurrency: Added lock protection to WebSocket writes, fixed timer leaks, and prevented config cache listener context leaks.
- Performance: Added memory cache in ObservabilityBufferStore, periodic cleaning in CH Deduplicator, and buffered ZIP batch download writes.
- Design: Introduced Redis caching for OAuth session/tokens, sanitized raw DB error messages, segregated handlers and logics, and standard CAP response envelopes.
Introduce internal/db/batchwriter as a reusable generic buffered writer
for per-domain ClickHouse flush pipelines, with unit tests and default
batch tuning aligned with audit log ingestion.
Add clickhouse-batchwriter agent skill and cross-references in AGENTS.md
and database-migration. Business layers are not wired yet.
- Return AccessLogView.id as string to avoid JS Number precision loss
- Store OpenFlareAccessLog IDs as uint64 with json id,string
- Update frontend AccessLogItem.id type to string
- Scan ClickHouse count()/countIf() aggregates as uint64 before int64 conversion
- Fix access log list count, region stats, and delete pre-count queries
- Aggregate dashboard 24h traffic totals from hourly trend buckets
- Show current hour value in traffic trend chart summary
- Rename docker-compose service to openflare
- Remove redundant config preview section from performance page
Migrate risk_control audit logs to internal/db/batchwriter and add
openflare/chwriter with per-table async flush for observability
timeseries and node access logs.
Replace single-row ClickHouse inserts and pre-insert SELECT count()
dedup with repository BatchInsert* APIs, in-process TTL dedup for
metric/report snapshots, and bootstrap initialization on API startup.
Introduce internal/db/batchwriter as a reusable generic buffered writer
for per-domain ClickHouse flush pipelines, with unit tests and default
batch tuning aligned with audit log ingestion.
Add clickhouse-batchwriter agent skill and cross-references in AGENTS.md
and database-migration. Business layers are not wired yet.
- Merge all files inside openflare-server to the repository root directory.
- Relocate agent, relay, and flared subprojects from internal/ to internal/apps/.
- Combine docker-compose files and update build context paths to root.
- Update GitHub workflows and Dockerfiles to refer to new directories and package names.
- Rewrite Go package imports across all files.
- Resolve database renew test race condition and clean up docs.
Move all ClickHouse DML for OpenFlare node access logs into
internal/repository/analytics; model layer keeps domain aggregation and
in-memory test store via a thin adapter. Aligns with goose-managed DDL
and openflare database startup dependency.
Add a separate ClickHouse OLAP pipeline with goose/clickhouse DDL as the
sole schema source, model/analytics for ORM mapping, and repository/analytics
for reads (ChDB/GORM) and batch writes (ChConn). Refactor risk_control and
admin/logs to use the repository layer instead of inline SQL. Remove the
manual support-files DDL and document the workflow in database-migration skill.
Move node access log storage from PostgreSQL/SQLite to ClickHouse
(database: openflare). System startup now requires a healthy ClickHouse
connection and auto-initializes the schema. Agent heartbeat writes access
logs via batch insert; goose migration drops the legacy relational table.
Delete the old monolithic openflare-server implementation and rename
Wavelet/ to openflare-server/ to complete the migration consolidation.
Update CI workflows, agent Dockerfiles, and deployment docs for the new
layout (frontend/, docker/Dockerfile).
ryan