Previously a per-row scan error was logged and skipped with continue, so a
decode failure (e.g. the UInt64->int64 regression) produced a 200 response
with partial or empty data and no signal to the client. Fail the request
instead so the HTTP handlers return 500. Also check rows.Err() after iteration.
The PostgreSQL model migration regenerated LogScore.ID, MonitorID and
ServerID as signed/pgtype fields, but the ClickHouse log_scores columns
are unsigned (id UInt64, monitor_id/server_id UInt32). clickhouse-go
refuses to scan UInt64 into *int64, so every score row failed to parse
and the scores endpoints returned empty.
Scan each row into locals matching the ClickHouse column types, then
convert into the model, mirroring the existing leap handling. Factor the
shared scan into scanLogScore used by both Logscores and
LogscoresTimeRange.
Replace MySQL driver with pgx/v5 and pgxpool:
- Update sqlc to use postgresql engine
- Convert query.sql to PostgreSQL syntax ($1 params, CASE WHEN,
ANY() arrays)
- Replace sql.DB with pgxpool.Pool throughout
- Change nullable types from sql.Null* to pgtype.*
- Update ID types from uint32 to int64 for PostgreSQL compatibility
- Delete MySQL-specific dynamic_connect.go
- Add opentelemetry.gowrap template for tracing
- Add /api/v2/server/scores/{server}/{mode} endpoint
- Support time range queries with from/to parameters
- Return data in Grafana table format for visualization
- Fix routing pattern to handle IP addresses correctly
- Add comprehensive parameter validation and error handling
ask