TSDB Architecture

Runtime Components

• ProxySQL_Admin main loop triggers three TSDB schedulers:
  • sampler (tsdb_sampler_loop)
  • downsampler (tsdb_downsample_metrics)
  • backend monitor (tsdb_monitor_loop)
• Data is stored in statsdb_disk SQLite tables.

Data Path

1. Sampler collects all metric families from GloVars.prometheus_registry->Collect().
2. Samples are normalized and inserted into tsdb_metrics.
3. Hourly job aggregates into tsdb_metrics_hour.
4. Backend monitor probes servers from runtime_mysql_servers and stores in tsdb_backend_health.
5. Retention cleanup removes old raw/probe data using configured days.

Metric Family Handling

• Counter/Gauge/Untyped/Info: one sample per metric point.
• Summary: quantiles plus _sum and _count companion metrics.
• Histogram: _bucket{le=...} plus _sum and _count companion metrics.

Storage Schema

• tsdb_metrics: PK (timestamp, metric_name, labels)
• tsdb_metrics_hour: PK (bucket, metric_name, labels)
• tsdb_backend_health: PK (timestamp, hostgroup, hostname, port)

Configuration Lifecycle

TSDB configuration is handled via standard global variables with the tsdb- prefix. Changes are applied and persisted using the following dedicated administrative commands:

• LOAD TSDB VARIABLES TO RUNTIME
• SAVE TSDB VARIABLES TO DISK

These commands function similarly to their MYSQL and PGSQL counterparts, moving configuration between memory and disk storage.