Faster, Truer Metrics Charts for Caches, Databases and Rapids
The metrics pipeline behind every chart was rebuilt for accuracy and speed, and the chart catalog grew in the process.
What's changed
- Charts now summarise every sample inside each step, so gauges keep their peaks and rates are averaged properly instead of showing one sample in thirty
- Long ranges render again: 72-hour and 7-day database charts and the 7-day and 30-day Rapids tabs no longer come back empty
- The granularity picker is honoured end to end, results are cached briefly and per-node database queries no longer multiply with the node count, so pages load noticeably faster
- Rapids grows from 6 to 10 charts, including response time p50/p95/p99, responses by status class, an error rate in percent, concurrency and network I/O
- Caches grow from 5 to 14 charts with hit rate, evictions, key counts, command latency, fragmentation and replica lag, and CPU and memory now come from container metrics, which fixes Dragonfly CPU readings
- Databases gain database size, blocked queries, backup archiving, connection pooler charts, and MySQL and MariaDB connection error and temp table charts