SYSTEM Cited by 1 source
AWS Performance Insights¶
What it is¶
AWS Performance Insights is the managed observability
surface Amazon RDS exposes for database performance monitoring.
It collects time-series metrics at two altitudes — OS-level
metrics under the os.* namespace (CPU utilisation, swap
activity, block-device IO, etc.) and database-engine-level
metrics under the db.* namespace (buffer-cache hit counts,
block read time, deadlock counter, transaction counts, SQL
row-fetch ratios). Metric paths follow a nested-name convention
(e.g. os.cpuUtilization.total, db.Cache.blks_hit,
os.diskIO.rdsdev.await).
Why it shows up on this wiki¶
Performance Insights is the data source for Zalando's concepts/golden-signals-rds|12 Golden Signals methodology and the underlying API that systems/rds-health queries to produce fleet-wide health reports. (Source: sources/2024-02-19-zalando-twelve-golden-signals.)
The canonical metric paths Zalando uses:
| Bucket | Metric name | Purpose |
|---|---|---|
| CPU | os.cpuUtilization.total |
Total CPU utilisation |
| CPU | os.cpuUtilization.await |
Linux IO-wait CPU fraction |
| Memory | os.swap.in, os.swap.out |
Swap activity (memory pressure) |
| Disk | os.diskIO.rdsdev.readIOsPS |
Read IOPS on RDS data volume |
| Disk | os.diskIO.rdsdev.writeIOsPS |
Write IOPS on RDS data volume |
| Disk | os.diskIO.rdsdev.await |
Block-device IO latency |
| DB | db.Cache.blks_hit / db.IO.blk_read |
Postgres buffer-cache hit ratio |
| DB | db.IO.blk_read_time |
Time spent reading blocks from storage |
| DB | db.Concurrency.deadlocks |
Deadlock count |
| DB | db.Transactions.xact_commit |
Committed transactions |
| DB | db.SQL.tup_fetched / db.SQL.tup_returned |
SQL efficiency ratio |
Relationship to CloudWatch¶
Performance Insights and CloudWatch are complementary. CloudWatch exposes coarser-grained metrics at host and service altitude; Performance Insights adds the engine-internal metric surface (Postgres/MySQL buffer-cache counters, SQL stats, deadlock counts) that CloudWatch alone cannot see. Zalando's 12-signal methodology draws primarily from Performance Insights for this reason — 5 of the 12 signals are engine-internal and only available via this API.
Seen in¶
- sources/2024-02-19-zalando-twelve-golden-signals —
Zalando's methodology queries Performance Insights for all
12 golden signals; the specific metric-path strings in the
article (
os.diskIO.rdsdev.await,db.Cache.blks_hit, etc.) are Performance Insights names.
Related¶
- systems/aws-rds · systems/aws-cloudwatch · systems/postgresql
- systems/rds-health — Zalando CLI that queries this API
- concepts/golden-signals-rds — the methodology built on this surface
- concepts/observability