Insights

Spot the drift before the incident.

Insights learns how each of your databases normally behaves and tells you the moment one starts to drift: a query slower than its own history, an unusual workload, a wasteful access pattern. You hear about it before your users do.

What Insights watches

Catch how your database drifts.

Insights works on the statistics the agent already sends. Nothing extra runs on your databases.

Workload anomalies

Insights knows what a normal Monday morning looks like for each database. When the mix of reads, writes, connections and transactions moves away from its usual shape, you see it while it is still unusual, not after it becomes an outage.

Query regressions

Every query is compared with its own history at the same time of day, so a nightly batch is not news and a real slowdown is. Only changes that cost real database time are reported.

Data access patterns

The same small query fired again and again, one row at a time, usually means a loop in the application that should be one batched query. Insights points to the query, so the fix lands in the right place.

How a finding reads

What changed, why, and what to do.

Every finding opens the same way, so a DBA on call at 3 a.m. knows where to look. Severity and confidence are separate: severity is how big the change is, confidence is how much history stands behind the comparison.

  • ActiveDetected in the latest analysis. Recurrences are counted on the same finding instead of opening new ones.
  • No longer detectedNot seen in recent analyses. The app says what that can mean: fixed, gone quiet, or absorbed as the new normal. If it returns, it comes back as the same finding.
  • DismissedAdmins and DBAs can dismiss a finding with an optional reason, and it stays quiet for a while.
Query regression orders_by_customer orders · pg-prod-01 · High confidence · 12× · 8 min ago
+42%

What changed

Mean latency
38 ms → 54 ms
I/O per call
210 → 430 blocks
Deviation
3.4σ

Why

Data growthRows read per call doubled against the baseline.

Recommended fix

  1. Review the execution plan for this query to check for plan changes.
  2. Check indexes on the filtered columns.
First seen 2 h ago · Last seen 8 min ago · 12 occurrences Open databaseDismiss
Sample data. Layout as in the app; the app shows no chart inside a finding.
Signal over noise

Fewer findings, each one worth reading.

A detector that fires on every wobble trains people to ignore it. These guards exist because each of them removed a class of findings nobody acted on.

Further reading: How we stopped one anomaly finding from firing hundreds of times

Time of day

Comparisons follow the rhythm of your day, so the morning peak is not a regression every morning.

Recent trend

A change is reported once, when it starts, not again every day after.

Materiality

Changes too small to cost anything are ignored, so every finding is worth a look.

Known patterns

Routine traffic from monitoring tools and schedulers is recognised and left out.

Statistics resets

When a database's statistics are reset, Insights notices and stays quiet instead of raising false alarms.

Missing data never silences

When there is not enough history yet, the safeguards step aside. A finding is never hidden because of it.

Plainly

Insights in short.

  • Learns each database's normal on its own. There are no thresholds to maintain.
  • Included for every connected database. There is nothing to configure.
  • Honest about the end of a finding: when it stops firing, it is marked as no longer detected, not as fixed.
  • Points you to the exact query and what changed, so you know where to start.

See what your databases have been doing.

Insights learns from each database's own history and gets sharper every day. Start a trial and see what it finds on yours.