Data Observability in the PI System: The Four Pillars

Data observability gives PI teams continuous visibility into the health, context, use, and behavior of operational data.

For a PI environment, four practical pillars are data quality, metadata and configuration, usage and lineage, and operational logs and change history.

1. Data quality

Data-quality checks evaluate whether the signal behaves as expected.

Useful checks include:

• Freshness

• Bad states

• Data gaps

• Unexpected flatlines

• Out-of-range values

• Abnormal update volume

No single check proves that data is correct. A tag can be current and numerically valid while it is mapped to the wrong source.

2. Metadata and configuration

Metadata explains what a signal means and how PI handles it.

Important fields can include:

• Description

• Engineering units

• Span

• Point source

• Source address

• Exception settings

• Compression settings

• AF mappings

Configuration drift can change data behavior without changing the physical process. Observability must therefore include configuration as well as values.

3. Usage and lineage

Usage shows where a data object is consumed. Lineage shows how data moves and transforms across the system.

Useful questions include:

• Which displays use this tag?

• Which AF attributes reference it?

• Which analyses depend on it?

• What produces this calculated output?

• What can break if the source changes?

This context supports troubleshooting, cleanup, and impact analysis.

4. Logs and change history

Logs provide evidence of runtime problems. Change history provides evidence of configuration changes.

Together they help answer:

• When did the problem start?

• What changed before the problem?

• Which service or analysis reported an error?

• Who changed the configuration when that information is available?

Do not collect logs without a clear use. Focus on events that help teams detect and diagnose operational data problems.

Combine the pillars

The four pillars are most useful when they work together.

A stale critical tag is more important when lineage shows that it feeds several analyses and operator displays. A failed analysis is easier to investigate when change history shows that an input mapping changed shortly before the failure.

This combined context reduces time spent moving between tools and reconstructing the event manually.

The objective

Data observability does not replace PI System administration. It makes administration more systematic.

A mature program helps the team detect important issues early, understand what is affected, and identify the likely cause with less manual investigation.