Azure Synapse — What We Collect - DQLabs PRIZM

Prizm reads from Synapse system views, the Synapse workspace APIs, and the storage behind your external tables. It never writes to Synapse. One physical table is one Prizm asset, so a table seen from two engines isn’t duplicated.

Catalog
Observability
Lineage
Quality & Criticality

Synapse object	What Prizm stores
Database / pool	Database asset
Schema	Schema asset
Table, view, and column	Table and view assets with each column’s data type and nullability
Distribution (HASH, ROUND_ROBIN, REPLICATE)	Distribution strategy and, for HASH, the distribution key column (Dedicated pool)
Index type	Index type (Dedicated pool)
Row count	Row count, summed across distributions (Dedicated pool)
Keys	Primary key, unique, and foreign key hints. Foreign keys aren’t enforced in Synapse, so Prizm treats them as hints
External table	File format, storage path, and declared columns. The credential is stored as governance metadata only
Delta table	Schema, version, partitions, and commit history
Pipeline	Pipeline, with its folder, annotations, owner, schedule, last run and status, and activity list
Activity	Child of its pipeline
Trigger	Schedule metadata
Data flow	Child of its pipeline

Signal	How Prizm measures it
Volume	Partition statistics on Dedicated pools. On Serverless external tables the count can be estimated
Freshness	Dedicated pools have no native last-modified time, so Prizm compares row-count snapshots between runs. For Serverless external tables it uses the latest file modified time, and for Delta tables the latest commit time
Schema changes	Column differences between runs
Performance and usage	Query duration percentiles, waits, queue contention, and distribution skew on Dedicated pools. Query duration on Serverless
Pipeline runs	Status, duration, activity-level failures, and triggers that have stopped firing

Synapse keeps only recent query history, so Prizm harvests it regularly and keeps its own record.

Prizm builds table-level lineage from these sources, preferring what Synapse declares over what it infers:

Pipeline definitions (Copy and Data Flow activities)
View definitions, foreign keys, and external tables that point at a Delta path
CTAS and CETAS statements found in request history
Delta operation metadata

Lineage across surfaces joins on storage location, so multi-hop lineage and downstream impact are supported. Column-level lineage isn’t covered.

What Prizm Does Not Collect

Prizm doesn’t collect cost data, column-level lineage, or credentials stored in linked services. Request-history retention limits usage and CTAS/CETAS lineage to what Prizm has already harvested. CSV scans can’t prune columns, so they use a stricter bytes budget.

Next Steps

Setup

Grant the permissions each item needs.

FAQ

Common questions about the Synapse connector.