Data Reconciliation
How does cross-database diffing work?
How does cross-database diffing work?
Datafold’s data diff connects to source and target databases and performs fast, accurate and detailed comparison of datasets providing aggregate summaries, column, and value-level insights into any discrepancies.
- Datafold connects to any SQL source and target databases, similar to how BI tools do.
- Datafold does not need to extract the entirety of the datasets for comparisons. Instead, Datafold relies on stochastic checksumming to identify discrepancies and only extract those for analysis.
What kind of information does Datafold output?
What kind of information does Datafold output?
Datafold’s cross-database diffing will produce the following results:
- High-Level Summary:
- Total number of different rows
- Total number of rows (primary keys) that are present in one database, but not the other
- Aggregate schema differences
- Schema Differences: Per-column mapping of data types, column order, etc.
- Primary Key Differences: Sample of specific rows that are present in one database, but not the other
- Value-Level Differences: Sample of differing values for each column with identified discrepancies; full dataset of differences can be downloaded or materialized to the warehouse
You can check out what the results look like in the App.
How does a user run a data diff?
How does a user run a data diff?
- Via Datafold’s interactive UI
- Via the Datafold API
- On a schedule (as a monitor) with optional alerting via Slack, email, PagerDuty, etc.
Can I run multiple data diffs at the same time?
Can I run multiple data diffs at the same time?
Yes, users can run as many diffs as they would like with concurrency limited by the underlying database.
How can I ensure accurate data comparison if my data is changing and being replicated in real-time?
How can I ensure accurate data comparison if my data is changing and being replicated in real-time?
In such cases, we recommend using watermarking – diffing data within a specified time window of row creation / update (e.g. updated_at timestamp
).
What if the data types do not match between source and target?
What if the data types do not match between source and target?
Datafold performs best-effort type matching for cases when deterministic type casting is possible, e.g. comparing VARCHAR
type with STRING
type. When automatic type casting without information loss is not possible, the user can define type casting manually using diffing in Query mode.
Can data diff help if the source and target datasets have a different shape/schema/column naming?
Can data diff help if the source and target datasets have a different shape/schema/column naming?
Yes, users can reshape the input dataset by writing a SQL query and diffing in Query mode to bring the dataset to a shape that can be compared with another. Datafold also supports column remapping for datasets with different column names between tables.
How can data diffs be provisioned at scale, e.g. we need to create hundreds / thousands of data diffs?
How can data diffs be provisioned at scale, e.g. we need to create hundreds / thousands of data diffs?
To make the provisioning at scale easier, you can create data diffs via the Datafold API.