Skip to main content

table-validator

Cross-platform data migration validator. Compares a source table against a target Databricks table/catalog and reports whether the migration is correct: matching schema, row counts, column statistics, and (where a difference is found) the exact row/column that changed. Produces a multi-sheet Excel report and a pass/fail summary.

Currently supported sources (target is always Databricks):

  • Databricks catalog → Databricks catalog
  • Azure Blob Storage (CSV / Excel / Parquet) → Databricks table
  • Azure SQL Database → Databricks catalog
  • Azure Synapse SQL pool (dedicated or serverless) → Databricks catalog

More source platforms can be added behind the same connector/validator interfaces described below.

Install

pip install table-validator

From a local checkout (the directory containing pyproject.toml):

pip install .

For local development (editable install, so code changes take effect without reinstalling):

pip install -e ".[dev]"

Either way, this installs the tablevalidator command on your PATH.

CLI usage

tablevalidator info

Prints what the tool does and the commands below, in the order you'd normally run them.

tablevalidator configure

Interactive wizard that walks you through:

  1. Azure Storage account + container (optional, skip if you don't have a Blob source) and account key
  2. Azure SQL server + database (optional, skip if you don't have a SQL source) and username/password
  3. Azure Synapse SQL pool endpoint + database (optional, skip if you don't have a Synapse source), then either a SQL login (username/password) or a Microsoft Entra ID service principal (tenant ID + client ID + client secret)
  4. Databricks workspace URL, SQL Warehouse HTTP path, and personal access token
  5. Source table (catalog / schema / table)
  6. Target table (catalog / schema / table)
  7. Which validations to run (catalog / schema / column / row)
tablevalidator validate

Runs the comparison using the saved configuration and writes validation_report.xlsx in the current directory, printing a pass/fail summary to the console. Exit code is 0 if the overall result is PASS, non-zero otherwise (useful in CI). Useful flags:

tablevalidator validate --config-path /path/to/config.yaml --output /path/to/report.xlsx
tablevalidator open

Opens the most recently generated report in your default spreadsheet app.

Notebook usage (zero config, inside Databricks)

From inside a Databricks notebook cell, validate_tables() compares two tables using the notebook's own already-authenticated Spark session - no workspace URL, personal access token, or SQL Warehouse HTTP path to set up. This is a second, independent way to use the package alongside the CLI above (pick whichever fits: the CLI for a scheduled/scripted run against a SQL Warehouse, this for an ad-hoc in-notebook check) - both run the exact same full-depth comparison engine.

%pip install table-validator

from table_validator import validate_tables

result = validate_tables(
    "catalog1.schema1.table1",
    "catalog1.schema1.table2",
)
print(result)  # compact summary
Overall status: FAIL
Tables: 1 total, 0 passed, 1 failed, 0 error, 0 skipped

Call validate_tables.help() any time (no arguments, no live Spark connection needed) to print a full syntax reference for every keyword argument shown below, directly in the notebook cell.

result also exposes each of the Excel report's own sheets as a plain-text table - print only the one you want:

print(result.table_validation)    # one row per table
print(result.column_validation)   # one row per column
print(result.data_mismatches)     # one row per mismatched cell (FULL mode)
print(result.row_hash_mismatches) # one row per mismatched primary key
print(result.mismatch_categories) # root-cause breakdown (NULL_MISMATCH, etc.)
print(result.suggestions)         # plain-English fix suggestions
Source Schema Source Table Overall Status  Row Count (Src)  Row Count (Tgt)  ...
         sch1           t1           FAIL                5              100  ...

Each of these is a small ResultTable object - .headers/.rows for programmatic access, or .to_dataframe() if you want a real pandas DataFrame to filter/sort/export yourself. result.response is the raw CatalogValidationResponse, for full programmatic access beyond the sheet breakdown.

Optional keyword arguments to validate_tables(): primary_key (a real key for row-level comparison instead of the ROW_NUMBER() fallback, single- table mode only), ignore_columns, only_columns, column_map (for a renamed column), ignore_datatype_columns (skip a real type mismatch on these columns rather than failing the table on it). Outside a Databricks notebook (e.g. local development against a real Spark session), install the spark extra: pip install "table-validator[spark]" - inside an actual Databricks notebook, pyspark and an active session already exist, so the bare %pip install table-validator above is sufficient.

Schema-wide sweep (no single table named)

Leave the table off both source/target (just "catalog.schema") to compare every identically-named table in that schema in one call - the same auto-discovery the CLI's own blank-table config triggers, with zero further setup:

result = validate_tables("catalog1.bronze", "catalog2.silver")
print(result)                  # lists every matched table's status
print(result.table_validation) # one row per matched table

If some tables were renamed between source and target, pass table_map (source name → target name) - unmapped tables are still matched by identical name as usual:

result = validate_tables(
    "catalog1.bronze", "catalog2.silver",
    table_map={"cust": "customers", "ord": "orders"},
)

primary_key isn't valid in this mode (a single key can't apply to every table in the sweep) - compare one table at a time ("catalog.schema.table" on both sides) if you need row-level detail via a real key.

Row filtering (compare only matching rows)

Pass row_filter to restrict comparison - row counts, statistics, fingerprint, and row-hash/column diff - to only the rows matching a SQL condition, instead of the whole table:

result = validate_tables(
    "catalog1.bronze.orders", "catalog2.silver.orders",
    row_filter="id > 20 and id < 100",
)
result = validate_tables(
    "catalog1.bronze.customers", "catalog2.silver.customers",
    row_filter="gender = 'male'",
)

If the condition needs to differ per side, use source_row_filter/ target_row_filter instead (or alongside row_filter, which then applies to both and combines with AND):

result = validate_tables(
    "catalog1.bronze.orders", "catalog2.silver.orders",
    source_row_filter="id > 20",
    target_row_filter="id > 15",
)

Each is a raw SQL WHERE-clause fragment, used as-is - not parsed or validated, so a typo surfaces as a normal SQL error when the query runs. Works in both single-table and schema-wide sweep mode (e.g. filtering every matched table in a sweep by the same condition).

Quickstart (Python API)

Everything the CLI does is available as a library, built from the same public API exported by table_validator/__init__.py:

from table_validator import (
    load_config,
    CatalogValidator,
    CatalogValidationRequest,
    DatabricksConnector,
)
from table_validator.auth.databricks_auth import get_databricks_token

# Non-secret settings from ~/.table_validator/config.yaml
# (see `tablevalidator configure`); secrets from ~/.table_validator/.env.
config = load_config()
token = get_databricks_token(config)

# DatabricksConnector wants a bare hostname, not the full workspace URL
host = config.databricks.workspace_url.replace("https://", "").split("/")[0]

databricks = DatabricksConnector(
    host=host,
    token=token,
    http_path=config.databricks.http_path,
)

validator = CatalogValidator(databricks)

request = CatalogValidationRequest(
    source_catalog="source_catalog_name",
    target_catalog="target_catalog_name",
    schemas=["sales"],          # optional: restrict scope
    primary_keys={"sales.orders": ["order_id"]},  # optional: enables row-level diffing
)

result = validator.compare_catalogs(request)

print(result.status)  # PASS / FAIL / ERROR / SKIPPED

Other public entry points exported from table_validator:

  • AzureCsvValidator / AzureSqlValidator — the Blob-CSV and Azure-SQL equivalents of CatalogValidator, returning the same CatalogValidationResponse shape.
  • BlobCatalogValidator — validates every file in an Azure Blob container against like-named Databricks tables.
  • AzureConnector / AzureSqlConnector — the Azure-side connectors, for building your own validation flow against a connector directly.
  • ValidatorConfig, default_config, save_config, require_config, ConfigNotFoundError — the same config load/save layer the CLI wizard uses, if you want to construct or persist configuration programmatically.
  • validate_tables / SparkConnector — the notebook-native entry point described above (lazily imported: importing table_validator itself never requires pyspark to be installed).

Where config and secrets are stored

Everything lives outside the repo, under your home directory:

  • ~/.table_validator/config.yaml — non-secret configuration (table references, workspace URL, which validations are enabled). Safe to inspect or version-control separately if you want.
  • ~/.table_validator/.env — credentials (Azure Storage key, Azure SQL username/password, Databricks personal access token), written in plaintext and restricted to owner read/write only (chmod 600, best-effort on Windows since NTFS doesn't map POSIX permission bits). Never commit this file or add it inside the project repo — it isn't, by construction, since it's written under your home directory rather than the working directory.

Optional: DATABRICKS_RETRY_TIMEOUT_SECONDS (a plain environment variable, not part of config.yaml) raises the CloudFetch HTTP retry timeout above its 300-second default. On a slow or unstable network, downloading a large row-hash result set can legitimately take longer than that, causing a Retry request would exceed Retry policy max retry duration failure even though the query itself succeeded. Set it higher (e.g. 900 for 15 minutes) if you hit this on large tables.

A future version will replace manual credential entry with Azure CLI / Service Principal auth and Databricks CLI / OAuth login, without changing the config file format or any command usage above.

Development

pip install -e ".[dev]"
pytest

License

MIT — see LICENSE.

Metadata

Release files for table-validator 0.1.23

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for table-validator 0.1.23
File Size Uploaded Provenance Actions
table_validator-0.1.23.tar.gz 214.1 kB With provenance Details for table_validator-0.1.23.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for table-validator 0.1.23
File Interpreter ABI Platform Actions
table_validator-0.1.23-py3-none-any.whl Python 3 none any Details for table_validator-0.1.23-py3-none-any.whl

Total release size: 367.5 kB

Release files / table_validator-0.1.23.tar.gz

Download URL table_validator-0.1.23.tar.gz
Size 214.1 kB
Tags Source
SHA-256 checksum
How to use checksums
23a45ebbce26cbfff44e331c2a7cbb065869f6816cdb1162272c9b30045e0b30
BLAKE2b-256 checksum
How to use checksums
5a2838e2dab76ff3c6ffbb65dc9fb46ca1d3b7e7bb4d4a548509cd4670e3efdc
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 10, 2026.

Transparency log

Release files / table_validator-0.1.23-py3-none-any.whl

Download URL table_validator-0.1.23-py3-none-any.whl
Size 153.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
e21027e4b9cad07f8131e455563881cae7161d794fc00ec5ce7619bdb965118f
BLAKE2b-256 checksum
How to use checksums
e1130a0868a16b4364e30808999e580e476ea6f86f413935dd32fce8959b0eb4
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 10, 2026.

Transparency log

Security for table-validator 0.1.23

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for table_validator-0.1.23.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 10, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page