great_expectations

Compatibility: Python <3.14,>=3.10 → <3.15,>=3.10; numpy added (python_version >= "3.14"); pandas added (python_version >= "3.14"); pydantic added (python_version >= "3.14")

Highlights

• Python 3.14 support — Great Expectations now installs and runs on Python 3.14, and the supported range is Python 3.10 through 3.14. On Python 3.14 installs resolve numpy 2.3.2 or later, pandas 2.3.3 or later, and pydantic 2.13.0 or later. The clickhouse and teradata extras are not supported on 3.14, and Python 3.15 and later still require GX_PYTHON_EXPERIMENTAL. (#12280)

  pip install great_expectations  # now works on Python 3.14
  

• New Expectation: ExpectColumnTypeToBe — A schema-level Expectation that checks a column's declared type, separate from the row-level ExpectColumnValuesToBeOfType. It works on pandas, SQLAlchemy and Spark, returns only observed_value in its result, and reports success=False with observed_value=None for a missing column. Note that pandas 3 infers different default dtypes than pandas 2 for string and datetime columns, so re-check type_ when upgrading pandas. (#12186)

  import great_expectations.expectations as gxe
  
  batch.validate(gxe.ExpectColumnTypeToBe(column="order_id", type_="int64"))
  

• Escape character for the LIKE pattern Expectations — The four LIKE pattern Expectations accept an optional single-character escape, which emits an ESCAPE clause so _ and % can be matched literally and identically across backends. Omitting it produces exactly the SQL generated before. BigQuery and ClickHouse raise a clear error naming the dialect, because neither supports a usable ESCAPE clause; escape wildcards with a backslash inside the pattern there instead. Both renderers now disclose the escape character. (#12154)

  import great_expectations.expectations as gxe
  
  gxe.ExpectColumnValuesToMatchLikePattern(column="c", like_pattern="a!_b", escape="!")
  

• Objects now resolve through the Data Context that owns them — With more than one Data Context in a process, expectation suites, validation definitions, checkpoints, batch definitions, the validator and a checkpoint's actions now read, write and run against the context they came from rather than whichever context happens to be current. Data Docs are built into the owning context's site, ${VARIABLE} references in Slack, Teams and Email settings resolve from the owning context's configuration, and a "not found" error for an object that belongs to no context now names the context the lookup consulted. An object that belongs to no context still resolves through the current one. (#12297, #12298, #12302, #12303)

  c1 = gx.get_context(mode="ephemeral")
  suite = c1.suites.add(gx.ExpectationSuite(name="my_suite"))
  c2 = gx.get_context(mode="ephemeral")  # c2 is now current
  suite.add_expectation(...)  # still lands in c1
  

• Database URLs are masked without being mangled — PasswordMasker.mask_db_url keeps every part of a database URL except the secrets: query strings, fragments, IPv6 brackets and empty authorities survive, absent usernames and hosts are no longer printed as None, and passwords containing /, ? or # no longer raise. Credentials passed as query parameters (password=, odbc_connect=, private_key_file_pwd=) are now masked on both code paths, with an allowlist of safe-to-display parameter names, and any value that is not a <scheme>://... URL is masked whole. ([#12242](https://github.com/fivetran

Compatibility: sqlalchemy now <2.1 (extras snowflake, databricks)

Highlights

• Fixes GX on SQLAlchemy 2.1 — SQLAlchemy 2.1.0, released 2026-09-24, broke GX 1.23.1 and earlier on Python 3.11+, where every SQL extra resolves it by default. Depending on the backend, import great_expectations failed whenever snowflake-sqlalchemy was installed, every Databricks query failed, driverless postgresql:// URLs could not load a driver, BigQuery queries comparing against a float failed, SQL Server reported mixed-case and upper-case tables as missing, and expect_column_values_to_be_of_type(type_="Numeric") failed on float columns. 1.23.2 fixes all of these: the snowflake and databricks extras stay below SQLAlchemy 2.1 until their dialects support it, and every other SQL extra runs on 2.1. Python 3.10 is unaffected, since SQLAlchemy 2.1 requires Python 3.11. If you can't upgrade yet, pin sqlalchemy<2.1; do the same if you install snowflake-sqlalchemy or databricks-sqlalchemy outside GX's extras. (#12269)

  pip install --upgrade 'great_expectations[snowflake]'  # include your extras so the SQLAlchemy cap applies
  

• Regex Expectations work on ClickHouse — The four regex Expectations now run on ClickHouse, which does not support regexp_like(). ClickHouse is covered by integration tests for these Expectations. (#12222)

  gx.expectations.ExpectColumnValuesToMatchRegex(column="name", regex="^A")
  

• Correct substring matching for regex Expectations on Snowflake — Snowflake's REGEXP operator anchors patterns to the whole value, so unanchored patterns behaved differently there than on other backends. Regex Expectations on Snowflake now use substring semantics that match every other GX backend, while preserving the user's pattern. (#12221)

  gx.expectations.ExpectColumnValuesToMatchRegex(column="name", regex="ell")
  

• Nested struct columns supported in Spark value-counts Expectations — On the Spark engine, Expectations that rely on the column.value_counts metric — including ExpectColumnMostCommonValueToBeInSet and ExpectColumnKLDivergenceToBeLessThan — now work for dotted nested struct paths such as address.city, instead of returning an empty result with an unresolved-column error. (#12231)

  gx.expectations.ExpectColumnMostCommonValueToBeInSet(column="address.city", value_set=["Springfield"])
  

Changes

Bug fixes

• GX works on SQLAlchemy 2.1, which broke 1.23.1 on several backends: the snowflake and databricks extras are capped below 2.1, a broken snowflake-sqlalchemy install no longer prevents import great_expectations, driverless postgresql:// URLs fall back to psycopg2 when psycopg is unavailable, BigQuery renders Double as FLOAT64, SQL Server reflects mixed-case tables, the Numeric type name matches float columns again, and database URL masking keeps the database and query string verbatim. (#12269)
• Validation results containing an infinite Decimal value — for example the maximum, mean or sum of a PostgreSQL numeric column or a pandas column of Decimal values — now serialize as float infinity instead of raising decimal.InvalidOperation. (#12254)
• Regex Expectations on Snowflake now match substrings, consistent with other backends, rather than requiring the pattern to match the entire column value. (#12221)
• UnexpectedRowsExpectation no longer misreads a query as containing a JOIN when the letters appear inside a string literal, a column name such as join_date,

Highlights

• Spark now evaluates each regex independently with match_on="all" — On Spark, ExpectColumnValuesToMatchRegexList with match_on="all" now checks every regex separately against each column value, so patterns anchored at different positions (such as ^A and [0-9]{3}$) both match a value that satisfies them. This matches the behavior already seen on Pandas and SQL. (#12198)

  gxe.ExpectColumnValuesToMatchRegexList(
      column="id",
      regex_list=["^A", "[0-9]{3}$"],
      match_on="all",
  )
  

• Each Validator reports results for its own Batch when a datasource is reused — Validators built on the same datasource no longer borrow one another's Batch. Running two validation definitions on threads, or creating two validators from one datasource on a single thread, now evaluates and reports each validator's own data, with the correct batch_id, batch_spec and batch_definition on the result. This fixes a long-standing latent bug made reachable by #12148 in the 1.23.0 release. (#12211)

  validator_a = context.get_validator(batch_request=request_a)
  validator_b = context.get_validator(batch_request=request_b)
  # validator_a still validates request_a's batch
  result = validator_a.expect_table_row_count_to_equal(value=3)
  

• A Spark schema saved in great_expectations.yml reloads correctly — A persisted spark_schema is now read back through StructType.fromJson, so reopening a File Data Context round-trips the schema instead of failing inside PySpark. Values that are not an accepted schema form now raise a validation error naming the field and the accepted types. (#12200)

  context = gx.get_context(mode="file")
  asset = context.data_sources.get("spark_ds").get_asset("my_asset")
  assert asset.spark_schema is not None
  

• GX config files are read and written as UTF-8 regardless of locale — config_variables.yml and great_expectations.yml, and the .gitignore read while scaffolding a project, are now opened with an explicit UTF-8 encoding. Projects containing non-ASCII values or comments can be created and reloaded on hosts with a non-UTF-8 locale, and a project YAML file that is not valid UTF-8 now raises an error naming the file instead of a bare decode error. (#12182, #12204)

• A gallery-wide test tier for data sources — A new gallery support tier asserts a measured test result across the entire shipped expectation library — one case per registered expectation, each pairing a passing and a failing configuration — and nine data sources (pandas in-memory and filesystem CSV, SQLite, MySQL, PostgreSQL, Trino, BigQuery, Databricks and Redshift) now declare it after being measured against the full gallery. (#12150)

Changes

Features

• Adds a gallery support tier that asserts a passing test result over the whole shipped expectation library, with nine data sources declaring it, a required CI lane per member, membership and coverage guards, and a measurement mode for evaluating new candidate backends. (#12150)

Bug fixes

• ExpectColumnValuesToMatchRegexList with match_on="all" on Spark now evaluates each regex independently, so patterns anchored at different positions no longer fail on values that satisfy them all. (#12198)
• A Validator now keeps the identity of the Batches it loaded even when its execution engine is shared, so concurrent validations and multiple validators on one datasource e

• [BUGFIX] Cache the SQL execution engine across calls instead of rebuilding it every time (#12148)
• [BUGFIX] query asset SQL ending in a comment fails validation (#12124) (thanks @nanjeshramesh)
• [BUGFIX] Wrap a query asset's SQL whole so Oracle does not add FROM DUAL (#12162)
• [BUGFIX] TupleFilesystemStoreBackend reads with ambient locale encoding (#12125) (thanks @nanjeshramesh)
• [BUGFIX] Make ExpectColumnValueZScoresToBeLessThan pass rather than fail or raise on zero or undefined variance (#12145) (thanks @Star-cloud626, Claude Opus 5 (1M context))
• [BUGFIX] expect_column_values_to_be_unique raises KeyError on mixed-case column names (#12180)
• [BUGFIX] Report an undefined SQLite standard deviation as None instead of raising (#12168) (thanks @siddharthgaur1, Claude Opus 5)
• [DOCS] Update repository links to the fivetran org (#12126) (thanks @iamfeldman)
• [MAINTENANCE] Declare metric values, render-content payloads, and metric keys as what they actually are (#12116)
• [MAINTENANCE] Generalize the data source declaration record beyond SQL backends (#12110)
• [MAINTENANCE] Pin the fluent datasource management API with a per-type CRUD contract suite (#12141)
• [MAINTENANCE] Say what the shared fixture column types mean (#12147)
• [MAINTENANCE] Type-check tests/test_utils.py, tests/actions/ and tests/checkpoint/ (#12146) (thanks @MannXo)
• [MAINTENANCE] Clean up check-actor-permissions and dead CI workflow code (#12159)
• [MAINTENANCE] Bump browserslist from 4.28.1 to 4.28.8 in /docs/docusaurus (#12155)
• [MAINTENANCE] Bump fast-uri from 3.1.5 to 3.1.7 in /docs/docusaurus (#12151)
• [MAINTENANCE] Type-check tests/render (#12152) (thanks @Ryota-Di)
• [MAINTENANCE] Type-check tests/core/ #12130 (#12140) (thanks @AnandkumarMall)
• [MAINTENANCE] Type-check tests/execution_engine/partition_and_sample/ (#12169) (thanks @siddharthgaur1, Claude Opus 5)
• [MAINTENANCE] Bump colord from 2.9.3 to 2.10.0 in /docs/docusaurus (#12170)
• [MAINTENANCE] Type-check tests/data_context/ (#12171) (thanks @nanjeshramesh)
• [MAINTENANCE] Remove unused checkpoint test fixtures (#12174)
• [MAINTENANCE] Bump svgo from 3.3.4 to 3.3.5 in /docs/docusaurus (#12175)
• [MAINTENANCE] Bump joi from 17.13.4 to 17.13.7 in /docs/docusaurus (#12176)
• [MAINTENANCE] Publish the oracle ext