Skip to main content
Version: 1.8.0

DataHub Releases

Summary​

VersionRelease DateLinks
v1.9.02026-11-16Release Notes, View on GitHub
v1.6.0.52026-11-12Release Notes, View on GitHub
v1.6.0.42026-10-29Release Notes, View on GitHub
v1.8.02026-10-12Release Notes, View on GitHub
v1.6.0.32026-09-25Release Notes, View on GitHub
v1.7.0.12026-09-03Release Notes, View on GitHub

v1.9.0​

Released on 2026-11-16 by @datahub-release-bot[bot].

v1.9.0​

Release Availability Date​

16 Nov 2026

Requirements​

  • CLI / Python SDK: 1.7.0.19
  • Helm chart: 1.1.10

Upgrade actions are in Breaking Changes for v1.9.0.

Highlights​

  • Metrics are on by default. METRICS_ENABLED now defaults to true, which turns on the Metrics page, the /metrics route, and the backend wiring for Metric and Semantic Model. v1.8.0 defaulted to false. Set METRICS_ENABLED=false to turn it off.
  • Google Drive source (Alpha). Ingests Google Docs from chosen folders and shared drives as Document entities, so they are searchable, including with semantic search.
  • datahub recipe probe. Checks what a recipe would ingest before a run. probe, test-connection, and validate now use exit code 2 for bad input, 3 for a source failure, and 1 for DataHub, the connector, or a missing package.
  • Elasticsearch 7.x is no longer supported. Use Elasticsearch 8.x, or OpenSearch 2.x or 3.x.
  • System client credentials go only to the configured server. A client created without a token sends DATAHUB_SYSTEM_CLIENT_ID and DATAHUB_SYSTEM_CLIENT_SECRET only to the server in DATAHUB_GMS_URL. UI-managed ingestion refuses recipes that would send them elsewhere.

Product​

Platform​

Ingestion​

  • The UI-managed ingestion executor validates extra_pip_requirements before it installs them. ([#21203](https://github.com/datahub-project/datahub/pull/21203))
  • The ingestion framework runs on SQLAlchemy 2.0. acryl-datahub requires click>=8.3.3 and anyio>=4.14.2.
  • DataHub Lite uses SQLite for new instances. DuckDB is a separate extra.
  • The hive and presto sources no longer use PyHive.

Security​

Deprecations​

  • Audit stamps that only repeat the DataHub write are deprecated in favor of systemMetadata.aspectCreated and systemMetadata.aspectModified. The old fields remain and are still indexed.
  • GET /admin and GET /health on the Play port (9002) are deprecated as health checks. They still return 200 GOOD, or 503 during graceful shutdown.

Breaking Changes​

The full list, with actions, is in Breaking Changes for v1.9.0. The main changes are:

  • Authorization. Changing an entity's platform instance or domain is checked against both the old and the new value. Assertion APIs apply the same asset-level permissions as the UI. Scheduling an ingestion source requires permission. Invite tokens that assign a role require Manage User Credentials and Manage Policies. Privilege-constraint denials on tags and structured properties return 403 instead of 400.
  • Search. Elasticsearch 7.x is removed. Search V3 keyword scoring, default query fields, indexed fields, and the lineage and browse templates changed. Timeseries indices refresh every 60 seconds by default.
  • Hard rollback. The UI Rollback button and datahub ingest rollback --nuke no longer delete an entity once another run or a user has written to it.
  • Conditional writes. If-Version-Match and If-Modified-Since apply to generic JSON patches.
  • Ingestion. Airflow plugin datasets use the same URNs as the S3, GCS, and Azure Blob sources, so lineage recorded under old URNs stays on the old datasets. dbt manifest_path treats *, ?, [, and ] as glob characters, and semantic models are no longer modeled as warehouse tables. Sigma runs fail when an entity listing fails. The profiling config classes lost their GE prefix.
  • CLI. datahub recipe commands return the exit codes above. The cloud pip extra was removed from acryl-datahub.
  • Logging and startup. GMS no longer writes DEBUG logs to gms.debug.log by default. GMS, MAE, and MCE no longer wait on the DataHubUpgradeHistory_v1 marker.

Full Changelog: https://github.com/datahub-project/datahub/compare/v1.8.0...v1.9.0

v1.6.0.5​

Released on 2026-11-12 by @supersingh05.

DataHub v1.6.0.5​

Patch release for the v1.6.0 LTS line. Focus is security: November CVE dependency bumps, Jackson 2.22.4, and hardening of the datahub-documents source, which now also works without outbound network access.

Full diff: https://github.com/datahub-project/datahub/compare/v1.6.0.4...v1.6.0.5


Operator notes​

  • datahub-documents no longer parses Markdown links or images, and partitioning has a CPU-time limit. Link and image syntax is kept as literal text, which removes the quadratic parsing path behind CVE-2019-25781. Partitioning one document stops after partition_timeout_seconds (default 30 s) (#21204, from #21125).
    • A document that hits the limit is skipped with a warning and counted in num_documents_partition_timeout. Its hash is still recorded, so it is not retried until its content changes. Embeddings written by an earlier version are left in place.
  • The spaCy model en_core_web_sm is now bundled in the images. unstructured loads this model the first time it partitions text. Earlier v1.6.0 images did not include it, so datahub-documents (and other unstructured-based sources) downloaded it from github.com on first use. Without outbound access, each document failed after a 120 s timeout with Failed to download spaCy model, and the run still reported success with 0 chunks. The model is now installed in the bundled ingestion venvs of the datahub-actions images and in the full datahub-ingestion image (#21210, from #19554).
    • Documents that failed this way on v1.6.0.4 or earlier are not re-processed automatically. Their hashes were recorded, so they are skipped as unchanged until their content changes. Edit those documents, or reset the source's state, to have them chunked.
    • Loading the model raises the memory used by datahub-documents runs. The Helm chart's default acryl-datahub-actions memory limit (512Mi) leaves little headroom, so consider raising it if you use Documents, especially if runs can overlap.
  • Fresh installs of acryl-datahub work again. sqlglot[c] is pinned to 30.11.0. PyPI no longer serves sqlglotc wheels for 30.7.0–30.10.0, so with the previous 30.8.0 pin every fresh install of this line tried to build sqlglotc from source and failed (#21204).

Security​

AreaFix
Jackson2.22.4: jackson-dataformat-smile CVE-2026-104895 (GHSA-h3c2-2m49-r49w, unbounded allocation in the async parser) and jackson-dataformat-cbor CVE-2026-106554 (GHSA-cjg7-h3pr-p927) (#21208)
Java runtime librariesorg.json 20260814 (CVE-2026-59171), netty-tcnative 2.0.84.Final (CVE-2026-116541), json-smart / accessors-smart 2.6.1 (CVE-2026-113602) (#21204)
unstructured / filelockunstructured 0.27.37 lifts the filelock<4 cap; filelock 4.2.10 (CVE-2026-112940); snowflake-connector-python resolves to 4.8.1 (#21204)
Other Pythonpyasn1 ≥0.6.5 (CVE-2026-119194 / 119195 / 119196 / 119197), markdown ≥3.11.3 (#21204)

markdown 3.11.3 does not fix CVE-2019-25781 by itself. Advisories list 3.0–3.11.2 as affected, so the floor clears scanners, but 3.11.3 still parses unclosed brackets in quadratic time. The datahub-documents change in the operator notes is the actual fix.

Not included:

  • Jackson 2.21.2 / 3.1.2 and org.json 20250107 shaded inside hazelcast-5.7.0.jar. 5.7.0 is the latest Hazelcast OSS release.
  • Guava CVE-2026-102554 in the copies shaded into graphql-java and Iceberg. The standalone guava is 33.7.2. Since v1.6.0.4, the embedded Hazelcast rejects Guava classes during deserialization.
  • json-smart 1.3.2 and older Hadoop / Hive jars in the datahub-actions full image. They come from the bundled PySpark venv; the slim and locked images do not include it.

Full Changelog: https://github.com/datahub-project/datahub/compare/v1.6.0.4...v1.6.0.5

v1.6.0.4​

Released on 2026-10-29 by @supersingh05.

DataHub v1.6.0.4​

Patch release for the v1.6.0 LTS line. Focus is security: October CVE dependency bumps, an allow-list for Java deserialization in GMS's embedded Hazelcast, and tighter handling of system client credentials in the Python client.

Full diff: https://github.com/datahub-project/datahub/compare/v1.6.0.3...v1.6.0.4


Operator notes​

  • Embedded Hazelcast only deserializes allow-listed classes from its peers. Allowed: JDK types, com.hazelcast.*, com.linkedin.metadata.* and org.javatuples.* (search cache keys). Guava classes, including the copies bundled in graphql-java and Iceberg, are always rejected. Before, any process that could reach the cluster port (5701 by default) could trigger readObject on any Serializable class on the GMS classpath (#20783, from #20776).
    • No action needed for a standard deployment. During a rolling upgrade from an older version, an upgraded GMS may not be able to read search cache keys that older pods wrote. Eviction skips those keys with a warning, and they expire with the cache TTL (CACHE_TTL_SECONDS, 10 minutes by default).
    • If an extension stores its own Serializable types in Hazelcast, allow them with SEARCH_SERVICE_HAZELCAST_SERIALIZATION_ALLOWED_CLASSES or SEARCH_SERVICE_HAZELCAST_SERIALIZATION_ALLOWED_PACKAGE_PREFIXES (comma-separated), or register a HazelcastSerializationAllowListContributor bean. Rejected types fail with HazelcastSerializationException ("Resolving class ... is not allowed").
  • Breaking (Python client): system client credentials are only sent to the configured GMS. A client created without a token (REST emitter, DataHubGraph, datahub-rest sink) now attaches DATAHUB_SYSTEM_CLIENT_ID / DATAHUB_SYSTEM_CLIENT_SECRET only when its server matches DATAHUB_GMS_URL (or DATAHUB_GMS_HOST / DATAHUB_GMS_PORT / DATAHUB_GMS_PROTOCOL). Otherwise it connects without credentials and logs a warning. Hosts are compared as written, so http://localhost:8080 and http://127.0.0.1:8080 do not match (#20783, from #20717).
  • Breaking (Python SDK): the cloud pip extra is removed. pip install 'acryl-datahub[cloud]' no longer installs the cloud package; pip only warns about the missing extra. Install acryl-datahub-cloud directly; it pins the matching acryl-datahub (#20676, from #20325).
  • Default managed-ingestion CLI version is now 1.7.0.17, so UI-triggered ingestion installs a CLI that includes the dependency fixes below (#20676).

Security​

AreaFix
Frontend SSOpac4j 6.5.8, play-pac4j 13.0.3 (CVE-2026-82461 / 82462 / 82463 / 82464 / 82465); brings nimbus-jose-jwt 10.9.1, oauth2-oidc-sdk 11.38.2, json-smart 2.6.0
Java runtime librariesnetty 4.2.19, vert.x 4.5.35, reactor-netty 1.3.8, guava 33.7.2, snappy-java 1.1.10.12, protobuf 4.36.2, hadoop 3.4.2 (brings avro 1.11.4), google-api-client 2.9.2, lz4-java 1.11.4 (CVE-2026-106451)
starlette / fastapistarlette ≥1.3.1 (CVE-2026-48710 / 48817 / 48818 / 54282 / 54283), fastapi ≥0.133
aiohttp and friendsaiohttp ≥3.14.4 (CVE-2026-111064), multidict ≥6.9.1 (CVE-2026-104874), mako ≥1.4.2 (CVE-2026-102991), fsspec ≥2026.6.0 (CVE-2026-104851)
Auth librariesauthlib ≥1.9.0 (CVE-2026-96760), joserfc ≥1.7.4 (CVE-2026-102946), PyJWT ≥2.15.0
pip / certifi / setuptoolspip ≥26.2.2 and certifi ≥2026.10.19 (GHSA-6q48-6c7w-9gw2), setuptools ≥84.1.0 (GHSA-xmj3-9gf7-h52w)
Other Pythonpypdf ≥6.19.0 (CVE-2026-103000), snowflake-sqlalchemy ≥1.11.0 (CVE-2026-15736), opentelemetry-api ≥1.45.1 (CVE-2026-114603), urllib3 ≥2.8.0
Node.js22.23.4 (October 2026 security release)
ImagesCA trust store refresh and Python floors in the Docker ingestion constraints

Not included: oauthlib 4.0.0 (CVE-2026-49264 / 49265). This line uses databricks-sql-connector 2.9.6, which caps oauthlib below 4; oauthlib stays at 3.3.1.


Fixes and CI​

  • The backported semanticText document tests now match the v1.6.0 source, so ci (3.11/3.12, testQuick) passes on this branch again (#20676).
  • pytest capped below 10 in the Airflow and Prefect plugin setups; Grafana and Prometheus images pinned in the Grafana integration test (#20783).

Full Changelog: https://github.com/datahub-project/datahub/compare/v1.6.0.3...v1.6.0.4

v1.8.0​

Released on 2026-10-12 by @datahub-release-bot[bot].

v1.8.0​

Release Availability Date​

22 Oct 2026

Requirements​

  • CLI / Python SDK: 1.7.0.14
  • Helm chart: 1.1.8

Upgrade actions are in Updating DataHub.

Highlights​

  • OpenSearch 3.x. Supported next to OpenSearch 2.x and Elasticsearch 8.x. The engine version is detected automatically. OpenSearch 3.x will not create new indexes with the nmslib kNN engine, and it enforces the 512-byte document-id limit on bulk writes.
  • Metrics and Semantic Models. Charts, dashboards, and datasets can take a Metric as an upstream, with column-level lineage through metric upstreams. Metrics and Semantic Models are included in global search. The feature ships off. Set METRICS_ENABLED=true on GMS to show it.
  • Dark mode. A dark theme is in this release and stays off until THEME_DARK_MODE_ENABLED=true. With that set, the toggle is under Settings → Appearance.
  • Structured Properties. Creating, editing, and viewing a structured property is a page rather than a side drawer. Allowed values can be reordered, each with an optional description. Leaving the page with unsaved edits prompts before discarding them. The properties table adds Type and Platforms columns.
  • Data Products. Data Products can nest under a parent Data Product. Data Products are filterable in global search. On upgrade, a one-time system update fills the asset-side dataProducts search field from stored membership so that filter matches existing members. It is on by default (BOOTSTRAP_SYSTEM_UPDATE_DATA_PRODUCT_ASSETS_ENABLED=true).
  • New ingestion sources. IBM/HCL Informix, Monte Carlo, and SQLMesh.

Product​

  • Compliance forms can be assigned to owners of a specific ownership type.
  • Deprecation date input is editable.
  • Languages in public beta: Japanese, Simplified Chinese, Traditional Chinese, and Russian.
  • Sidebars. Domains, glossary, metrics, and documents sidebars can sort and filter. The home page Domains module lists up to 25 domains.
  • Views can filter on structured properties.
  • Incidents can be raised on ML models, ML features, ML feature tables, and schema fields.

Platform​

  • Search document ids (Search V3). Search V3 documents use a SHA-256 of the entity URN as _id. V2 indexes are unchanged. If an environment already wrote V3 documents under URL-encoded ids, hashed ids are not an in-place rewrite. Run RestoreIndices with -a clean, or reindex into a fresh index. A fill without clean leaves the old _id in place and the next update creates a duplicate.
  • GraphQL worker pool. When GRAPHQL_CONCURRENCY_SEPARATE_THREAD_POOL=true, pool sizes no longer follow the node’s CPU count. Defaults are the 8-core sizes: core 40, max 800. The separate pool stays off by default.
  • AWS IRSA. GMS keeps one process-wide default credentials provider, built once and closed on shutdown.
  • Timeseries reads. Dataset profile, usage, and operations, and dashboard usage statistics, require the matching View privileges on Rest.li and OpenAPI as well as GraphQL.
  • Access tokens. Creating a token with no expiry is disabled. Allowed durations default to 1 hour, 1 day, 7 days, 30 days, 90 days, 180 days, and 365 days. Tokens that already exist, including never-expiring ones, keep working.
  • Query SQL. Reading a query, a view definition, or topSqlQueries requires View Entity Queries on a subject dataset, or the platform privilege View All Queries. On upgrade, policies that already grant View Entity Page are backfilled with both. A policy created after the upgrade that only grants View Entity Page does not reveal query SQL.

Ingestion​

New sources: IBM/HCL Informix, Monte Carlo, SQLMesh.

Connectors:

  • Snowflake. Logical tables in semantic views keep their stored casing. preserve_column_case keeps Snowflake’s column casing in field paths and column-level lineage. It is off by default. Turning it on changes every column’s schemaField URN.
  • BigQuery. Sharing linked datasets, multi-column partition info, and dataset profiles for materialized views. The profiler lives in its own package.
  • Cube. Cube views are ingested as Semantic Models. emit_semantic_model_entities: true emits first-class semantic model, metric, and semantic-model dataset entities. That path needs METRICS_ENABLED=true.
  • Dataplex. Metadata Export API extraction.
  • Metabase. Column-level lineage, model ingestion, and collection tags and containers.
  • MicroStrategy. Project schema can be ingested as a Semantic Model.
  • Power BI. Lineage for BigQuery EXTERNAL_QUERY.
  • SAP Analytics Cloud. Column-level lineage to SAP Datasphere for DWC live models, and SAC to SAP Datasphere live-model lineage.
  • SAP Datasphere. Calculated column formulas are written into the column description.
  • Sigma. Dataset lineage is recovered from Sigma’s connection metadata. See Breaking Changes.
  • Kafka Connect. SQL Server is a three-level platform. See Breaking Changes.
  • Looker. Per-field explore usage from System Activity.
  • Airbyte. Inactive connections can be ingested by opting in. Doris can ingest Iceberg tables in non-default catalogs.

Ingestion infrastructure:

  • Connector support tiers are Alpha, Beta, and GA. The old names Certified, Incubating, and Testing are removed.
  • SQL lineage timeout is configurable.
  • Structured property settings can be ingested from the Python SDK.
  • Stdin envelope secrets are registered for masking when they arrive.
  • confluent-kafka is 2.15.1. A custom oauth_cb must return an absolute expiry time, not a lifetime in seconds.

Breaking Changes​

Full actions are in Updating DataHub.

  • Sigma ingestion. Sigma Dataset lineage URNs are built from connection_to_platform_map rather than chart_sources_platform_mapping. Add the Sigma connectionId to that map with the env and platform_instance your warehouse connector used. Set convert_urns_to_lowercase: true if you want the previous spelling for those table names.
  • Structured property hard-delete. An active structured property cannot be hard-deleted, and its propertyDefinition aspect cannot be deleted directly, until the property is soft-deleted. The UI does both steps. Automation that issued one hard delete must soft-delete, then hard-delete.
  • Timeseries read authorization. API clients that read dataset profile, usage, or operations with only entity GET need View Dataset Profile, View Dataset Usage, or View Dataset Operations. Dedicated timeseries APIs also require Get Timeseries Aspect API when REST API authorization is enabled.
  • Role and group membership. Adding a role (roleMembership) requires Manage Policies. Adding a group member requires Edit Group Members. Adding a group owner requires Edit Owners. When the acting user is the one being added, both require Manage Users & Groups. Edit Entity on a user is no longer enough to grant that user Admin.
  • Analytics API. POST /openapi/v2/analytics/datahub_usage_events/_search requires Analytics API access or Manage System Operations. The in-app analytics dashboard still uses View Analytics.
  • Compliance form assignment. batchAssignForm, batchRemoveForm, createDynamicFormAssignment, and writes that change which forms are on an asset, require Manage Compliance Forms. Completing an assigned form is unchanged.
  • Asset summary settings. updateAssetSettings and writes to assetSettings require Edit Entity or Manage Asset Summary on the asset. OpenAPI and Rest.li still treat the write as a generic entity update, so Manage Asset Summary alone is not enough there.
  • Never-expiring access tokens. Disabled unless ACCESS_TOKEN_ALLOW_NO_EXPIRY=true. Set ACCESS_TOKEN_ALLOWED_DURATIONS to change the allowed TTLs.
  • OpenAPI create is per URN. A principal with only Create Entity cannot overwrite an existing entity through create or UPSERT. Grant Edit Entity to callers that update existing entities.
  • Domain-scoped create. Domain-scoped Create and Edit policies are evaluated against the domains proposed on a new entity. A domains PATCH requires Edit Entity, and domain-scoped Edit must allow both the previous and the new domain membership.
  • Group member removal. removeGroupMembers removes members who only have the legacy groupMembership aspect.
  • Documents. The default Editor role no longer has Manage Documents. Editors can create, edit, and move documents, and can delete documents they own.
  • Password hashes. Reading corpUserCredentials requires Manage User Credentials.
  • Kafka Connect. SQL Server sink URNs gain a dbo schema when schema.name is unset (database.table becomes database.dbo.table). A SQL Server source dataset with no resolvable schema is dropped with a warning. Oracle JDBC sinks emit schema.table.
  • Elasticsearch ingestion. acryl-datahub[elasticsearch] requires opensearch-py 3.x. The 3.x client still talks to OpenSearch 2 clusters.
  • Great Expectations SQL profiler. Removed. The SQLAlchemy profiler is the only SQL profiler. profiling.method and the acryl-datahub[profiling-ge] extra are gone. This does not affect the Great Expectations integration that ingests GX validation results.
  • MySQL, MariaDB, Doris, and TiDB profiling limits. profiling.profile_table_row_limit and profiling.profile_table_size_limit are enforced. They were previously accepted and ignored. Both still default to unset on these sources.
  • sql-queries. A run where every input line is unparseable, or every query fails, exits non-zero. enable_lazy_schema_loading is removed. Report fields num_queries_processed_sequential and num_temp_tables_detected are renamed.
  • Python SDK circuit breakers. Assertion and Operation clients take graph= instead of transport=. GraphQL errors raise GraphError.
  • Connector support status names. SupportStatus.CERTIFIED, INCUBATING, and TESTING are removed. Use GA, BETA, and ALPHA.
  • Python extras. notion, confluence, unstructured, and datahub-documents require Python 3.11+ and are no longer installed by acryl-datahub[all].
  • Kafka OAuth callbacks. Since confluent-kafka 2.13, the second value returned by oauth_cb is an absolute expiry in epoch seconds. A callback that returns a lifetime such as 3600 now looks already expired.
  • Search indexing startup. GMS refuses to start when it consumes MCLs and both UI indexing paths are off (PRE_PROCESS_HOOKS_UI_ENABLED=false and PRE_PROCESS_HOOKS_REPROCESS_ENABLED unset or false).
  • pgQueue idle sleep. DATAHUB_PGQUEUE_EMPTY_POLL_SLEEP_MS and DATAHUB_PGQUEUE_MCL_EMPTY_POLL_SLEEP_MS are removed. Use the new min and per-topic max settings. pgQueue maintenance intervals must divide the clock (for example 15 minutes, not 7).
  • OpenAPI token recipes. get_token is a typed object and cannot be combined with token or bearer_token.
  • Actions Kafka. A pipeline whose Kafka source cannot connect stops, and the actions process exits, instead of staying up while consuming nothing.
  • Log line format. datahub-frontend, datahub-upgrade, and GMS log lines no longer include the source line number.

Deprecations​

  • OpenAPI v2 entity and relationship APIs (/openapi/v2/entity, /openapi/v2/relationship) are deprecated in favor of /openapi/v3/entity and /openapi/v3/relationship. Timeseries, platform entity ingest, and timeline under v2 are not deprecated by this.
  • semanticModelInfo.datasets. Membership is stored on the member. Datasets use semanticModelProperties.semanticModel. Metrics use metricInfo.semanticModel. Re-ingest Semantic Models after upgrade so metric-to-dataset lineage is populated. It is missing between the upgrade and that ingest.

Security​

  • Spring Framework 7.0.9, Spring Boot 4.0.8, Spring Security 7.0.7.
  • Apache Parquet 1.18.1 (CVE-2026-73334).
  • Apache Log4j 2.25.5 (CVE-2026-49844).
  • Logback 1.5.38 (CVE-2026-9828, CVE-2026-10532).
  • Apache Thrift libthrift 0.24.0 (CVE-2026-48586).
  • pac4j 6.5.8 and play-pac4j 13.0.3 (CVE-2026-82461 through CVE-2026-82465).
  • acryl-datahub no longer caps setuptools, so it installs alongside setuptools>=83 (CVE-2026-59890).

Known Issues​

None recorded in the v1.8.0 upgrade notes.

v1.6.0.3​

Released on 2026-09-25 by @david-leifker.

DataHub v1.6.0.3​

Patch release for the v1.6.0 LTS line. Focus is security and authorization hardening, plus a few ingestion and platform fixes.

Full diff: https://github.com/datahub-project/datahub/compare/v1.6.0.2...v1.6.0.3


Operator notes​

  • Python 3.11 is required on this hotfix line (ingestion CLI/images and related packages). Python 3.10 is no longer supported for v1.6.0.x (#19703).
  • SecretService caller guard defaults to ENFORCE. Secret encrypt/decrypt now requires an OperationContext; misconfigured callers fail closed instead of silently succeeding (#17995).
  • Opt-in frontend security headers. Set DATAHUB_SECURITY_HEADERS_FRAME_OPTIONS, DATAHUB_SECURITY_HEADERS_CONTENT_TYPE_OPTIONS, and/or DATAHUB_SECURITY_HEADERS_REFERRER_POLICY to emit X-Frame-Options, X-Content-Type-Options, and Referrer-Policy. Unset vars omit those headers (#19848).

Security​

Dependency floors and library bumps, including:

AreaFix
Snowflake connectorFloor raised to 4.7.3 (CVE-2026-15925, plus 4.7.1/4.7.2 TLS SAN regression for account locators with _)
aiohttp3.14.3 (CVE-2026-69244 / 59881 / 69243)
pip26.2+ (CVE-2026-13346)
click≥8.3.3 (CVE-2026-7246)
deepdiff≥8.6.2 (CVE-2026-33155)
unstructured0.24.1 (CVE-2026-71428)
libthrift0.24.0 (CVE-2026-48586)
MariaDB Connector/J2.7.14+ (CVE-2026-55856/557/558)
Bouncy Castle1.86
Wolfi actions imageWOLFI_PKG_REFRESH bump (CVE-2026-85091 / zlib)
Java stackSpring Framework 7.0.9, Boot 4.0.8, Security 7.0.7; Jackson 2.22.x; Parquet 1.18.1; JJWT 0.13.0; httpclient5, OTel, GitPython, cryptography / pyOpenSSL, anyio, soupsieve

Also: setuptools ≥83 is allowed so consumers can take CVE-2026-59890 (#19601).


Authorization and secrets​

  • patchEntity / patchEntities authorize against the URN’s entity type, not the client-supplied entityType argument (#19872).
  • Manage User Credentials is required to write corpUserCredentials (GraphQL, OpenAPI, Rest.li) (#19967).
  • Asset settings, compliance form assignment, and raw analytics search require the intended privileges (#19750).
  • getSecretValues no longer fails the whole batch when one secret cannot be decrypted (#19539).

Features​

  • semanticText aspect — optional curated text used as the embedding source for semantic search, instead of the entity’s full body (#18654).

Ingestion​

  • confluent-kafka raised to 2.15.1.
  • urllib3 2.x (needed by unstructured 0.24.1); vcrpy ≥8.2.0 for aiohttp 3.14 compatibility.

Full Changelog: https://github.com/datahub-project/datahub/compare/v1.6.0.2...v1.6.0.3

v1.7.0.1 🖖​

Released on 2026-09-03 by @david-leifker.

DataHub v1.7.0.1 🖖​

Patch release for the v1.7.0 line: authorization tightening, CVE dependency bumps, entity-graph-cache and lineage correctness fixes, optional optimistic locking for aspect writes, and the remainder of the custom-assertion stack.

Requirements

  • CLI / Python SDK: 1.7.0.9
  • Helm Chart: 1.1.0

Full upgrade guidance, including every breaking change and migration step: Updating DataHub — v1.7.0.1.

Upgrade path

  • From v1.7.0: drop-in image bump. No ZDU or Helm chart change is required. System-update runs a resumable backfill that copies embedded assertion notes to the new assertionNote aspect (tunable with SYSTEM_UPDATE_ASSERTION_NOTE_MIGRATION_ENABLED, _BATCH_SIZE, _DELAY_MS, _LIMIT).
  • From pre-1.7.0: the v1.7.0 path still applies — you must upgrade to v1.6.0 first (Helm chart 1.0.3, let system-update complete), then upgrade to v1.7.0.1 with Helm chart 1.1.0. Do not skip 1.6.0.

Model changes: additive only. The assertionNote aspect is now registered on the assertion entity so ingestion can fully upsert assertionInfo without clobbering user-authored notes. Assertion URNs and run history are unchanged.


Action required before upgrading​

AreaWhat changedWho is affected
System client secretThe hardcoded systemClientSecret default (JohnSnowKnowsNothing) is removed from the server configs (#18886)Set DATAHUB_SYSTEM_CLIENT_SECRET on GMS, MAE/MCE/PE consumers, the frontend, and Actions before upgrading if you still rely on the built-in default
Timeseries readsDataset profile / usage / operations timeseries (and dashboard usage stats) now require the matching View Dataset Profile / Usage / Operations privileges on Rest.li and OpenAPI as well as GraphQL (#19508)API clients that previously read these aspects with only entity GET — grant the privileges or rely on the default view-dataset-sensitive policy
Role & group membership writesroleMembership requires Manage Policies; groupMembership / nativeGroupMembership requires Edit Group Members; corpGroup ownership requires Edit Owners — checked at the aspect layer across GraphQL, OpenAPI, and Rest.li (#19360)Automation writing membership outside the UI. Only additions are checked; removals, unchanged re-ingestion, and system writes pass. Toggle with metadataChangeProposal.validation.aspectAuthorization.privilegeGrant.enabled (default true)
View authorization (VBAC)container and schemaField are view-restricted by default; schema-field View Entity Page inherits from the parent dataset (#19212)Deployments with VIEW_AUTHORIZATION_ENABLED=true — ensure View Entity Page policies cover containers and parent datasets

Security and authorization​

  • Default self policy (#19297) — limited to VIEW_ENTITY_PAGE and GET_ENTITY on the actor's own entity.
  • GMS /auth helpers (#19316) — signUp, resetNativeUserCredentials, verifyNativeUserCredentials, and getSsoSettings now require system client credentials, matching generateSessionTokenForUser. Frontend login, signup, password reset, and SSO are unchanged.
  • System-write identification (#19384) — aspect authorization no longer treats client-supplied appSource as proof of a system write.
  • Schema-field incidents (#19405) — authorized against the parent entity.
  • Lineage scroll (#19463) — unauthorized OpenAPI lineage scroll endpoints are redacted as restricted.
  • UI sanitization (#19298, #19299, #19300) — query and incident descriptions, documentation PDF preview iframe src, and embed renderUrl schemes.
  • Ingestion secret masking (#19515, #19526, #19561, #19332) — masking is unified, idempotent, and fail-closed (mask before truncation; install once per process). Recorded HTTP cassettes are scrubbed.
  • AWS credentials (#19483) — OpenSearch IAM signing, Bedrock embeddings, STS, and object storage share a process-wide credential provider and client lifecycle.

Reliability and performance​

  • Entity-graph cache (#19498, #19288, #19301, #19412) — relationships are no longer silently dropped; indexed search fields are fetched for the cache; null relationship endpoints are skipped; hierarchy graph fallback batches frontier scrolls.
  • Container contents (#19489) — Container.relationships(types: [IsPartOf], direction: INCOMING) again returns all contained entities from the live graph.
  • Lineage traversal (#19413) — scrollAcrossLineage is bounded to prevent GMS OOM on wide graphs.
  • Case-insensitive relatedEntityTypes (#19268).
  • Hot-path caching and batching (#19404, #19476, #18266) — corpGroup roleMembership cached via SystemEntityClient, per-request CorpUserFlags served from a cached aspect, and Dashboard usageStats buckets and metrics batch-loaded across a page.
  • Optional optimistic locking for aspect writes (#19249) — OPTIMISTIC_LOCKING_ENABLED replaces SELECT FOR UPDATE with compare-and-set on SystemMetadata.version for Ebean storage. SCOPED_RETRY_ENABLED retries only the conflicted URN's branch, and ENTITY_WRITE_LOCK_BACKEND=hazelcast serializes same-(urn, aspect) writers off the DB connection pool. PostgreSQL aspect writes also take row locks in consistent primary-key order. All of these default off.

Assertions​

  • Custom-assertion stack completed on the 1.7 line (#18882) — API/SDK, UI, the assertionNote aspect, and the dbt / Great Expectations / YAML data-contract writers. Those writers emit AssertionType.CUSTOM instead of the deprecated DATASET shape on the next run; assertion URNs are unchanged and run history is preserved. Assertions that are never re-ingested keep their stored DATASET aspects and continue to render. See Custom Assertions.

Ingestion​

  • nltk 3.10.1 excluded (#18889) — that version silently breaks document chunking.
  • GitPython floor raised to 3.1.58 (#19506).
  • Bundled default CLI / Python SDK is 1.7.0.9 (#19605).

Dependency CVE bumps​

  • Java: httpclient5 5.6.3 (CVE-2026-64607), Spring Boot / Spring Kafka 4.0.7 (CVE-2026-41001), OpenTelemetry Java agent 2.28.0 (CVE-2026-54704), mariadb-java-client 2.7.14 (CVE-2026-55856 / 55857 / 55858), Logback 1.5.38 (CVE-2026-9828, CVE-2026-10532), Log4j 2.25.5 (CVE-2026-49844), Netty 4.2.17.Final (CVE-2026-59902), libthrift 0.23.0 (CVE-2026-43869), Jetty 12.1.10 (CVE-2026-10050), httpcore5 5.4.3 (CVE-2026-54399), micrometer-core 1.16.6 (CVE-2026-40983, CVE-2026-40984), wire-runtime 6.3.0 (CVE-2026-45799), reactor-netty-core 1.3.6, netty-reactive-streams 3.0.9.
  • Apache Parquet 1.18.0, which refreshes its shaded Jackson to jackson-databind 2.22.1 (CVE-2026-54512, CVE-2026-54513).
  • Python / images: nltk 3.10.3 (CVE-2026-12075); the datahub-actions image python layer is cache-busted so a stale cached layer stops shipping setuptools 70.3.0 (CVE-2025-47273).

Quickstart​

datahub docker quickstart (and --version v1.7.0) resolves to the v1.7.0.1 images once the version mapping update lands.


What's Changed​

Full Changelog: https://github.com/datahub-project/datahub/compare/v1.7.0...v1.7.0.1