Profiles 0.26 Changelog

Version 0.26.2

7 September 2026
Schema version: 99

Release Schedule

What’s New

• Material cleanup is rearchitected for scale — registry deletions are batched, warehouse objects are dropped in dependency order without cascading deletes, per-column cleanup is skipped when the whole table is dropped, and deletions run concurrently. Cleanup of projects with hundreds of entity vars now completes in seconds instead of hours.
• Partial cleanup (--expired) always preserves the latest version of every model along with its frontier views, including the bases that incremental models build on.
• Materials without a declared retention period now inherit the project-level (or default 30-day) retention instead of being treated as immediately expired. Input-var materials continue to be cleaned up after every run.
• The --concurrency flag of pb cleanup materials defaults to 4 on all warehouses. See pb cleanup for more details.

Bug Fixes

• Concurrent or repeated cleanup runs no longer fail on objects that are already gone (for example, cache lookup failed for relation or could not open relation errors on Redshift). Each object is dropped exactly once, and objects removed by an earlier pass are treated as successfully cleaned.

Version 0.26.1

29 July 2026
Schema version: 99

Release Schedule

Bug Fixes

• Output schemas are normalized consistently in the warehouse client constructors, fixing schema resolution issues in projects that configure a separate output schema.
• ID Stitcher temp-table pruning is idempotent across retries, so retrying a run at the same sequence number no longer fails with already exists errors.

Version 0.26.0

23 July 2026
Schema version: 99

Release Schedule

What’s New

• Rule-based filtering (ID Graph Cardinality Rules) is now supported in ID Stitcher for BigQuery, in addition to Snowflake
• ID Stitcher metadata logging (log_direct_edge_info and log_node_metadata) is now supported for BigQuery, in addition to Snowflake. See ID Stitcher
• Per-source edge attribution: Each bucket in the edges_by_type metadata column now carries a sources[] array listing the source models that contributed the edge. See Identity Graph
• The <material>_<entity>_CARDINALITY_AUDIT table includes a new source_models column, so you can trace a capped or dropped edge back to the source models that produced it
• Incremental entity var bundling: Support to compute incremental entity vars in a bundle is now available.
• Profiles writes user-facing output tables to a schema separate from its internal working tables.
• The pb CLI supports shell autocompletion

Miscellaneous

• BigQuery queries now enforce GoogleSQL, with normalized JSON and time expressions
• S3 configuration resolution and the Redshift S3 write path support IAM role authentication, in addition to key-based authentication
• The pb version is included in the task summary
• Snowflake OCSP cache miss errors can be suppressed using a PB environment toggle

Bug Fixes

• Rebase-affected default materials are now reused at compile time
• rebase_incremental is no longer silently skipped in already-materialized timegrain buckets
• Material cleanup now tolerates dependency cycles instead of failing
• Connection leaks and broken pipe errors caused by unbounded connection pools are fixed
• Query tags work again on Snowflake

Known Issues

BigQuery

• pb validate access command does not work for BigQuery.

Redshift

• If two different users create material objects on the same schema, RudderStack gives an error during cleanup when trying to drop views created by the other user, like user_var_table.
• Cross-database references can fail on Redshift for a few clusters.
• While creating activations, validation for Redshift does not work correctly in the RudderStack dashboard.

Other issues

• Linux users might see this warning for all command runs — you can ignore it:

WARN[0000]log.go:228 gosnowflake.(*defaultLogger).Warn DBUS_SESSION_BUS_ADDRESS envvar looks to be not set, this can lead to runaway dbus-daemon processes. 

To avoid this, set envvar DBUS_SESSION_BUS_ADDRESS=$XDG_RUNTIME_DIR/bus (if it exists) or DBUS_SESSION_BUS_ADDRESS=/dev/null

• pb insert does not work for Redshift, Databricks, and BigQuery.
• If you are referring a public package in the project and get the ssh: handshake failed error, then you will have to manually remove the entire folder from WhtGitCache to make it work.