dagster-io/dagster 1.13.26

1.13.26 (core) / 0.29.26 (libraries)

on GitHub

Breaking Changes

• [dagster-dbt] dagster-dbt no longer installs dbt-core. It is now available as the dagster-dbt[dbt-core] extra. Do not install dbt-core into an environment that also has dbt Fusion's dbt PyPI package: both claim the dbt import namespace and overwrite each other's files, and pip reports no error.

New

• Sensor ticks that request a run larger than the instance's run size limit now fail with an explicit error naming the offending run requests, instead of storing those requests in tick history.
• The SF_PARTNER_ID environment variable is now set by default, to allow for partner attribution on Snowflake connections made through third-party libraries.
• Run monitoring now includes the run worker's debug info (for example, a Kubernetes OOMKilled termination reason) when it fails a started run because the run worker failed. (Thanks, @hosaka2!)
• [dg] dg list envs, dg plus create env, and dg plus pull env now accept a --deployment option (or the DAGSTER_CLOUD_DEPLOYMENT environment variable) to target a deployment other than the configured default.
• [dagster-cloud] The published dagster-cloud-agent image now includes the Azure Container Apps launcher.
• [dagster-dbt] A dbt Fusion binary on PATH is now used as-is, instead of being shadowed by the dbt entry point that dbt-core installs into the virtualenv.
• [dagster-dbt] fetch_column_metadata() and fetch_row_counts() no longer fail the step when running on the dbt Fusion engine, or when dbt adapter initialization fails. They now log a warning and emit events without the metadata.
• [dagster-dbt] dagster-dbt's sqlglot dependency no longer pins an upper bound. Column lineage generation now works on sqlglot 28.1 and later, including the performance improvements in 30.x for wide models.
• [dagster-dbt] Asset materializations from dbt runs now carry the node's dbt status as metadata.

Bugfixes

• GraphQL event-log, tick, and run-record queries no longer fail when the requested limit exceeds the server-side default. The limit is clamped to the default, and the remaining rows are available by paginating.
• Fixed a bug where a run's asset list could be under-reported in the UI for runs with a large number of materializations, and where an asset could be reported as failed even though it was materialized by a different step than the one that planned it.
• Fixed an issue where the UI could keep displaying a code location's previous container image after a rolling update, until browser storage was cleared. (Thanks, @HynekBlaha!)
• [ui] Run filters now keep tag values that contain =, such as partition keys like region=west.
• [ui] Selection inputs no longer collapse repeated spaces inside quoted values, and now have an accessible name.
• [ui] On cursor-paginated lists such as Runs, "Newer" now returns to the previous page after you switch tabs and press browser Back, instead of jumping to the first page.
• [dagster-dbt] Fixed a bug where dbt models that finished with a no-op, reused, or warn status emitted no materialization, so the asset silently stopped updating and downstream freshness checks and automation conditions never fired. partial success and skipped are still treated as failures, since dbt marks a partial-success node's dependents as errored, and dbt Fusion reports both a genuine no-op and an upstream-failure skip as skipped.
• [dagster-dbt] Fixed an issue where dbt Fusion models with no ref() or source() calls were still left out of asset selection in environments without dbt-core installed, which the fix in 1.13.25 did not cover. Selections other than the default select are still affected.
• [dagster-dbt] Fixed a bug where, when running on dbt Fusion without dbt-core installed, every passing dbt test was reported as a failed asset check.
• [dagster-dbt] Fixed a bug where a dbt Cloud run containing seeds executed under dbt Fusion failed event translation with KeyError: 'materialized', dropping materializations for every asset after the first seed in the run.
• [dagster-dbt] Fixed a failure reloading a dagster-dbt project on Windows, where the local defs state snapshot could not be rebuilt if it contained read-only files such as a git checkout's object files.

Documentation

• Corrected the Dagster+ credit usage documentation: credits are counted per step execution plus per asset materialization event, and the page now covers retries, failures, and how Insights attributes credits across assets materialized in one step.