Skip to content

feat(ci): try to speed-up pipelines - #69

Merged
giggsoff merged 8 commits into
develop/4.4.0/4.2.0.2from
feature/ci-change-detection
Sep 22, 2026
Merged

giggsoff merged 8 commits into
develop/4.4.0/4.2.0.2from
feature/ci-change-detection

Conversation

@giggsoff

@giggsoff giggsoff commented Sep 17, 2026 •

Copy link
Copy Markdown
Collaborator
  • skip heavy CI jobs if PR not affect them
  • optimize torch and arrow install
  • build spark in separate job and store for a day to consume from the rest of jobs
  • use public mirror

A precondition job uses dev/is-changed.py to gate the pyspark, SparkR, Docker, Kubernetes and TPC-DS jobs. Pushes, manual runs and PRs to release branches still run everything.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
The PyPI torch and torchvision builds pull in several GB of CUDA libraries that the runners never use.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
Posit Package Manager serves a binary for Ubuntu 24.04 when the R user agent is set, so the package is no longer built from CRAN sources.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
The build-spark job uploads the installed Maven artifacts and the build output. The build matrix, pyspark, SparkR, Docker and Kubernetes jobs restore them instead of rebuilding Spark.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
@giggsoff giggsoff changed the title Feature/ci change detection feat(ci): try to speed-up pipelines Sep 18, 2026
The mirror proxies GitHub Packages without credentials, so builds, Hive client tests and CI no longer need a token or the custom Ivy settings. Deploys still go to GitHub Packages.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
ADD JAR ivy:// and spark.jars.packages resolve through spark.jars.repositories rather than IsolatedClientLoader, so the tests need the mirror there as well now that dev/ivysettings.xml is gone.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
The Spark Connect client pushes every spark.sql.* system property to its server, so a global spark.sql.maven.additionalRemoteRepositories made each UnsupportedFeaturesSuite test retry for about ten minutes.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
SparkSQLEnvSuite resolves Hive through IsolatedClientLoader with metastore jars from Maven, so it needs the Arenadata mirror just like the sql/hive tests.

Signed-off-by: Petr Fedchenkov <giggsoff@gmail.com>
@giggsoff
giggsoff marked this pull request as ready for review September 21, 2026 07:43
@giggsoff
giggsoff merged commit dc79ada into develop/4.4.0/4.2.0.2 Sep 22, 2026
47 of 48 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants