Skip to content
startreedataPublic
forked from trinodb/trino

About

StarTree patches for the Trino Pinot connector. Connector-only releases matching upstream Trino versions.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

 
 

Repository files navigation

StarTree Pinot connector for Trino

This fork maintains patches to Trino's Pinot connector while changes make their way upstream. Releases contain only the complete Pinot plugin ZIP, including its runtime dependencies. Continue using the official Trino server or container.

The default development branch is pinot-dev, based on upstream master. Its initial patch stack contains all four commits from Trino PR #30881: authoritative server host lookup, background instance refresh, coordinator-side instance resolution, and synchronized gRPC client registration. Connector releases initially start at Trino 429 and continue with every later stable release.

Known limitation in the imported PR: worker instance metadata is shared between splits. During server host changes, overlapping splits can reuse another split's cached address. This fork initially preserves the selected upstream PR as-is.

Compatibility and downloads

Download from GitHub Releases. Each release is named <trino-version>-startree.<patch>, for example 483-startree.1, and is intended only for that exact Trino version. Do not use a 483 connector with a different Trino release. Trino's SPI can change between releases; see the upstream compatibility guidance.

Each release includes the plugin ZIP, a SHA-256 checksum, and JSON build metadata identifying its upstream version and source commits. The Maven project and Trino dependency versions remain the upstream release number; the StarTree suffix identifies the downloadable patched build. We do not publish over upstream Maven artifacts.

Installation

The following example assumes 483-startree.1 has been published and your cluster runs Trino 483. Substitute the release matching your cluster.

release=483-startree.1
archive="trino-pinot-${release}.zip"
base="https://github.com/startreedata/trino/releases/download/${release}"
curl --fail --location --remote-name "${base}/${archive}"
curl --fail --location --remote-name "${base}/${archive}.sha256"
sha256sum --check "${archive}.sha256"  # macOS: shasum -a 256 -c "${archive}.sha256"
unzip "${archive}" -d staged-pinot
  1. Stop Trino on every coordinator and worker.
  2. Move the current plugin/pinot directory to a backup outside plugin/.
  3. Copy the extracted directory containing the plugin JAR and dependency JARs into plugin/pinot on every node. Replace the entire directory; copying a single JAR or overlaying old dependencies can leave incompatible classes.
  4. Keep your existing Pinot catalog configuration (connector.name=pinot).
  5. Start all nodes and run your Pinot query checks. To roll back, stop all nodes, restore the backup directory on each node, and restart.

For the official Docker image the destination is /usr/lib/trino/plugin/pinot. Use a derived image or a directory mount to replace it. Each plugin directory must contain the JAR files directly; avoid an extra nested extraction directory. Do not retain a second copy of the Pinot plugin under another name in plugin/. See the upstream installation instructions.

Development and automatic releases

Commit connector development to pinot-dev; keep its patch history linear. Changes may touch plugin/trino-pinot and the Pinot connector documentation. Engine or SPI changes cannot be shipped as a drop-in connector. Upstream Trino workflow definitions are removed from downstream branches; only the connector workflows run.

The Pinot connector workflow checks daily for new upstream releases starting at 429, including historical releases from Maven Central that predate Trino's GitHub release downloads. It rebases pinot-dev directly onto upstream master. The release coordinator maintains the fork's master mirror separately, because copying upstream workflow changes requires permissions unavailable to the Actions token. A real source conflict stops the run for maintainer resolution; it never silently drops a conflicting patch. Pushes and pull requests to pinot-dev validate the complete patch stack against the latest stable Trino release without publishing.

For each new upstream release, automation:

  1. Pins the complete pinot-dev snapshot and its upstream base.
  2. Starts from the exact upstream release tag, retaining that release's Maven and dependency versions, and replays every downstream connector commit in order. Patches already present upstream are recorded as skipped. Tooling is copied from the same pinned development snapshot.
  3. Runs connector unit tests and loads the resulting complete plugin ZIP into the matching official Trino Docker image. If that image is unavailable, it uses the released server archive verified against Maven Central's checksum. The load check verifies packaging and catalog initialization; it does not validate queries against a live Pinot cluster. Existing cluster integration suites are excluded from this small packaging pipeline.
  4. Publishes <trino-version>-startree.<patch> with only the connector assets. A failed replay, build, or runtime check prevents publication. Existing release tags and assets are never replaced.

There is no per-release commit selection: every release receives the whole applicable development stack. Keep fixes in separate commits and send them upstream when practical. Add focused regression coverage for each fix.

After upstream incorporates the initial fixes, releases can continue without a downstream production diff. Automation requires matching commits (or their combined squash) in the target's history and verifies that their complete production diff is still present. An unfamiliar rewrite or a revert stops publication for review.

Manual releases and reproducibility

To publish the current development stack for a specific Trino version, run the workflow on pinot-dev. For example, a second patched build of Trino 483:

gh workflow run pinot-connector.yml --repo startreedata/trino --ref pinot-dev \
  -f trino_version=483 -f patch_number=2

The defaults are trino_version=latest and patch_number=1. Automatic runs skip versions that already have a published release. If an incomplete publication reserved a tag, the next automatic attempt uses a higher patch number. Use the next patch number when publishing additional fixes for an existing Trino version. Comma-separated version numbers can be used to retry a specific batch.

To backfill all missing releases from 429 onward:

gh workflow run pinot-connector.yml --repo startreedata/trino --ref pinot-dev \
  -f trino_version=all-missing

Historical builds use each upstream tag's Java version and dependencies. Compatibility adaptations preserve the complete development patch stack across older connector APIs and test frameworks; provenance records every adaptation. Missing Trino dependencies are built from that exact release's source. Trino 477 also requires its unpublished dependency modules to be built locally for validation; Trino 484 and later build the unpublished trino-json runtime dependency locally. Only the Pinot plugin is packaged for download.

Each release retains an immutable annotated development snapshot tag, pinot-dev-snapshot/<full-development-sha>. Its provenance records that snapshot, the upstream base and release SHAs, and the complete ordered commit list with replay/skip status. Automation also retains startree/dev-snapshots/<sha> branches before building, so GitHub can publish the pinned workflow files after development moves without a personal access token. To rebuild an older development stack after pinot-dev has moved or dropped fixes incorporated upstream, pass its retained snapshot:

gh workflow run pinot-connector.yml --repo startreedata/trino --ref pinot-dev \
  -f trino_version=483 -f patch_number=3 \
  -f dev_snapshot=pinot-dev-snapshot/<full-development-sha>

The reusable build workflow builds the connector and the unpublished dependencies described above, using released Trino artifacts for other dependencies. A master development version such as 484-SNAPSHOT is never relabeled as a stable release build. The workflow selects the JDK required by the targeted upstream release.

Upstream and license

Source is forked from trinodb/trino, under the Apache License 2.0. The full upstream source remains available for development and provenance. For general Trino development, see the upstream README and development guide.

About

StarTree patches for the Trino Pinot connector. Connector-only releases matching upstream Trino versions.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages