Choosing a JAR¶

Flare is published once per Spark line and Scala version. The wrong one fails at class-load time, often with an unhelpful error, so match it to what the cluster runs.

Spark 3.3 is deprecated in 1.4 and its JARs end with 2.0; see Upgrading.

All are under the Maven group io.github.neutrinic, for example:

io.github.neutrinic:flare-spark-3-5_2.12:1.4.0

The JAR is self-contained: it bundles the OpenTelemetry API it needs and nothing else. See Downloading the JARs.

Finding your versions¶

spark-submit --version

prints both, for example version 3.5.1 and Using Scala version 2.12.18. On a managed platform, the runtime's release notes list them. The ones that catch people out:

Dataproc Serverless 2.3 is Scala 2.13 while Dataproc on Compute Engine 2.3 is 2.12. A 2.12 JAR on a 2.13 runtime fails with NoClassDefFoundError: scala/Serializable.

Spark 4.1 and later¶

There is no separate artifact yet. The 4-0_2.13 JAR has been run on Spark 4.1.2 (Dataproc 3.0) with a complete trace. Later Spark releases have not been run.

Java¶

Every Spark 3.x artifact runs on Java 8 and later, and the 4.0 artifact on Java 17 and later, the same floors as Spark itself. This matters on Databricks Runtime 14 and 15, which default to Java 8, and on Dataproc on Compute Engine 2.x, which runs Java 11. Dataproc Serverless 2.3 runs Java 17. Releases before 1.3.0 exported nothing on Java 8 or 11; see Upgrading.