[Bug]: FnHarness.main not thread-safe on JvmInitializer calls

[Abacn]

What happened?

FnHarness.main usually started from a fresh jvm on worker. However, there is also an EmbeddedEnvironmentFactory that creates an environment where FnHarness is embedded, as a result a single JVM is used. This becomes a problem when running various pipelines at the same time, like in a gradle test task:

A suspected racing condition:

• Pipeline A loaded some java class at runtime

• Pipeline B execution startup, in JvmInitializers.beforeProcessing, gathered all JvmInitializers implementations

beam/sdks/java/harness/src/main/java/org/apache/beam/fn/harness/FnHarness.java

Line 400 in d3a841c

JvmInitializers.runBeforeProcessing(options);

beam/sdks/java/core/src/main/java/org/apache/beam/sdk/util/common/ReflectHelpers.java

Line 197 in d3a841c

ServiceLoader<T> loader = ServiceLoader.load(iface, classLoader);

• Pipeline A clean up, removed those classes from temp dir

• Pipeline B trying to fetch those class instances from ServiceLoader, however they are gone

In general, the JvmInitializer mechanism did not consider the scenario of embedded FnHarness. Running multiple tests in parallel using portable runner "initializes" jvm each time a FnHarness started.

This causes #30512 (comment)

Issue Priority

Priority: 2 (default / most bugs should be filed as P2)

Issue Components

• Component: Python SDK
• Component: Java SDK
• Component: Go SDK
• Component: Typescript SDK
• Component: IO connector
• Component: Beam YAML
• Component: Beam examples
• Component: Beam playground
• Component: Beam katas
• Component: Website
• Component: Infrastructure
• Component: Spark Runner
• Component: Flink Runner
• Component: Samza Runner
• Component: Twister2 Runner
• Component: Hazelcast Jet Runner
• Component: Google Cloud Dataflow Runner