kaxil commented on code in PR #65958: URL: https://github.com/apache/airflow/pull/65958#discussion_r3295487547
########## task-sdk/src/airflow/sdk/coordinators/java/coordinator.py: ########## @@ -0,0 +1,261 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Java runtime coordinator that launches a JVM subprocess for Dag file processing and task execution.""" + +from __future__ import annotations + +import email +import os +import pathlib +import selectors +import socket +import subprocess +import time +import zipfile +from typing import TYPE_CHECKING, cast + +import attrs +import psutil +import structlog + +from airflow.sdk.execution_time.coordinator import BaseCoordinator +from airflow.sdk.execution_time.supervisor import ActivitySubprocess + +if TYPE_CHECKING: + from collections.abc import Sequence + + from structlog.typing import FilteringBoundLogger + from typing_extensions import Self + + from airflow.sdk.api.client import Client + from airflow.sdk.api.datamodels._generated import BundleInfo + from airflow.sdk.execution_time.workloads.task import TaskInstanceDTO + +log: FilteringBoundLogger = structlog.get_logger(logger_name="coordinators.java") + + +def _start_server() -> socket.socket: + server = socket.socket() + server.bind(("127.0.0.1", 0)) + server.setblocking(True) + server.listen(1) # Just need to listen to the child process. + return server + + +def _calculate_classpath(jars_root: Sequence[pathlib.Path]) -> str: + jars = (p.as_posix() for root in jars_root for p in root.iterdir() if p.suffix == ".jar") + return os.pathsep.join(jars) + + [email protected] +class _MainJar: + path: pathlib.Path + main_class: str + schema_version: str | None + + @classmethod + def find(cls, jars_root: Sequence[pathlib.Path]) -> Self: + for root in jars_root: + for p in root.iterdir(): Review Comment: `_calculate_classpath` (line 61) builds the JVM classpath via `pathlib.Path.iterdir()`, whose order is filesystem-dependent. When two JARs in `jars_root` contain a class with the same fully-qualified name (common with shaded/relocated user JARs), the JVM picks whichever JAR is *earlier* on the classpath -- so two machines with identical JARs but different filesystem ordering get different class-resolution behaviour silently. Suggest `sorted(root.iterdir())` for reproducible classpath ordering. (Edit: my original draft also raised the `_MainJar.find` brittleness -- `first JAR with Main-Class` being filesystem-order-dependent, with the picked JAR's `Airflow-SDK-Supervisor-Schema-Version` flowing to the migrator. The active thread on line 83 covers that: @uranusjr's planned explicit `main_class` kwarg deterministically anchors discovery, so the schema-version selection becomes deterministic too. Narrowing this comment to just the classpath ordering.) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
