[ML] Deprecate machine learning on Intel macOS (#104087)

PyTorch is no longer going to provide macOS x86_64 builds after
version 2.2. This doesn't instantly affect us, as we build PyTorch
from source ourselves and they've said that they will not
deliberately break macOS x86_64. They just won't build or test on
that platform themselves. As a result it's inevitable that we'll
have to make some tweaks to the PyTorch code to get it to build,
and as the years go by it will become harder and harder to make
the code compile on an unsupported platform. Since PyTorch is
such a critical component of Elastic ML we won't be able to keep
it running on macOS x86_64 for more than a few releases after
PyTorch drops support. This change gives notice of our intentions.

Author: David Roberts

docs/changelog/104087.yaml

@@ -0,0 +1,13 @@
+pr: 104087
+summary: Deprecate machine learning on Intel macOS
+area: Machine Learning
+type: deprecation
+issues: []
+deprecation:
+  title: Deprecate machine learning on Intel macOS
+  area: Packaging
+  details: The machine learning plugin will be permanently disabled on macOS x86_64
+    in new minor versions released from December 2024 onwards.
+  impact: To continue to use machine learning functionality on macOS please switch to
+    an arm64 machine (Apple silicon). Alternatively, it will still be possible to run
+    Elasticsearch with machine learning enabled in a Docker container on macOS x86_64.

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MachineLearning.java

@@ -32,6 +32,8 @@
 import org.elasticsearch.cluster.service.ClusterService;
 import org.elasticsearch.common.breaker.CircuitBreaker;
 import org.elasticsearch.common.io.stream.NamedWriteableRegistry;
+import org.elasticsearch.common.logging.DeprecationCategory;
+import org.elasticsearch.common.logging.DeprecationLogger;
 import org.elasticsearch.common.settings.ClusterSettings;
 import org.elasticsearch.common.settings.IndexScopedSettings;
 import org.elasticsearch.common.settings.Setting;
@@ -68,6 +70,7 @@
 import org.elasticsearch.plugins.IngestPlugin;
 import org.elasticsearch.plugins.MapperPlugin;
 import org.elasticsearch.plugins.PersistentTaskPlugin;
+import org.elasticsearch.plugins.Platforms;
 import org.elasticsearch.plugins.Plugin;
 import org.elasticsearch.plugins.SearchPlugin;
 import org.elasticsearch.plugins.ShutdownAwarePlugin;
@@ -753,6 +756,7 @@ public void loadExtensions(ExtensionLoader loader) {
     public static final int MAX_LOW_PRIORITY_MODELS_PER_NODE = 100;
 
     private static final Logger logger = LogManager.getLogger(MachineLearning.class);
+    private static final DeprecationLogger deprecationLogger = DeprecationLogger.getLogger(MachineLearning.class);
 
     private final Settings settings;
     private final boolean enabled;
@@ -919,6 +923,15 @@ public Collection<?> createComponents(PluginServices services) {
             return List.of(new JobManagerHolder(), new MachineLearningExtensionHolder());
         }
 
+        if ("darwin-x86_64".equals(Platforms.PLATFORM_NAME)) {
+            String msg = "The machine learning plugin will be permanently disabled on macOS x86_64 in new minor versions released "
+                + "from December 2024 onwards. To continue to use machine learning functionality on macOS please switch to an arm64 "
+                + "machine (Apple silicon). Alternatively, it will still be possible to run Elasticsearch with machine learning "
+                + "enabled in a Docker container on macOS x86_64.";
+            logger.warn(msg);
+            deprecationLogger.warn(DeprecationCategory.PLUGINS, "ml-darwin-x86_64", msg);
+        }
+
         machineLearningExtension.get().configure(environment.settings());
 
         this.mlUpgradeModeActionFilter.set(new MlUpgradeModeActionFilter(clusterService));