Move shared cache pre-allocation and support macOS (elastic#70285)

This commit moves the shared cache pre-allocation code out of bootstrap,
and instead to the searchable snapshots code. We go out of our way to
only grant permissions to a specific library used for the
pre-allocation.

Additionally, to ensure our interfaces are sound, we add a macOS
implementation based on fcntl and ftruncate.

Co-authored-by: Yannick Welsch <[email protected]>

Author: jasontedor

server/src/main/java/org/elasticsearch/bootstrap/Bootstrap.java

@@ -42,7 +42,6 @@
 import org.elasticsearch.node.InternalSettingsPreparer;
 import org.elasticsearch.node.Node;
 import org.elasticsearch.node.NodeValidationException;
-import org.elasticsearch.snapshots.SnapshotsService;
 
 import java.io.ByteArrayOutputStream;
 import java.io.IOException;
@@ -172,18 +171,6 @@ private void setup(boolean addShutdownHook, Environment environment) throws Boot
                 BootstrapSettings.SYSTEM_CALL_FILTER_SETTING.get(settings),
                 BootstrapSettings.CTRLHANDLER_SETTING.get(settings));
 
-        final long cacheSize = SnapshotsService.SNAPSHOT_CACHE_SIZE_SETTING.get(settings).getBytes();
-        final long regionSize = SnapshotsService.SNAPSHOT_CACHE_REGION_SIZE_SETTING.get(settings).getBytes();
-        final int numRegions = Math.toIntExact(cacheSize / regionSize);
-        final long fileSize = numRegions * regionSize;
-        if (fileSize > 0) {
-            try {
-                Natives.tryCreateCacheFile(environment, fileSize);
-            } catch (Exception e) {
-                throw new BootstrapException(e);
-            }
-        }
-
         // initialize probes before the security manager is installed
         initializeProbes();
 

server/src/main/java/org/elasticsearch/bootstrap/JNAFalloc.java

@@ -11,20 +11,11 @@
 import com.sun.jna.Native;
 import com.sun.jna.Pointer;
 import com.sun.jna.WString;
-
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
-import org.apache.logging.log4j.message.ParameterizedMessage;
 import org.apache.lucene.util.Constants;
-import org.elasticsearch.common.SuppressForbidden;
-import org.elasticsearch.env.Environment;
 import org.elasticsearch.monitor.jvm.JvmInfo;
-import org.elasticsearch.snapshots.SnapshotUtils;
 
-import java.io.FileOutputStream;
-import java.io.IOException;
-import java.lang.reflect.Field;
-import java.nio.file.Files;
 import java.nio.file.Path;
 
 import static org.elasticsearch.bootstrap.JNAKernel32Library.SizeT;
@@ -269,39 +260,4 @@ static void tryInstallSystemCallFilter(Path tmpFile) {
         }
     }
 
-    @SuppressForbidden(reason = "need access to fd on FileOutputStream")
-    static void fallocateSnapshotCacheFile(Environment environment, long fileSize) throws IOException {
-        final JNAFalloc falloc = JNAFalloc.falloc();
-        if (falloc == null) {
-            logger.debug("not trying to create a shared cache file using fallocate because native fallocate library could not be loaded.");
-            return;
-        }
-
-        Path cacheFile = SnapshotUtils.findCacheSnapshotCacheFilePath(environment, fileSize);
-        if (cacheFile == null) {
-            throw new IOException("could not find a directory with adequate free space for cache file");
-        }
-        boolean success = false;
-        try (FileOutputStream fileChannel = new FileOutputStream(cacheFile.toFile())) {
-            long currentSize = fileChannel.getChannel().size();
-            if (currentSize < fileSize) {
-                final Field field = fileChannel.getFD().getClass().getDeclaredField("fd");
-                field.setAccessible(true);
-                final int result = falloc.fallocate((int) field.get(fileChannel.getFD()), currentSize, fileSize - currentSize);
-                if (result == 0) {
-                    success = true;
-                    logger.info("allocated cache file [{}] using fallocate", cacheFile);
-                } else {
-                    logger.warn("failed to initialize cache file [{}] using fallocate errno [{}]", cacheFile, result);
-                }
-            }
-        } catch (Exception e) {
-            logger.warn(new ParameterizedMessage("failed to initialize cache file [{}] using fallocate", cacheFile), e);
-        } finally {
-            if (success == false) {
-                // if anything goes wrong, delete the potentially created file to not waste disk space
-                Files.deleteIfExists(cacheFile);
-            }
-        }
-    }
 }

server/src/main/java/org/elasticsearch/bootstrap/JNANatives.java

@@ -10,9 +10,7 @@
 
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
-import org.elasticsearch.env.Environment;
 
-import java.io.IOException;
 import java.nio.file.Path;
 
 /**
@@ -135,19 +133,4 @@ static boolean isSystemCallFilterInstalled() {
         return JNANatives.LOCAL_SYSTEM_CALL_FILTER;
     }
 
-    /**
-     * On Linux, this method tries to create the searchable snapshot frozen cache file using fallocate if JNA is available. This enables
-     * a much faster creation of the file than the fallback mechanism in the searchable snapshots plugin that will pre-allocate the cache
-     * file by writing zeros to the file.
-     *
-     * @throws IOException on failure to determine free disk space for a data path
-     */
-    public static void tryCreateCacheFile(Environment environment, long fileSize) throws IOException {
-        if (JNA_AVAILABLE == false) {
-            logger.warn("cannot use fallocate to create cache file because JNA is not available");
-            return;
-        }
-        JNANatives.fallocateSnapshotCacheFile(environment, fileSize);
-    }
-
 }

server/src/main/java/org/elasticsearch/bootstrap/Natives.java

@@ -9,15 +9,16 @@
 
 import org.elasticsearch.action.support.IndicesOptions;
 import org.elasticsearch.cluster.metadata.IndexNameExpressionResolver;
-import org.elasticsearch.common.Nullable;
 import org.elasticsearch.common.regex.Regex;
-import org.elasticsearch.env.Environment;
 import org.elasticsearch.index.IndexNotFoundException;
 
+<<<<<<< HEAD
 import java.util.ArrayList;
 import java.io.IOException;
 import java.nio.file.Files;
 import java.nio.file.Path;
+=======
+>>>>>>> 36683a4581a... Move shared cache pre-allocation and support macOS (#70285)
 import java.util.Arrays;
 import java.util.Collections;
 import java.util.HashSet;
@@ -115,28 +116,4 @@ public static List<String> filterIndices(List<String> availableIndices, String[]
         return Collections.unmodifiableList(new ArrayList<>(result));
     }
 
-    /**
-     * Tries to find a suitable path to a searchable snapshots shared cache file in the data paths founds in the environment.
-     *
-     * @return path for the cache file or {@code null} if none could be found
-     */
-    @Nullable
-    public static Path findCacheSnapshotCacheFilePath(Environment environment, long fileSize) throws IOException {
-        Path cacheFile = null;
-        for (Path path : environment.dataFiles()) {
-            Files.createDirectories(path);
-            // TODO: be resilient to this check failing and try next path?
-            long usableSpace = Environment.getUsableSpace(path);
-            Path p = path.resolve(SnapshotsService.CACHE_FILE_NAME);
-            if (Files.exists(p)) {
-                usableSpace += Files.size(p);
-            }
-            // TODO: leave some margin for error here
-            if (usableSpace > fileSize) {
-                cacheFile = p;
-                break;
-            }
-        }
-        return cacheFile;
-    }
 }

server/src/main/java/org/elasticsearch/snapshots/SnapshotUtils.java

@@ -37,9 +37,6 @@
 import org.elasticsearch.cluster.RestoreInProgress;
 import org.elasticsearch.cluster.SnapshotDeletionsInProgress;
 import org.elasticsearch.cluster.SnapshotsInProgress;
-import org.elasticsearch.common.unit.ByteSizeValue;
-import org.elasticsearch.common.util.CollectionUtils;
-import org.elasticsearch.repositories.RepositoryShardId;
 import org.elasticsearch.cluster.SnapshotsInProgress.ShardSnapshotStatus;
 import org.elasticsearch.cluster.SnapshotsInProgress.ShardState;
 import org.elasticsearch.cluster.SnapshotsInProgress.State;
@@ -67,7 +64,11 @@
 import org.elasticsearch.common.regex.Regex;
 import org.elasticsearch.common.settings.Setting;
 import org.elasticsearch.common.settings.Settings;
+<<<<<<< HEAD
 import org.elasticsearch.common.util.concurrent.AbstractRunnable;
+=======
+import org.elasticsearch.common.util.CollectionUtils;
+>>>>>>> 36683a4581a... Move shared cache pre-allocation and support macOS (#70285)
 import org.elasticsearch.index.Index;
 import org.elasticsearch.index.shard.ShardId;
 import org.elasticsearch.indices.SystemIndices;
@@ -77,6 +78,7 @@
 import org.elasticsearch.repositories.RepositoryData;
 import org.elasticsearch.repositories.RepositoryException;
 import org.elasticsearch.repositories.RepositoryMissingException;
+import org.elasticsearch.repositories.RepositoryShardId;
 import org.elasticsearch.repositories.ShardGenerations;
 import org.elasticsearch.threadpool.ThreadPool;
 import org.elasticsearch.transport.TransportService;
@@ -143,26 +145,6 @@ public class SnapshotsService extends AbstractLifecycleComponent implements Clus
 
     public static final String UPDATE_SNAPSHOT_STATUS_ACTION_NAME = "internal:cluster/snapshot/update_snapshot_status";
 
-    public static final String SHARED_CACHE_SETTINGS_PREFIX = "xpack.searchable.snapshot.shared_cache.";
-
-    public static final Setting<ByteSizeValue> SHARED_CACHE_RANGE_SIZE_SETTING = Setting.byteSizeSetting(
-        SHARED_CACHE_SETTINGS_PREFIX + "range_size",
-        ByteSizeValue.ofMb(16),                                 // default
-        Setting.Property.NodeScope
-    );
-    public static final Setting<ByteSizeValue> SNAPSHOT_CACHE_REGION_SIZE_SETTING = Setting.byteSizeSetting(
-        SHARED_CACHE_SETTINGS_PREFIX + "region_size",
-        SHARED_CACHE_RANGE_SIZE_SETTING,
-        Setting.Property.NodeScope
-    );
-    public static final Setting<ByteSizeValue> SNAPSHOT_CACHE_SIZE_SETTING = Setting.byteSizeSetting(
-        SHARED_CACHE_SETTINGS_PREFIX + "size",
-        ByteSizeValue.ZERO,
-        Setting.Property.NodeScope
-    );
-
-    public static final String CACHE_FILE_NAME = "shared_snapshot_cache";
-
     public static final String NO_FEATURE_STATES_VALUE = "none";
 
     private final ClusterService clusterService;

server/src/main/java/org/elasticsearch/snapshots/SnapshotsService.java

@@ -12,7 +12,8 @@ archivesBaseName = 'x-pack-searchable-snapshots'
 
 dependencies {
   compileOnly project(path: xpackModule('core'))
+  implementation project(path: 'preallocate')
   internalClusterTestImplementation(testArtifact(project(xpackModule('core'))))
 }
 
-addQaCheckDependencies()
+addQaCheckDependencies()

x-pack/plugin/searchable-snapshots/build.gradle

@@ -0,0 +1,16 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+apply plugin: 'elasticsearch.build'
+
+archivesBaseName = 'preallocate'
+
+dependencies {
+  compileOnly project(":server")
+}
+
+tasks.named("testingConventions").configure { enabled = false }

x-pack/plugin/searchable-snapshots/preallocate/build.gradle

@@ -0,0 +1,55 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+package org.elasticsearch.xpack.searchablesnapshots.preallocate;
+
+import com.sun.jna.Native;
+import com.sun.jna.Platform;
+
+import java.security.AccessController;
+import java.security.PrivilegedAction;
+
+final class LinuxPreallocator implements Preallocator {
+
+    @Override
+    public boolean available() {
+        return Natives.NATIVES_AVAILABLE;
+    }
+
+    @Override
+    public int preallocate(final int fd, final long currentSize, final long fileSize) {
+        final int rc = Natives.fallocate(fd, 0, currentSize, fileSize - currentSize);
+        return rc == 0 ? 0 : Native.getLastError();
+    }
+
+    @Override
+    public String error(int errno) {
+        return Natives.strerror(errno);
+    }
+
+    private static class Natives {
+
+        public static final boolean NATIVES_AVAILABLE;
+
+        static {
+            NATIVES_AVAILABLE = AccessController.doPrivileged((PrivilegedAction<Boolean>) () -> {
+                try {
+                    Native.register(Natives.class, Platform.C_LIBRARY_NAME);
+                } catch (final UnsatisfiedLinkError e) {
+                    return false;
+                }
+                return true;
+            });
+        }
+
+        static native int fallocate(int fd, int mode, long offset, long length);
+
+        static native String strerror(int errno);
+
+    }
+
+}