drop index.shard.check_on_startup: fix (#32279)

Relates #31389

(cherry picked from commit 3d82a30)

Author: vladimirdolzhenko

docs/reference/index-modules.asciidoc

@@ -65,9 +65,7 @@ corruption is detected, it will prevent the shard from being opened. Accepts:
 
 `fix`::
 
-    Check for both physical and logical corruption.  Segments that were reported
-    as corrupted will be automatically removed. This option *may result in data loss*.
-    Use with extreme caution!
+    The same as `false`. This option is deprecated and will be completely removed in 7.0.
 
 WARNING: Expert only. Checking shards may take a lot of time on large indices.
 --

server/src/main/java/org/elasticsearch/index/shard/IndexShard.java

@@ -297,6 +297,10 @@ public IndexShard(
         logger.debug("state: [CREATED]");
 
         this.checkIndexOnStartup = indexSettings.getValue(IndexSettings.INDEX_CHECK_ON_STARTUP);
+        if ("fix".equals(checkIndexOnStartup)) {
+            deprecationLogger.deprecated("Setting [index.shard.check_on_startup] is set to deprecated value [fix], "
+                + "which has no effect and will not be accepted in future");
+        }
         this.translogConfig = new TranslogConfig(shardId, shardPath().resolveTranslog(), indexSettings, bigArrays);
         final String aId = shardRouting.allocationId().getId();
         this.globalCheckpointListeners = new GlobalCheckpointListeners(shardId, threadPool.executor(ThreadPool.Names.LISTENER), logger);
@@ -1354,7 +1358,7 @@ private void innerOpenEngineAndTranslog() throws IOException {
         }
         recoveryState.setStage(RecoveryState.Stage.VERIFY_INDEX);
         // also check here, before we apply the translog
-        if (Booleans.isTrue(checkIndexOnStartup)) {
+        if (Booleans.isTrue(checkIndexOnStartup) || "checksum".equals(checkIndexOnStartup)) {
             try {
                 checkIndex();
             } catch (IOException ex) {
@@ -1958,6 +1962,9 @@ void checkIndex() throws IOException {
         if (store.tryIncRef()) {
             try {
                 doCheckIndex();
+            } catch (IOException e) {
+                store.markStoreCorrupted(e);
+                throw e;
             } finally {
                 store.decRef();
             }
@@ -2001,18 +2008,7 @@ private void doCheckIndex() throws IOException {
                     return;
                 }
                 logger.warn("check index [failure]\n{}", os.bytes().utf8ToString());
-                if ("fix".equals(checkIndexOnStartup)) {
-                    if (logger.isDebugEnabled()) {
-                        logger.debug("fixing index, writing new segments file ...");
-                    }
-                    store.exorciseIndex(status);
-                    if (logger.isDebugEnabled()) {
-                        logger.debug("index fixed, wrote new segments file \"{}\"", status.segmentsFileName);
-                    }
-                } else {
-                    // only throw a failure if we are not going to fix the index
-                    throw new IllegalStateException("index check failure but can't fix it");
-                }
+                throw new IOException("index check failure");
             }
         }
 

server/src/main/java/org/elasticsearch/index/store/Store.java

@@ -134,7 +134,8 @@ public class Store extends AbstractIndexShardComponent implements Closeable, Ref
     static final int VERSION_STACK_TRACE = 1; // we write the stack trace too since 1.4.0
     static final int VERSION_START = 0;
     static final int VERSION = VERSION_WRITE_THROWABLE;
-    static final String CORRUPTED = "corrupted_";
+    // public is for test purposes
+    public static final String CORRUPTED = "corrupted_";
     public static final Setting<TimeValue> INDEX_STORE_STATS_REFRESH_INTERVAL_SETTING =
         Setting.timeSetting("index.store.stats_refresh_interval", TimeValue.timeValueSeconds(10), Property.IndexScope);
 
@@ -360,18 +361,6 @@ public CheckIndex.Status checkIndex(PrintStream out) throws IOException {
         }
     }
 
-    /**
-     * Repairs the index using the previous returned status from {@link #checkIndex(PrintStream)}.
-     */
-    public void exorciseIndex(CheckIndex.Status status) throws IOException {
-        metadataLock.writeLock().lock();
-        try (CheckIndex checkIndex = new CheckIndex(directory)) {
-            checkIndex.exorciseIndex(status);
-        } finally {
-            metadataLock.writeLock().unlock();
-        }
-    }
-
     public StoreStats stats() throws IOException {
         ensureOpen();
         return new StoreStats(directory.estimateSize());

server/src/test/java/org/elasticsearch/index/shard/IndexShardTests.java

@@ -22,6 +22,7 @@
 import org.apache.lucene.index.CorruptIndexException;
 import org.apache.lucene.index.DirectoryReader;
 import org.apache.lucene.index.IndexCommit;
+import org.apache.lucene.index.IndexWriter;
 import org.apache.lucene.index.Term;
 import org.apache.lucene.search.IndexSearcher;
 import org.apache.lucene.search.TermQuery;
@@ -110,6 +111,7 @@
 import org.elasticsearch.snapshots.SnapshotId;
 import org.elasticsearch.snapshots.SnapshotInfo;
 import org.elasticsearch.snapshots.SnapshotShardFailure;
+import org.elasticsearch.test.CorruptionUtils;
 import org.elasticsearch.test.DummyShardLock;
 import org.elasticsearch.test.FieldMaskingReader;
 import org.elasticsearch.test.VersionUtils;
@@ -118,7 +120,11 @@
 
 import java.io.IOException;
 import java.nio.charset.Charset;
+import java.nio.file.FileVisitResult;
+import java.nio.file.Files;
 import java.nio.file.Path;
+import java.nio.file.SimpleFileVisitor;
+import java.nio.file.attribute.BasicFileAttributes;
 import java.util.ArrayList;
 import java.util.Arrays;
 import java.util.Collections;
@@ -1229,7 +1235,7 @@ public String[] listAll() throws IOException {
         };
 
         try (Store store = createStore(shardId, new IndexSettings(metaData, Settings.EMPTY), directory)) {
-            IndexShard shard = newShard(shardRouting, shardPath, metaData, store,
+            IndexShard shard = newShard(shardRouting, shardPath, metaData, i -> store,
                     null, new InternalEngineFactory(), () -> {
                     }, EMPTY_EVENT_LISTENER);
             AtomicBoolean failureCallbackTriggered = new AtomicBoolean(false);
@@ -2569,6 +2575,143 @@ public void testReadSnapshotConcurrently() throws IOException, InterruptedExcept
         closeShards(newShard);
     }
 
+    public void testIndexCheckOnStartup() throws Exception {
+        final IndexShard indexShard = newStartedShard(true);
+
+        final long numDocs = between(10, 100);
+        for (long i = 0; i < numDocs; i++) {
+            indexDoc(indexShard, "_doc", Long.toString(i), "{}");
+        }
+        indexShard.flush(new FlushRequest());
+        closeShards(indexShard);
+
+        final ShardPath shardPath = indexShard.shardPath();
+
+        final Path indexPath = corruptIndexFile(shardPath);
+
+        final AtomicInteger corruptedMarkerCount = new AtomicInteger();
+        final SimpleFileVisitor<Path> corruptedVisitor = new SimpleFileVisitor<Path>() {
+            @Override
+            public FileVisitResult visitFile(Path file, BasicFileAttributes attrs) throws IOException {
+                if (Files.isRegularFile(file) && file.getFileName().toString().startsWith(Store.CORRUPTED)) {
+                    corruptedMarkerCount.incrementAndGet();
+                }
+                return FileVisitResult.CONTINUE;
+            }
+        };
+        Files.walkFileTree(indexPath, corruptedVisitor);
+
+        assertThat("corruption marker should not be there", corruptedMarkerCount.get(), equalTo(0));
+
+        final ShardRouting shardRouting = ShardRoutingHelper.initWithSameId(indexShard.routingEntry(),
+            RecoverySource.StoreRecoverySource.EXISTING_STORE_INSTANCE
+        );
+        // start shard and perform index check on startup. It enforce shard to fail due to corrupted index files
+        final IndexMetaData indexMetaData = IndexMetaData.builder(indexShard.indexSettings().getIndexMetaData())
+            .settings(Settings.builder()
+                .put(indexShard.indexSettings.getSettings())
+                .put(IndexSettings.INDEX_CHECK_ON_STARTUP.getKey(), randomFrom("true", "checksum")))
+            .build();
+
+        IndexShard corruptedShard = newShard(shardRouting, shardPath, indexMetaData,
+            null, null, indexShard.engineFactory,
+            indexShard.getGlobalCheckpointSyncer(), EMPTY_EVENT_LISTENER);
+
+        final IndexShardRecoveryException indexShardRecoveryException =
+            expectThrows(IndexShardRecoveryException.class, () -> newStartedShard(p -> corruptedShard, true));
+        assertThat(indexShardRecoveryException.getMessage(), equalTo("failed recovery"));
+
+        // check that corrupt marker is there
+        Files.walkFileTree(indexPath, corruptedVisitor);
+        assertThat("store has to be marked as corrupted", corruptedMarkerCount.get(), equalTo(1));
+
+        try {
+            closeShards(corruptedShard);
+        } catch (RuntimeException e) {
+            assertThat(e.getMessage(), equalTo("CheckIndex failed"));
+        }
+    }
+
+    public void testShardDoesNotStartIfCorruptedMarkerIsPresent() throws Exception {
+        final IndexShard indexShard = newStartedShard(true);
+
+        final long numDocs = between(10, 100);
+        for (long i = 0; i < numDocs; i++) {
+            indexDoc(indexShard, "_doc", Long.toString(i), "{}");
+        }
+        indexShard.flush(new FlushRequest());
+        closeShards(indexShard);
+
+        final ShardPath shardPath = indexShard.shardPath();
+
+        final ShardRouting shardRouting = ShardRoutingHelper.initWithSameId(indexShard.routingEntry(),
+            RecoverySource.StoreRecoverySource.EXISTING_STORE_INSTANCE
+        );
+        final IndexMetaData indexMetaData = indexShard.indexSettings().getIndexMetaData();
+
+        final Path indexPath = shardPath.getDataPath().resolve(ShardPath.INDEX_FOLDER_NAME);
+
+        // create corrupted marker
+        final String corruptionMessage = "fake ioexception";
+        try(Store store = createStore(indexShard.indexSettings(), shardPath)) {
+            store.markStoreCorrupted(new IOException(corruptionMessage));
+        }
+
+        // try to start shard on corrupted files
+        final IndexShard corruptedShard = newShard(shardRouting, shardPath, indexMetaData,
+            null, null, indexShard.engineFactory,
+            indexShard.getGlobalCheckpointSyncer(), EMPTY_EVENT_LISTENER);
+
+        final IndexShardRecoveryException exception1 = expectThrows(IndexShardRecoveryException.class,
+            () -> newStartedShard(p -> corruptedShard, true));
+        assertThat(exception1.getCause().getMessage(), equalTo(corruptionMessage + " (resource=preexisting_corruption)"));
+        closeShards(corruptedShard);
+
+        final AtomicInteger corruptedMarkerCount = new AtomicInteger();
+        final SimpleFileVisitor<Path> corruptedVisitor = new SimpleFileVisitor<Path>() {
+            @Override
+            public FileVisitResult visitFile(Path file, BasicFileAttributes attrs) throws IOException {
+                if (Files.isRegularFile(file) && file.getFileName().toString().startsWith(Store.CORRUPTED)) {
+                    corruptedMarkerCount.incrementAndGet();
+                }
+                return FileVisitResult.CONTINUE;
+            }
+        };
+        Files.walkFileTree(indexPath, corruptedVisitor);
+        assertThat("store has to be marked as corrupted", corruptedMarkerCount.get(), equalTo(1));
+
+        // try to start another time shard on corrupted files
+        final IndexShard corruptedShard2 = newShard(shardRouting, shardPath, indexMetaData,
+            null, null, indexShard.engineFactory,
+            indexShard.getGlobalCheckpointSyncer(), EMPTY_EVENT_LISTENER);
+
+        final IndexShardRecoveryException exception2 = expectThrows(IndexShardRecoveryException.class,
+            () -> newStartedShard(p -> corruptedShard2, true));
+        assertThat(exception2.getCause().getMessage(), equalTo(corruptionMessage + " (resource=preexisting_corruption)"));
+        closeShards(corruptedShard2);
+
+        // check that corrupt marker is there
+        corruptedMarkerCount.set(0);
+        Files.walkFileTree(indexPath, corruptedVisitor);
+        assertThat("store still has a single corrupt marker", corruptedMarkerCount.get(), equalTo(1));
+    }
+
+    private Path corruptIndexFile(ShardPath shardPath) throws IOException {
+        final Path indexPath = shardPath.getDataPath().resolve(ShardPath.INDEX_FOLDER_NAME);
+        final Path[] filesToCorrupt =
+            Files.walk(indexPath)
+                .filter(p -> {
+                    final String name = p.getFileName().toString();
+                    return Files.isRegularFile(p)
+                        && name.startsWith("extra") == false // Skip files added by Lucene's ExtrasFS
+                        && IndexWriter.WRITE_LOCK_NAME.equals(name) == false
+                        && name.startsWith("segments_") == false && name.endsWith(".si") == false;
+                })
+                .toArray(Path[]::new);
+        CorruptionUtils.corruptFile(random(), filesToCorrupt);
+        return indexPath;
+    }
+
     /**
      * Simulates a scenario that happens when we are async fetching snapshot metadata from GatewayService
      * and checking index concurrently. This should always be possible without any exception.
@@ -2592,7 +2735,7 @@ public void testReadSnapshotAndCheckIndexConcurrently() throws Exception {
         final IndexMetaData indexMetaData = IndexMetaData.builder(indexShard.indexSettings().getIndexMetaData())
             .settings(Settings.builder()
                 .put(indexShard.indexSettings.getSettings())
-                .put(IndexSettings.INDEX_CHECK_ON_STARTUP.getKey(), randomFrom("false", "true", "checksum", "fix")))
+                .put(IndexSettings.INDEX_CHECK_ON_STARTUP.getKey(), randomFrom("false", "true", "checksum")))
             .build();
         final IndexShard newShard = newShard(shardRouting, indexShard.shardPath(), indexMetaData,
                 null, null, indexShard.engineFactory, indexShard.getGlobalCheckpointSyncer(), EMPTY_EVENT_LISTENER);
@@ -2634,6 +2777,16 @@ public void testReadSnapshotAndCheckIndexConcurrently() throws Exception {
         closeShards(newShard);
     }
 
+    public void testCheckOnStartupDeprecatedValue() throws Exception {
+        final Settings settings = Settings.builder().put(IndexSettings.INDEX_CHECK_ON_STARTUP.getKey(), "fix").build();
+
+        final IndexShard newShard = newShard(true, settings);
+        closeShards(newShard);
+
+        assertWarnings("Setting [index.shard.check_on_startup] is set to deprecated value [fix], "
+            + "which has no effect and will not be accepted in future");
+    }
+
     class Result {
         private final int localCheckpoint;
         private final int maxSeqNo;