More detailed logging for snapshot-based recovery (#93469)

The logic for planning a snapshot-based recovery is a little intricate and lacks much logging, making it hard to determine the reason why a snapshot-based recovery might not occur in production. This commit adds some extra logging showing details of the branches that ES takes.
elastic · Feb 6, 2023 · 1fb3a1b · 1fb3a1b
1 parent d3d1bbc
commit 1fb3a1b
Show file tree

Hide file tree

Showing 3 changed files with 48 additions and 3 deletions.
diff --git a/server/src/main/java/org/elasticsearch/indices/recovery/plan/ShardSnapshotsService.java b/server/src/main/java/org/elasticsearch/indices/recovery/plan/ShardSnapshotsService.java
@@ -75,7 +75,7 @@ public ShardSnapshotsService(
     }
 
     public void fetchLatestSnapshotsForShard(ShardId shardId, ActionListener<Optional<ShardSnapshot>> listener) {
-        assert shardId != null : "SharId was null but a value was expected";
+        assert shardId != null : "ShardId was null but a value was expected";
 
         final RepositoriesMetadata currentReposMetadata = clusterService.state()
             .metadata()
@@ -104,24 +104,30 @@ public void fetchLatestSnapshotsForShard(ShardId shardId, ActionListener<Optiona
         client.execute(
             GetShardSnapshotAction.INSTANCE,
             request,
-            new ThreadedActionListener<>(threadPool.generic(), listener.map(this::fetchSnapshotFiles))
+            new ThreadedActionListener<>(
+                threadPool.generic(),
+                listener.map(shardSnapshotResponse -> fetchSnapshotFiles(shardId, shardSnapshotResponse))
+            )
         );
     }
 
-    private Optional<ShardSnapshot> fetchSnapshotFiles(GetShardSnapshotResponse shardSnapshotResponse) {
+    private Optional<ShardSnapshot> fetchSnapshotFiles(ShardId shardId, GetShardSnapshotResponse shardSnapshotResponse) {
         assert ThreadPool.assertCurrentThreadPool(ThreadPool.Names.GENERIC);
 
         final Optional<ShardSnapshotInfo> latestShardSnapshotOpt = shardSnapshotResponse.getLatestShardSnapshot();
         if (latestShardSnapshotOpt.isEmpty()) {
+            logger.debug("{} no latest shard snapshot found", shardId);
             return Optional.empty();
         }
 
         final ShardSnapshotInfo latestShardSnapshot = latestShardSnapshotOpt.get();
         try {
             final Snapshot snapshot = latestShardSnapshot.getSnapshot();
+            logger.debug("{} considering recovery from [{}][{}]", shardId, snapshot.getRepository(), snapshot.getSnapshotId());
 
             final Repository repository = repositoriesService.repository(snapshot.getRepository());
             if (repository instanceof BlobStoreRepository == false) {
+                logger.debug("{} not recovering from snapshot in non-blobstore repository [{}]", shardId, snapshot.getRepository());
                 return Optional.empty();
             }
 

diff --git a/server/src/main/java/org/elasticsearch/repositories/IndexSnapshotsService.java b/server/src/main/java/org/elasticsearch/repositories/IndexSnapshotsService.java
@@ -8,6 +8,8 @@
 
 package org.elasticsearch.repositories;
 
+import org.apache.logging.log4j.LogManager;
+import org.apache.logging.log4j.Logger;
 import org.elasticsearch.action.ActionListener;
 import org.elasticsearch.action.StepListener;
 import org.elasticsearch.cluster.metadata.IndexMetadata;
@@ -28,6 +30,9 @@
 import java.util.Optional;
 
 public class IndexSnapshotsService {
+
+    private static final Logger logger = LogManager.getLogger(IndexSnapshotsService.class);
+
     private static final Comparator<Tuple<SnapshotId, RepositoryData.SnapshotDetails>> START_TIME_COMPARATOR = Comparator.<
         Tuple<SnapshotId, RepositoryData.SnapshotDetails>>comparingLong(pair -> pair.v2().getStartTimeMillis()).thenComparing(Tuple::v1);
 
@@ -62,6 +67,7 @@ public void getLatestSuccessfulSnapshotForShard(
 
         repositoryDataStepListener.whenComplete(repositoryData -> {
             if (repositoryData.hasIndex(indexName) == false) {
+                logger.debug("{} repository [{}] has no snapshots of this index", shardId, repositoryName);
                 listener.onResponse(Optional.empty());
                 return;
             }
@@ -81,11 +87,13 @@ public void getLatestSuccessfulSnapshotForShard(
                 // have the start/end date populated in RepositoryData. We could fetch all the backups and find out if there is
                 // a valid candidate, but for simplicity we just consider that we couldn't find any valid snapshot. Existing
                 // snapshots start/end timestamps should appear in the RepositoryData eventually.
+                logger.debug("{} could not determine latest snapshot of this shard in repository [{}]", shardId, repositoryName);
                 listener.onResponse(Optional.empty());
                 return;
             }
 
             final SnapshotId snapshotId = latestSnapshotId.get();
+            logger.debug("{} fetching details of [{}][{}]", shardId, repositoryName, snapshotId);
             repository.getSnapshotInfo(
                 snapshotId,
                 snapshotInfoStepListener.map(
@@ -100,6 +108,7 @@ public void getLatestSuccessfulSnapshotForShard(
 
             if (snapshotInfo == null || snapshotInfo.state() != SnapshotState.SUCCESS) {
                 // We couldn't find a valid candidate
+                logger.debug("{} failed to retrieve snapshot details from [{}]", shardId, repositoryName);
                 listener.onResponse(Optional.empty());
                 return;
             }

diff --git a/...icsearch/xpack/snapshotbasedrecoveries/recovery/plan/SnapshotsRecoveryPlannerService.java b/...icsearch/xpack/snapshotbasedrecoveries/recovery/plan/SnapshotsRecoveryPlannerService.java
@@ -11,8 +11,10 @@
 import org.apache.logging.log4j.Logger;
 import org.elasticsearch.Version;
 import org.elasticsearch.action.ActionListener;
+import org.elasticsearch.common.Strings;
 import org.elasticsearch.core.Nullable;
 import org.elasticsearch.index.shard.ShardId;
+import org.elasticsearch.index.snapshots.blobstore.BlobStoreIndexShardSnapshot;
 import org.elasticsearch.index.store.Store;
 import org.elasticsearch.index.store.StoreFileMetadata;
 import org.elasticsearch.indices.recovery.RecoverySettings;
@@ -68,6 +70,7 @@ public void computeRecoveryPlan(
             latestSnapshotOpt -> ActionListener.completeWith(
                 listener,
                 () -> computeRecoveryPlanWithSnapshots(
+                    shardId,
                     shardStateIdentifier,
                     sourceMetadata,
                     targetMetadata,
@@ -80,6 +83,7 @@ public void computeRecoveryPlan(
     }
 
     private ShardRecoveryPlan computeRecoveryPlanWithSnapshots(
+        ShardId shardId,
         @Nullable String shardStateIdentifier,
         Store.MetadataSnapshot sourceMetadata,
         Store.MetadataSnapshot targetMetadata,
@@ -91,6 +95,7 @@ private ShardRecoveryPlan computeRecoveryPlanWithSnapshots(
         List<StoreFileMetadata> filesMissingInTarget = concatLists(sourceTargetDiff.missing, sourceTargetDiff.different);
 
         if (latestSnapshotOpt.isEmpty()) {
+            logger.debug("{} no snapshot suitable for recovery found, falling back to recovery from the primary", shardId);
             // If we couldn't find any valid snapshots, fallback to the source
             return getRecoveryPlanUsingSourceNode(sourceMetadata, sourceTargetDiff, filesMissingInTarget, startingSeqNo, translogOps);
         }
@@ -102,6 +107,18 @@ private ShardRecoveryPlan computeRecoveryPlanWithSnapshots(
             && latestSnapshot.hasDifferentPhysicalFiles(sourceMetadata)
             && isSnapshotVersionCompatible(latestSnapshot)
             && sourceTargetDiff.identical.isEmpty()) {
+
+            logger.debug(
+                () -> Strings.format(
+                    "%s primary has changed since snapshot completed, but snapshot still looks ok for recovery: %s",
+                    shardId,
+                    latestSnapshot.getSnapshotFiles()
+                        .stream()
+                        .map(BlobStoreIndexShardSnapshot.FileInfo::physicalName)
+                        .collect(Collectors.joining(", ", "[", "]"))
+                )
+            );
+
             // Use the current primary as a fallback if the download fails half-way
             ShardRecoveryPlan fallbackPlan = getRecoveryPlanUsingSourceNode(
                 sourceMetadata,
@@ -141,6 +158,18 @@ && isSnapshotVersionCompatible(latestSnapshot)
             );
         }
 
+        logger.debug(
+            () -> Strings.format(
+                "%s attempting snapshot-based recovery of %s based on %s",
+                shardId,
+                snapshotFilesToRecover.getSnapshotFiles()
+                    .stream()
+                    .map(BlobStoreIndexShardSnapshot.FileInfo::physicalName)
+                    .collect(Collectors.joining(", ", "[", "]")),
+                snapshotDiff
+            )
+        );
+
         return new ShardRecoveryPlan(
             snapshotFilesToRecover,
             concatLists(snapshotDiff.missing, snapshotDiff.different),
@@ -183,6 +212,7 @@ private ShardRecoveryPlan getRecoveryPlanUsingSourceNode(
 
     private void fetchLatestSnapshotsIgnoringErrors(ShardId shardId, boolean useSnapshots, Consumer<Optional<ShardSnapshot>> listener) {
         if (useSnapshots == false) {
+            logger.debug("{} recovery will not attempt to use snapshots", shardId);
             listener.accept(Optional.empty());
             return;
         }