[Remote Store] Rate limiter for low priority uploads (opensearch-proj…

…ect#14374)

---------
Signed-off-by: Gaurav Bafna <gbbafna@amazon.com>

CHANGELOG.md

@@ -6,6 +6,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 ## [Unreleased 2.x]
 ### Added
 - Add fingerprint ingest processor ([#13724](https://github.com/opensearch-project/OpenSearch/pull/13724))
+- [Remote Store] Rate limiter for remote store low priority uploads ([#14374](https://github.com/opensearch-project/OpenSearch/pull/14374/))
 
 ### Dependencies
 - Bump `org.gradle.test-retry` from 1.5.8 to 1.5.9 ([#13442](https://github.com/opensearch-project/OpenSearch/pull/13442))

server/src/internalClusterTest/java/org/opensearch/action/admin/indices/create/RemoteCloneIndexIT.java

@@ -42,24 +42,32 @@
 import org.opensearch.Version;
 import org.opensearch.action.admin.cluster.health.ClusterHealthRequest;
 import org.opensearch.action.admin.cluster.health.ClusterHealthResponse;
+import org.opensearch.action.admin.cluster.repositories.get.GetRepositoriesRequest;
+import org.opensearch.action.admin.cluster.repositories.get.GetRepositoriesResponse;
 import org.opensearch.action.admin.indices.settings.get.GetSettingsResponse;
 import org.opensearch.action.admin.indices.shrink.ResizeType;
 import org.opensearch.action.admin.indices.stats.IndicesStatsResponse;
 import org.opensearch.client.Requests;
+import org.opensearch.cluster.metadata.RepositoryMetadata;
 import org.opensearch.cluster.routing.allocation.decider.EnableAllocationDecider;
 import org.opensearch.common.settings.Settings;
+import org.opensearch.common.unit.TimeValue;
+import org.opensearch.core.common.unit.ByteSizeValue;
 import org.opensearch.core.xcontent.MediaTypeRegistry;
 import org.opensearch.index.query.TermsQueryBuilder;
 import org.opensearch.indices.recovery.RecoverySettings;
 import org.opensearch.remotestore.RemoteStoreBaseIntegTestCase;
+import org.opensearch.repositories.RepositoriesService;
 import org.opensearch.test.OpenSearchIntegTestCase;
 import org.opensearch.test.VersionUtils;
+import org.junit.Before;
 
 import java.util.concurrent.ExecutionException;
 
 import static org.opensearch.test.hamcrest.OpenSearchAssertions.assertAcked;
 import static org.opensearch.test.hamcrest.OpenSearchAssertions.assertHitCount;
 import static org.hamcrest.Matchers.equalTo;
+import static org.hamcrest.Matchers.greaterThan;
 
 @OpenSearchIntegTestCase.ClusterScope(scope = OpenSearchIntegTestCase.Scope.TEST, numDataNodes = 0)
 public class RemoteCloneIndexIT extends RemoteStoreBaseIntegTestCase {
@@ -69,6 +77,11 @@ protected boolean forbidPrivateIndexSettings() {
         return false;
     }
 
+    @Before
+    public void setup() {
+        asyncUploadMockFsRepo = true;
+    }
+
     public void testCreateCloneIndex() {
         Version version = VersionUtils.randomIndexCompatibleVersion(random());
         int numPrimaryShards = randomIntBetween(1, 5);
@@ -140,6 +153,79 @@ public void testCreateCloneIndex() {
 
     }
 
+    public void testCreateCloneIndexLowPriorityRateLimit() {
+        Version version = VersionUtils.randomIndexCompatibleVersion(random());
+        int numPrimaryShards = 1;
+        prepareCreate("source").setSettings(
+            Settings.builder().put(indexSettings()).put("number_of_shards", numPrimaryShards).put("index.version.created", version)
+        ).get();
+        final int docs = randomIntBetween(0, 128);
+        for (int i = 0; i < docs; i++) {
+            client().prepareIndex("source").setSource("{\"foo\" : \"bar\", \"i\" : " + i + "}", MediaTypeRegistry.JSON).get();
+        }
+        ByteSizeValue shardSize = client().admin().indices().prepareStats("source").execute().actionGet().getShards()[0].getStats()
+            .getStore()
+            .size();
+        logger.info("Shard size is {}", shardSize);
+        internalCluster().ensureAtLeastNumDataNodes(2);
+        // ensure all shards are allocated otherwise the ensure green below might not succeed since we require the merge node
+        // if we change the setting too quickly we will end up with one replica unassigned which can't be assigned anymore due
+        // to the require._name below.
+        ensureGreen();
+        // relocate all shards to one node such that we can merge it.
+        client().admin().indices().prepareUpdateSettings("source").setSettings(Settings.builder().put("index.blocks.write", true)).get();
+        ensureGreen();
+
+        // disable rebalancing to be able to capture the right stats. balancing can move the target primary
+        // making it hard to pin point the source shards.
+        client().admin()
+            .cluster()
+            .prepareUpdateSettings()
+            .setTransientSettings(Settings.builder().put(EnableAllocationDecider.CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING.getKey(), "none"))
+            .get();
+        try {
+            // apply rate limiter
+            setLowPriorityUploadRate(REPOSITORY_NAME, "1kb");
+            assertAcked(
+                client().admin()
+                    .indices()
+                    .prepareResizeIndex("source", "target")
+                    .setResizeType(ResizeType.CLONE)
+                    .setSettings(Settings.builder().put("index.number_of_replicas", 0).putNull("index.blocks.write").build())
+                    .get()
+            );
+            ensureGreen();
+            long uploadPauseTime = 0L;
+            for (RepositoriesService repositoriesService : internalCluster().getDataNodeInstances(RepositoriesService.class)) {
+                uploadPauseTime += repositoriesService.repository(REPOSITORY_NAME).getLowPriorityRemoteUploadThrottleTimeInNanos();
+            }
+            assertThat(uploadPauseTime, greaterThan(TimeValue.timeValueSeconds(randomIntBetween(5, 10)).nanos()));
+        } catch (Exception e) {
+            throw new RuntimeException(e);
+        } finally {
+            // clean up
+            client().admin()
+                .cluster()
+                .prepareUpdateSettings()
+                .setTransientSettings(
+                    Settings.builder()
+                        .put(EnableAllocationDecider.CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING.getKey(), (String) null)
+                        .put(RecoverySettings.INDICES_INTERNAL_REMOTE_UPLOAD_TIMEOUT.getKey(), (String) null)
+                )
+                .get();
+        }
+    }
+
+    protected void setLowPriorityUploadRate(String repoName, String value) throws ExecutionException, InterruptedException {
+        GetRepositoriesRequest gr = new GetRepositoriesRequest(new String[] { repoName });
+        GetRepositoriesResponse res = client().admin().cluster().getRepositories(gr).get();
+        RepositoryMetadata rmd = res.repositories().get(0);
+        Settings.Builder settings = Settings.builder()
+            .put("location", rmd.settings().get("location"))
+            .put("max_remote_low_priority_upload_bytes_per_sec", value);
+        assertAcked(client().admin().cluster().preparePutRepository(repoName).setType(rmd.type()).setSettings(settings).get());
+    }
+
     public void testCreateCloneIndexFailure() throws ExecutionException, InterruptedException {
         asyncUploadMockFsRepo = false;
         Version version = VersionUtils.randomIndexCompatibleVersion(random());

server/src/internalClusterTest/java/org/opensearch/remotestore/translogmetadata/mocks/MockFsMetadataSupportedRepository.java

@@ -16,9 +16,9 @@
 import org.opensearch.core.xcontent.NamedXContentRegistry;
 import org.opensearch.env.Environment;
 import org.opensearch.indices.recovery.RecoverySettings;
-import org.opensearch.repositories.fs.FsRepository;
+import org.opensearch.repositories.fs.ReloadableFsRepository;
 
-public class MockFsMetadataSupportedRepository extends FsRepository {
+public class MockFsMetadataSupportedRepository extends ReloadableFsRepository {
 
     public static Setting<Boolean> TRIGGER_DATA_INTEGRITY_FAILURE = Setting.boolSetting(
         "mock_fs_repository.trigger_data_integrity_failure",

server/src/main/java/org/opensearch/index/store/RemoteDirectory.java

@@ -64,6 +64,8 @@ public class RemoteDirectory extends Directory {
 
     private final UnaryOperator<OffsetRangeInputStream> uploadRateLimiter;
 
+    private final UnaryOperator<OffsetRangeInputStream> lowPriorityUploadRateLimiter;
+
     private final UnaryOperator<InputStream> downloadRateLimiter;
 
     /**
@@ -76,15 +78,17 @@ public BlobContainer getBlobContainer() {
     }
 
     public RemoteDirectory(BlobContainer blobContainer) {
-        this(blobContainer, UnaryOperator.identity(), UnaryOperator.identity());
+        this(blobContainer, UnaryOperator.identity(), UnaryOperator.identity(), UnaryOperator.identity());
     }
 
     public RemoteDirectory(
         BlobContainer blobContainer,
         UnaryOperator<OffsetRangeInputStream> uploadRateLimiter,
+        UnaryOperator<OffsetRangeInputStream> lowPriorityUploadRateLimiter,
         UnaryOperator<InputStream> downloadRateLimiter
     ) {
         this.blobContainer = blobContainer;
+        this.lowPriorityUploadRateLimiter = lowPriorityUploadRateLimiter;
         this.uploadRateLimiter = uploadRateLimiter;
         this.downloadRateLimiter = downloadRateLimiter;
     }
@@ -357,13 +361,23 @@ private void uploadBlob(
             remoteIntegrityEnabled = ((AsyncMultiStreamBlobContainer) getBlobContainer()).remoteIntegrityCheckSupported();
         }
         lowPriorityUpload = lowPriorityUpload || contentLength > ByteSizeUnit.GB.toBytes(15);
+        RemoteTransferContainer.OffsetRangeInputStreamSupplier offsetRangeInputStreamSupplier;
+        if (lowPriorityUpload) {
+            offsetRangeInputStreamSupplier = (size, position) -> lowPriorityUploadRateLimiter.apply(
+                new OffsetRangeIndexInputStream(from.openInput(src, ioContext), size, position)
+            );
+        } else {
+            offsetRangeInputStreamSupplier = (size, position) -> uploadRateLimiter.apply(
+                new OffsetRangeIndexInputStream(from.openInput(src, ioContext), size, position)
+            );
+        }
         RemoteTransferContainer remoteTransferContainer = new RemoteTransferContainer(
             src,
             remoteFileName,
             contentLength,
             true,
             lowPriorityUpload ? WritePriority.LOW : WritePriority.NORMAL,
-            (size, position) -> uploadRateLimiter.apply(new OffsetRangeIndexInputStream(from.openInput(src, ioContext), size, position)),
+            offsetRangeInputStreamSupplier,
             expectedChecksum,
             remoteIntegrityEnabled
         );

server/src/main/java/org/opensearch/index/store/RemoteSegmentStoreDirectoryFactory.java

@@ -77,6 +77,7 @@ public Directory newDirectory(String repositoryName, String indexUUID, ShardId s
             RemoteDirectory dataDirectory = new RemoteDirectory(
                 blobStoreRepository.blobStore().blobContainer(dataPath),
                 blobStoreRepository::maybeRateLimitRemoteUploadTransfers,
+                blobStoreRepository::maybeRateLimitLowPriorityRemoteUploadTransfers,
                 blobStoreRepository::maybeRateLimitRemoteDownloadTransfers
             );
 

server/src/main/java/org/opensearch/repositories/FilterRepository.java

@@ -142,6 +142,11 @@ public long getRemoteUploadThrottleTimeInNanos() {
         return in.getRemoteUploadThrottleTimeInNanos();
     }
 
+    @Override
+    public long getLowPriorityRemoteUploadThrottleTimeInNanos() {
+        return in.getRemoteUploadThrottleTimeInNanos();
+    }
+
     @Override
     public long getRemoteDownloadThrottleTimeInNanos() {
         return in.getRemoteDownloadThrottleTimeInNanos();

server/src/main/java/org/opensearch/repositories/Repository.java

@@ -207,10 +207,17 @@ default void deleteSnapshotsAndReleaseLockFiles(
     long getRestoreThrottleTimeInNanos();
 
     /**
-     * Returns restore throttle time in nanoseconds
+     * Returns upload throttle time in nanoseconds
      */
     long getRemoteUploadThrottleTimeInNanos();
 
+    /**
+     * Returns low priority upload throttle time in nanoseconds
+     */
+    default long getLowPriorityRemoteUploadThrottleTimeInNanos() {
+        return 0;
+    }
+
     /**
      * Returns restore throttle time in nanoseconds
      */

server/src/main/java/org/opensearch/repositories/blobstore/BlobStoreRepository.java

@@ -316,6 +316,8 @@ public abstract class BlobStoreRepository extends AbstractLifecycleComponent imp
 
     private volatile RateLimiter remoteUploadRateLimiter;
 
+    private volatile RateLimiter remoteUploadLowPriorityRateLimiter;
+
     private volatile RateLimiter remoteDownloadRateLimiter;
 
     private final CounterMetric snapshotRateLimitingTimeInNanos = new CounterMetric();
@@ -326,6 +328,8 @@ public abstract class BlobStoreRepository extends AbstractLifecycleComponent imp
 
     private final CounterMetric remoteUploadRateLimitingTimeInNanos = new CounterMetric();
 
+    private final CounterMetric remoteUploadLowPriorityRateLimitingTimeInNanos = new CounterMetric();
+
     public static final ChecksumBlobStoreFormat<Metadata> GLOBAL_METADATA_FORMAT = new ChecksumBlobStoreFormat<>(
         "metadata",
         METADATA_NAME_FORMAT,
@@ -445,6 +449,11 @@ private void readRepositoryMetadata(RepositoryMetadata repositoryMetadata) {
         snapshotRateLimiter = getRateLimiter(metadata.settings(), "max_snapshot_bytes_per_sec", new ByteSizeValue(40, ByteSizeUnit.MB));
         restoreRateLimiter = getRateLimiter(metadata.settings(), "max_restore_bytes_per_sec", ByteSizeValue.ZERO);
         remoteUploadRateLimiter = getRateLimiter(metadata.settings(), "max_remote_upload_bytes_per_sec", ByteSizeValue.ZERO);
+        remoteUploadLowPriorityRateLimiter = getRateLimiter(
+            metadata.settings(),
+            "max_remote_low_priority_upload_bytes_per_sec",
+            ByteSizeValue.ZERO
+        );
         remoteDownloadRateLimiter = getRateLimiter(metadata.settings(), "max_remote_download_bytes_per_sec", ByteSizeValue.ZERO);
         readOnly = READONLY_SETTING.get(metadata.settings());
         cacheRepositoryData = CACHE_REPOSITORY_DATA.get(metadata.settings());
@@ -1882,6 +1891,11 @@ public long getRemoteUploadThrottleTimeInNanos() {
         return remoteUploadRateLimitingTimeInNanos.count();
     }
 
+    @Override
+    public long getLowPriorityRemoteUploadThrottleTimeInNanos() {
+        return remoteUploadLowPriorityRateLimitingTimeInNanos.count();
+    }
+
     @Override
     public long getRemoteDownloadThrottleTimeInNanos() {
         return remoteDownloadRateLimitingTimeInNanos.count();
@@ -3177,6 +3191,20 @@ public OffsetRangeInputStream maybeRateLimitRemoteUploadTransfers(OffsetRangeInp
         );
     }
 
+    public OffsetRangeInputStream maybeRateLimitLowPriorityRemoteUploadTransfers(OffsetRangeInputStream offsetRangeInputStream) {
+        return maybeRateLimitRemoteTransfers(
+            maybeRateLimitRemoteTransfers(
+                offsetRangeInputStream,
+                () -> remoteUploadRateLimiter,
+                remoteUploadRateLimitingTimeInNanos,
+                BlobStoreTransferContext.REMOTE_UPLOAD
+            ),
+            () -> remoteUploadLowPriorityRateLimiter,
+            remoteUploadLowPriorityRateLimitingTimeInNanos,
+            BlobStoreTransferContext.REMOTE_UPLOAD
+        );
+    }
+
     public InputStream maybeRateLimitRemoteDownloadTransfers(InputStream inputStream) {
         return maybeRateLimit(
             maybeRateLimit(