[FLINK-39900] [forst] Add configuration option for setting the number of ForSt transfer threads

docs/layouts/shortcodes/generated/expert_forst_section.html

@@ -20,6 +20,12 @@
             <td>Integer</td>
             <td>When the number of eviction that a block in hot link is moved to cold link reaches this value, the block will be blocked from being promoted to the head of the LRU list. The default value is '3'.</td>
         </tr>
+        <tr>
+            <td><h5>state.backend.forst.checkpoint.transfer.thread.num</h5></td>
+            <td style="word-wrap: break-word;">4</td>
+            <td>Integer</td>
+            <td>The number of threads used to transfer files during checkpoint (writing or copying files to the checkpoint storage) and restore (transferring state files back to the ForSt working directory). Consider increasing this value when snapshotting or restoring large state. Note that the pool is created per state backend instance. Setting it too high can saturate network bandwidth or trigger rate limiting depending on the remote storage; setting it too low can lead to long checkpoint durations or timeouts for large state. The default value is '4'.</td>
+        </tr>
         <tr>
             <td><h5>state.backend.forst.executor.inline-coordinator</h5></td>
             <td style="word-wrap: break-word;">false</td>

docs/layouts/shortcodes/generated/forst_configuration.html

@@ -38,6 +38,12 @@
             <td>MemorySize</td>
             <td>An upper-bound of the size that can be used for cache. User should specify at least one cache size limit to enable the cache, either this option or the 'state.backend.forst.cache.reserve-size' option. They can be set simultaneously, and in this case, cache will grow if meet the requirements of both two options. The default value is '0 bytes', meaning that this option is disabled. </td>
         </tr>
+        <tr>
+            <td><h5>state.backend.forst.checkpoint.transfer.thread.num</h5></td>
+            <td style="word-wrap: break-word;">4</td>
+            <td>Integer</td>
+            <td>The number of threads used to transfer files during checkpoint (writing or copying files to the checkpoint storage) and restore (transferring state files back to the ForSt working directory). Consider increasing this value when snapshotting or restoring large state. Note that the pool is created per state backend instance. Setting it too high can saturate network bandwidth or trigger rate limiting depending on the remote storage; setting it too low can lead to long checkpoint durations or timeouts for large state. The default value is '4'.</td>
+        </tr>
         <tr>
             <td><h5>state.backend.forst.executor.inline-coordinator</h5></td>
             <td style="word-wrap: break-word;">false</td>

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/ForStConfigurableOptions.java

@@ -43,6 +43,7 @@
 import static org.apache.flink.state.forst.ForStOptions.CACHE_DIRECTORY;
 import static org.apache.flink.state.forst.ForStOptions.CACHE_RESERVED_SIZE;
 import static org.apache.flink.state.forst.ForStOptions.CACHE_SIZE_BASE_LIMIT;
+import static org.apache.flink.state.forst.ForStOptions.CHECKPOINT_TRANSFER_THREAD_NUM;
 import static org.apache.flink.state.forst.ForStOptions.EXECUTOR_COORDINATOR_INLINE;
 import static org.apache.flink.state.forst.ForStOptions.EXECUTOR_READ_IO_PARALLELISM;
 import static org.apache.flink.state.forst.ForStOptions.EXECUTOR_WRITE_IO_INLINE;
@@ -379,6 +380,7 @@ public class ForStConfigurableOptions implements Serializable {
                 LOG_MAX_FILE_SIZE,
                 LOG_FILE_NUM,
                 LOG_DIR,
+                CHECKPOINT_TRANSFER_THREAD_NUM,
 
                 // configurable ColumnFamilyOptions
                 COMPACTION_STYLE,
@@ -408,7 +410,8 @@ public class ForStConfigurableOptions implements Serializable {
                             MAX_BACKGROUND_THREADS,
                             LOG_FILE_NUM,
                             MAX_WRITE_BUFFER_NUMBER,
-                            MIN_WRITE_BUFFER_NUMBER_TO_MERGE));
+                            MIN_WRITE_BUFFER_NUMBER_TO_MERGE,
+                            CHECKPOINT_TRANSFER_THREAD_NUM));
 
     private static final Set<ConfigOption<?>> SIZE_CONFIG_SET =
             new HashSet<>(

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/ForStKeyedStateBackendBuilder.java

@@ -442,7 +442,7 @@ private ForStRestoreOperation getForStRestoreOperation(
             long lastCompletedCheckpointId) {
         ForStStateDataTransfer stateTransfer =
                 new ForStStateDataTransfer(
-                        ForStStateDataTransfer.DEFAULT_THREAD_NUM,
+                        optionsContainer.getDataTransferThreadNum(),
                         optionsContainer.getFileSystem());
 
         if (enableIncrementalCheckpointing) {

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/ForStOptions.java

@@ -300,4 +300,19 @@ public class ForStOptions {
                                     + " Only valid when '"
                                     + EXECUTOR_WRITE_IO_INLINE.key()
                                     + "' is false.");
+
+    @Documentation.Section(Documentation.Sections.EXPERT_FORST)
+    public static final ConfigOption<Integer> CHECKPOINT_TRANSFER_THREAD_NUM =
+            ConfigOptions.key("state.backend.forst.checkpoint.transfer.thread.num")
+                    .intType()
+                    .defaultValue(4)
+                    .withDescription(
+                            "The number of threads used to transfer files during checkpoint (writing or copying "
+                                    + "files to the checkpoint storage) and restore (transferring state files back to "
+                                    + "the ForSt working directory). Consider increasing this value when snapshotting or "
+                                    + "restoring large state. Note that the pool is created per state backend instance. "
+                                    + "Setting it too high can saturate network bandwidth or trigger rate limiting depending "
+                                    + "on the remote storage; setting it too low can lead to long checkpoint durations "
+                                    + "or timeouts for large state. "
+                                    + "The default value is '4'.");
 }

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/ForStResourceContainer.java

@@ -324,6 +324,10 @@ public int getWriteIoParallelism() {
         return configuration.get(ForStOptions.EXECUTOR_WRITE_IO_PARALLELISM);
     }
 
+    public int getDataTransferThreadNum() {
+        return configuration.get(ForStOptions.CHECKPOINT_TRANSFER_THREAD_NUM);
+    }
+
     /**
      * Prepare local and remote directories.
      *

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/datatransfer/ForStStateDataTransfer.java

@@ -64,9 +64,6 @@
 public class ForStStateDataTransfer implements Closeable {
     private static final Logger LOG = LoggerFactory.getLogger(ForStStateDataTransfer.class);
 
-    // TODO: Add ConfigOption replace this field after ForSt checkpoint implementation stable
-    public static final int DEFAULT_THREAD_NUM = 4;
-
     protected final ExecutorService executorService;
 
     @Nullable private final ForStFlinkFileSystem forStFs;

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/restore/ForStIncrementalRestoreOperation.java

@@ -265,7 +265,7 @@ public ForStRestoreResult restore() throws Exception {
     private void transferAllStateHandles(List<StateHandleTransferSpec> specs) throws Exception {
         try (ForStStateDataTransfer transfer =
                 new ForStStateDataTransfer(
-                        ForStStateDataTransfer.DEFAULT_THREAD_NUM,
+                        optionsContainer.getDataTransferThreadNum(),
                         optionsContainer.getFileSystem())) {
             transfer.transferAllStateDataToDirectory(
                     optionsContainer.getPathContainer(),

flink-state-backends/flink-statebackend-forst/src/main/java/org/apache/flink/state/forst/sync/ForStSyncKeyedStateBackendBuilder.java

@@ -465,7 +465,7 @@ public ForStSyncKeyedStateBackendBuilder<K> setRecoveryClaimMode(
 
         ForStStateDataTransfer stateTransfer =
                 new ForStStateDataTransfer(
-                        ForStStateDataTransfer.DEFAULT_THREAD_NUM,
+                        optionsContainer.getDataTransferThreadNum(),
                         optionsContainer.getFileSystem());
 
         if (enableIncrementalCheckpointing) {

flink-state-backends/flink-statebackend-forst/src/test/java/org/apache/flink/state/forst/ForStStateBackendConfigTest.java

@@ -815,6 +815,19 @@ void testConfigureQueryTimeAfterNumEntries() throws Exception {
         }
     }
 
+    @Test
+    public void testConfigureCheckpointTransferThreadNumber() throws Exception {
+        ForStStateBackend forStStateBackend = new ForStStateBackend();
+        Configuration configuration = new Configuration();
+        configuration.setString(ForStOptions.CHECKPOINT_TRANSFER_THREAD_NUM.key(), "10");
+        forStStateBackend = forStStateBackend.configure(configuration, getClass().getClassLoader());
+
+        try (ForStResourceContainer resourceContainer =
+                forStStateBackend.createOptionsAndResourceContainer(null)) {
+            assertThat(resourceContainer.getDataTransferThreadNum()).isEqualTo(10);
+        }
+    }
+
     private void verifySetParameter(Runnable setter) {
         assertThatThrownBy(setter::run).isInstanceOf(IllegalArgumentException.class);
     }

flink-state-backends/flink-statebackend-forst/src/test/java/org/apache/flink/state/forst/snapshot/ForStIncrementalSnapshotStrategyTest.java

@@ -33,6 +33,7 @@
 import org.apache.flink.runtime.state.v2.RegisteredKeyValueStateBackendMetaInfo;
 import org.apache.flink.state.forst.ForStExtension;
 import org.apache.flink.state.forst.ForStOperationUtils;
+import org.apache.flink.state.forst.ForStOptions;
 import org.apache.flink.state.forst.datatransfer.ForStStateDataTransfer;
 import org.apache.flink.testutils.junit.utils.TempDirUtils;
 
@@ -199,7 +200,8 @@ private ForStIncrementalSnapshotStrategy<?> createSnapshotStrategy()
                 CompositeKeySerializationUtils.computeRequiredBytesInKeyGroupPrefix(2),
                 UUID.randomUUID(),
                 new TreeMap<>(),
-                new ForStStateDataTransfer(ForStStateDataTransfer.DEFAULT_THREAD_NUM),
+                new ForStStateDataTransfer(
+                        ForStOptions.CHECKPOINT_TRANSFER_THREAD_NUM.defaultValue()),
                 -1);
     }
 
@@ -231,7 +233,8 @@ private ForStNativeFullSnapshotStrategy<?> createFullSnapshotStrategy()
                 new KeyGroupRange(0, 1),
                 CompositeKeySerializationUtils.computeRequiredBytesInKeyGroupPrefix(2),
                 UUID.randomUUID(),
-                new ForStStateDataTransfer(ForStStateDataTransfer.DEFAULT_THREAD_NUM));
+                new ForStStateDataTransfer(
+                        ForStOptions.CHECKPOINT_TRANSFER_THREAD_NUM.defaultValue()));
     }
 
     private FsCheckpointStreamFactory createFsCheckpointStreamFactory() throws IOException {