[SPARK-30900][SS] FileStreamSource: Avoid reading compact metadata log twice if the query restarts from compact batch

sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSource.scala

@@ -104,7 +104,7 @@ class FileStreamSource(
   // Visible for testing and debugging in production.
   val seenFiles = new SeenFilesMap(maxFileAgeMs, fileNameOnly)
 
-  metadataLog.allFiles().foreach { entry =>
+  metadataLog.restore().foreach { entry =>
     seenFiles.add(entry.path, entry.timestamp)
   }
   seenFiles.purge()

sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSourceLog.scala

@@ -36,6 +36,7 @@ class FileStreamSourceLog(
   extends CompactibleFileStreamLog[FileEntry](metadataLogVersion, sparkSession, path) {
 
   import CompactibleFileStreamLog._
+  import FileStreamSourceLog._
 
   // Configurations about metadata compaction
   protected override val defaultCompactInterval: Int =
@@ -122,8 +123,35 @@ class FileStreamSourceLog(
     }
     batches
   }
+
+  def restore(): Array[FileEntry] = {
+    val files = allFiles()
+
+    // When restarting the query, there is a case which the query starts from compaction batch,
+    // and the batch has source metadata file to read. One case is that the previous query
+    // succeeded to read from inputs, but not finalized the batch for various reasons.
+    // The below code finds the latest compaction batch, and put entries for the batch into the
+    // file entry cache which would avoid reading compact batch file twice.
+    // It doesn't know about offset / commit metadata in checkpoint so doesn't know which exactly
+    // batch to start from, but in practice, only couple of latest batches are candidates to
+    // be started. We leverage the fact to skip calculation if possible.
+    files.lastOption.foreach { lastEntry =>
+      val latestBatchId = lastEntry.batchId
+      val latestCompactedBatchId = getAllValidBatches(latestBatchId, compactInterval)(0)
+      if (latestCompactedBatchId > 0 &&
+          (latestBatchId - latestCompactedBatchId) < PREV_NUM_BATCHES_TO_READ_IN_RESTORE) {
+        val logsForLatestCompactedBatch = files.filter { entry =>
+          entry.batchId == latestCompactedBatchId
+        }
+        fileEntryCache.put(latestCompactedBatchId, logsForLatestCompactedBatch)
+      }
+    }
+
+    files
+  }
 }
 
 object FileStreamSourceLog {
   val VERSION = 1
+  val PREV_NUM_BATCHES_TO_READ_IN_RESTORE = 2
 }

sql/core/src/test/scala/org/apache/spark/sql/streaming/FileStreamSourceSuite.scala

@@ -1371,6 +1371,60 @@ class FileStreamSourceSuite extends FileStreamSourceTest {
     }
   }
 
+  test("restore from file stream source log") {
+    def createEntries(batchId: Long, count: Int): Array[FileEntry] = {
+      (1 to count).map { idx =>
+        FileEntry(s"path_${batchId}_$idx", 10000 * batchId + count, batchId)
+      }.toArray
+    }
+
+    withSQLConf(SQLConf.FILE_SOURCE_LOG_COMPACT_INTERVAL.key -> "5") {
+      withTempDir { chk =>
+        val _fileEntryCache = PrivateMethod[java.util.LinkedHashMap[Long, Array[FileEntry]]](
+          Symbol("fileEntryCache"))
+
+        val metadata = new FileStreamSourceLog(FileStreamSourceLog.VERSION, spark,
+          chk.getCanonicalPath)
+        val fileEntryCache = metadata invokePrivate _fileEntryCache()
+
+        (0 to 4).foreach { batchId =>
+          metadata.add(batchId, createEntries(batchId, 100))
+        }
+        val allFiles = metadata.allFiles()
+
+        // batch 4 is a compact batch which logs would be cached in fileEntryCache
+        fileEntryCache.containsKey(4)
+
+        val metadata2 = new FileStreamSourceLog(FileStreamSourceLog.VERSION, spark,
+          chk.getCanonicalPath)
+        val fileEntryCache2 = metadata2 invokePrivate _fileEntryCache()
+
+        // allFiles() doesn't restore the logs for the latest compact batch into file entry cache
+        assert(metadata2.allFiles() === allFiles)
+        assert(!fileEntryCache2.containsKey(4L))
+
+        // restore() will restore the logs for the latest compact batch into file entry cache
+        assert(metadata2.restore() === allFiles)
+        assert(fileEntryCache2.containsKey(4L))
+
+        (5 to 5 + FileStreamSourceLog.PREV_NUM_BATCHES_TO_READ_IN_RESTORE).foreach { batchId =>
+          metadata2.add(batchId, createEntries(batchId, 100))
+        }
+        val allFiles2 = metadata2.allFiles()
+
+        val metadata3 = new FileStreamSourceLog(FileStreamSourceLog.VERSION, spark,
+          chk.getCanonicalPath)
+        val fileEntryCache3 = metadata3 invokePrivate _fileEntryCache()
+
+        // restore() will not restore the logs for the latest compact batch into file entry cache
+        // if the latest batch is too far from latest compact batch, because it's unlikely Spark
+        // will request the batch for the start point.
+        assert(metadata2.restore() === allFiles2)
+        assert(!fileEntryCache3.containsKey(4L))
+      }
+    }
+  }
+
   test("get arbitrary batch from FileStreamSource") {
     withTempDirs { case (src, tmp) =>
       withSQLConf(