[SPARK-11740][STREAMING] Fix the race condition of two checkpoints in a batch

We will do checkpoint when generating a batch and completing a batch. When the processing time of a batch is greater than the batch interval, checkpointing for completing an old batch may run after checkpointing for generating a new batch. If this happens, checkpoint of an old batch actually has the latest information, so we want to recovery from it. This PR will use the latest checkpoint time as the file name, so that we can always recovery from the latest checkpoint file.

Author: Shixiong Zhu <shixiong@databricks.com>

Closes #9707 from zsxwing/fix-checkpoint.

1 files changed, 16 insertions, 2 deletions

diff --git a/streaming/src/main/scala/org/apache/spark/streaming/Checkpoint.scala b/streaming/src/main/scala/org/apache/spark/streaming/Checkpoint.scala
index 0cd55d9aec..fd0e8d5d69 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/Checkpoint.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/Checkpoint.scala
@@ -187,16 +187,30 @@ class CheckpointWriter(
   private var stopped = false
   private var fs_ : FileSystem = _
 
+  @volatile private var latestCheckpointTime: Time = null
+
   class CheckpointWriteHandler(
       checkpointTime: Time,
       bytes: Array[Byte],
       clearCheckpointDataLater: Boolean) extends Runnable {
     def run() {
+      if (latestCheckpointTime == null || latestCheckpointTime < checkpointTime) {
+        latestCheckpointTime = checkpointTime
+      }
       var attempts = 0
       val startTime = System.currentTimeMillis()
       val tempFile = new Path(checkpointDir, "temp")
-      val checkpointFile = Checkpoint.checkpointFile(checkpointDir, checkpointTime)
-      val backupFile = Checkpoint.checkpointBackupFile(checkpointDir, checkpointTime)
+      // We will do checkpoint when generating a batch and completing a batch. When the processing
+      // time of a batch is greater than the batch interval, checkpointing for completing an old
+      // batch may run after checkpointing of a new batch. If this happens, checkpoint of an old
+      // batch actually has the latest information, so we want to recovery from it. Therefore, we
+      // also use the latest checkpoint time as the file name, so that we can recovery from the
+      // latest checkpoint file.
+      //
+      // Note: there is only one thread writting the checkpoint files, so we don't need to worry
+      // about thread-safety.
+      val checkpointFile = Checkpoint.checkpointFile(checkpointDir, latestCheckpointTime)
+      val backupFile = Checkpoint.checkpointBackupFile(checkpointDir, latestCheckpointTime)
 
       while (attempts < MAX_ATTEMPTS && !stopped) {
         attempts += 1