[SPARK-18827] Cann't read broadcast if broadcast blocks are stored on-disk

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.1, 2.0.2, 2.1.0
• Fix Version/s: 2.0.3, 2.1.1
• Component/s: Spark Core
• Assignee: Yuming Wang
• Reporter: Yuming Wang
• Created: 12/Dec/16 05:18
• Updated: 23/Aug/17 03:58
• Resolved: 18/Dec/16 09:08

Description

How to reproduce it:

  test("Cache broadcast to disk") {
    val conf = new SparkConf()
      .setAppName("Cache broadcast to disk")
      .setMaster("local")
      .set("spark.memory.useLegacyMode", "true")
      .set("spark.storage.memoryFraction", "0.0")
    sc = new SparkContext(conf)
    val list = List[Int](1, 2, 3, 4)
    val broadcast = sc.broadcast(list)
    assert(broadcast.value.sum === 10)
  }

NoSuchElementException will throw since SPARK-17503 if a broadcast cannot cache in memory. The reason is that that change cannot cover !unrolled.hasNext in next() function.

Issue Links

• is duplicated by: SPARK-21794 exception about reading task serial data(broadcast) value when the storage memory is not enough to unroll
• links to: [Github] Pull Request #16252 (wangyum)