[SPARK-19659] Fetch big blocks to disk when shuffle-read

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.0
• Fix Version/s: 2.2.0
• Component/s: Shuffle, Spark Core
• Assignee: Jin Xing
• Reporter: Jin Xing
• Created: 19/Feb/17 10:04
• Updated: 17/May/20 18:30
• Resolved: 10/Jul/17 01:19

Description

Currently the whole block is fetched into memory(offheap by default) when shuffle-read. A block is defined by (shuffleId, mapId, reduceId). Thus it can be large when skew situations. If OOM happens during shuffle read, job will be killed and users will be notified to "Consider boosting spark.yarn.executor.memoryOverhead". Adjusting parameter and allocating more memory can resolve the OOM. However the approach is not perfectly suitable for production environment, especially for data warehouse.

Using Spark SQL as data engine in warehouse, users hope to have a unified parameter(e.g. memory) but less resource wasted(resource is allocated but not used),

It's not always easy to predict skew situations, when happen, it make sense to fetch remote blocks to disk for shuffle-read, rather than
kill the job because of OOM. This approach is mentioned during the discussion in SPARK-3019, by sandyr and mridulm80

Issue Links

• is blocked by: SPARK-19937 Collect metrics of block sizes when shuffle.
• is duplicated by: SPARK-6238 Support shuffle where individual blocks might be > 2G
• is duplicated by: SPARK-13510 Shuffle may throw FetchFailedException: Direct buffer memory
• is related to: SPARK-21253 Cannot fetch big blocks to disk
• is related to: SPARK-26590 make fetch-block-to-disk backward compatible
• links to: [Github] Pull Request #16989 (jinxing64)
• links to: [Github] Pull Request #18117 (cloud-fan)
• links to: [Github] Pull Request #18467 (zsxwing)