spark技术分享BlockStoreShuffleReader
BlockStoreShuffleReader is the one and only known ShuffleReader that reads the combined key-values for the reduce task (for a range of start and end reduce partitions) from a shuffle by requesting them from block managers.
BlockStoreShuffleReader is created exclusively when SortShuffleManager is requested for the ShuffleReader for a range of reduce partitions.
Reading Combined Key-Value Records For Reduce Task (using ShuffleBlockFetcherIterator) — read Method
|
1 2 3 4 5 |
read(): Iterator[Product2[K, C]] |
|
Note
|
read is part of ShuffleReader Contract.
|
Internally, read first creates a ShuffleBlockFetcherIterator (passing in the values of spark.reducer.maxSizeInFlight, spark.reducer.maxReqsInFlight and spark.shuffle.detectCorrupt Spark properties).
|
Note
|
read uses BlockManager to access ShuffleClient to create ShuffleBlockFetcherIterator.
|
|
Note
|
read uses MapOutputTracker to find the BlockManagers with the shuffle blocks and sizes to create ShuffleBlockFetcherIterator.
|
read creates a new SerializerInstance (using Serializer from ShuffleDependency).
read creates a key/value iterator by deserializeStream every shuffle block stream.
read updates the context task metrics for each record read.
|
Note
|
read uses CompletionIterator (to count the records read) and InterruptibleIterator (to support task cancellation).
|
If the ShuffleDependency has an Aggregator defined, read wraps the current iterator inside an iterator defined by Aggregator.combineCombinersByKey (for mapSideCombine enabled) or Aggregator.combineValuesByKey otherwise.
|
Note
|
run reports an exception when ShuffleDependency has no Aggregator defined with mapSideCombine flag enabled.
|
For keyOrdering defined in ShuffleDependency, run does the following:
-
Inserts all the records into the
ExternalSorter -
Updates context
TaskMetrics -
Returns a
CompletionIteratorfor theExternalSorter
Settings
| Spark Property | Default Value | Description |
|---|---|---|
|
|
Maximum size (in bytes) of map outputs to fetch simultaneously from each reduce task. Since each output requires a new buffer to receive it, this represents a fixed memory overhead per reduce task, so keep it small unless you have a large amount of memory. Used when |
|
|
(unlimited) |
The maximum number of remote requests to fetch blocks at any given point. When the number of hosts in the cluster increases, it might lead to very large number of in-bound connections to one or more nodes, causing the workers to fail under load. By allowing it to limit the number of fetch requests, this scenario can be mitigated. Used when |
|
|
|
Controls whether to detect any corruption in fetched blocks. Used when |
Creating BlockStoreShuffleReader Instance
BlockStoreShuffleReader takes the following when created:
BlockStoreShuffleReader initializes the internal registries and counters.
|
Note
|
BlockStoreShuffleReader uses SparkEnv to access the SerializerManager, BlockManager and MapOutputTracker.
|
