[CARBONDATA-1655] getSplits function is very slow !!!

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: data-query
• Assignee: Unassigned
• Reporter: cen yuhai
• Created: 28/Oct/17 15:51
• Updated: 28/Oct/17 16:15

Description

I have a table which has 4 billion records, I find that the getSplits function is too slow!
getSplit spent 20s!!!

"main" #1 prio=5 os_prio=0 tid=0x00007fcc94013000 nid=0x5ed5 runnable [0x00007fcc992b6000]
   java.lang.Thread.State: RUNNABLE
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getSizeInBytes(UnsafeDataMapRow.java:155)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getPosition(UnsafeDataMapRow.java:170)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getLengthInBytes(UnsafeDataMapRow.java:61)
        at org.apache.carbondata.core.indexstore.row.DataMapRow.getSizeInBytes(DataMapRow.java:80)
        at org.apache.carbondata.core.indexstore.row.DataMapRow.getTotalSizeInBytes(DataMapRow.java:70)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getSizeInBytes(UnsafeDataMapRow.java:161)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getPosition(UnsafeDataMapRow.java:170)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getRow(UnsafeDataMapRow.java:89)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getSizeInBytes(UnsafeDataMapRow.java:161)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getPosition(UnsafeDataMapRow.java:170)
        at org.apache.carbondata.core.indexstore.row.UnsafeDataMapRow.getByteArray(UnsafeDataMapRow.java:43)
        at org.apache.carbondata.core.indexstore.blockletindex.BlockletDataMap.createBlocklet(BlockletDataMap.java:310)
        at org.apache.carbondata.core.indexstore.blockletindex.BlockletDataMap.prune(BlockletDataMap.java:268)
        at org.apache.carbondata.core.datamap.TableDataMap.prune(TableDataMap.java:66)
        at org.apache.carbondata.hadoop.api.CarbonTableInputFormat.getDataBlocksOfSegment(CarbonTableInputFormat.java:524)
        at org.apache.carbondata.hadoop.api.CarbonTableInputFormat.getSplits(CarbonTableInputFormat.java:453)
        at org.apache.carbondata.hadoop.api.CarbonTableInputFormat.getSplits(CarbonTableInputFormat.java:324)
        at org.apache.carbondata.spark.rdd.CarbonScanRDD.getPartitions(CarbonScanRDD.scala:84)
        at org.apache.spark.rdd.RDD$$anonfun$partitions$2.apply(RDD.scala:260)
        at org.apache.spark.rdd.RDD$$anonfun$partitions$2.apply(RDD.scala:258)
        at scala.Option.getOrElse(Option.scala:121)
        at org.apache.spark.rdd.RDD.partitions(RDD.scala:258)
        at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:35)
        at org.apache.spark.rdd.RDD$$anonfun$partitions$2.apply(RDD.scala:260)
        at org.apache.spark.rdd.RDD$$anonfun$partitions$2.apply(RDD.scala:258)
        at scala.Option.getOrElse(Option.scala:121)
```

spark-sql> select dt from dm_test.table_carbondata limit 1;
NULL
Time taken: 20.94 seconds, Fetched 1 row(s)

If the query don't contains sort column, prune should return quickly