cassandra - DataStax 企业 : Spark Cassandra Batch Size

Question

我在我的 SparkConf 中设置参数 spark.cassandra.output.batch.size.rows 如下:

val conf = new SparkConf(true)
        .set("spark.cassandra.connection.host", "host")
        .set("spark.cassandra.auth.username", "cassandra")            
        .set("spark.cassandra.auth.password", "cassandra")
        .set("spark.cassandra.output.batch.size.rows", "5120")
        .set("spark.cassandra.output.concurrent.writes", "10")

但是当我表演的时候

saveToCassandra("data","ten_days")

我继续在 system.log 中看到警告

NFO [FlushWriter:7] 2014-11-20 11:11:16,498 Memtable.java (line 395) Completed flushing /var/lib/cassandra/data/system/hints/system-hints-jb-76-Data.db (5747287 bytes) for commitlog position ReplayPosition(segmentId=1416480663951, position=44882909)
 INFO [FlushWriter:7] 2014-11-20 11:11:16,499 Memtable.java (line 355) Writing Memtable-ten_days@1656582530(32979978/329799780 serialized/live bytes, 551793 ops)
 WARN [Native-Transport-Requests:761] 2014-11-20 11:11:16,499 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36825, exceeding specified threshold of 5120 by 31705.
 WARN [Native-Transport-Requests:777] 2014-11-20 11:11:16,500 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36813, exceeding specified threshold of 5120 by 31693.
 WARN [Native-Transport-Requests:822] 2014-11-20 11:11:16,501 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36823, exceeding specified threshold of 5120 by 31703.
 WARN [Native-Transport-Requests:835] 2014-11-20 11:11:16,500 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36817, exceeding specified threshold of 5120 by 31697.
 WARN [Native-Transport-Requests:781] 2014-11-20 11:11:16,501 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36817, exceeding specified threshold of 5120 by 31697.
 WARN [Native-Transport-Requests:755] 2014-11-20 11:11:16,501 BatchStatement.java (line 226) Batch of prepared statements for [data.ten_days] is of size 36822, exceeding specified threshold of 5120 by 31702.

我知道这只是警告，但我想了解为什么我的设置没有按预期工作。然后我可以在我的集群中看到很多提示。批量大小会影响集群中提示的数量吗？

谢谢

Answer 1

您已设置批量大小行而不是批量大小字节。这意味着连接器限制的是行数，而不是批处理的内存大小。

spark.cassandra.output.batch.size.rows: number of rows per singlebatch; default is 'auto' which means the connector will adjust thenumber of rows based on the amount of data in each row

spark.cassandra.output.batch.size.bytes: maximum total size of thebatch in bytes; defaults to 64 kB.

https://github.com/datastax/spark-cassandra-connector/blob/master/doc/5_saving.md

更重要的一点是，使用更大的批处理大小 (64kb) 并更改 cassandra.yaml 文件中的警告限制很可能会更好。

编辑:

最近我们发现，较大的批处理可能会导致某些 C* 配置不稳定，因此如果系统变得不稳定，请降低该值。