我正在将流式数据摄取到hbase中。我已经用kafka分区预先拆分了hbase表。使用复合rowkey和kafka分区的组合,timestamp后跟其他一些列使其唯一。使用这种方法,我在插入数据时有很好的吞吐量,但是我还必须每天聚合数据,这非常慢。
我们观察到 groupBy
或者 count
等于表所在的区域总数。
我做错什么了吗?如何限制hbase中表的区域数?
hbase create语句
create 'default:test', {NAME => 'data', VERSIONS => 1, TTL => '3888000'},{SPLITS=> ['10000000000000000000000000000000','20000000000000000000000000000000','30000000000000000000000000000000','40000000000000000000000000000000','50000000000000000000000000000000','60000000000000000000000000000000','70000000000000000000000000000000','80000000000000000000000000000000','90000000000000000000000000000000']}
插入时目录
def catalog = s"""{
|"table":{"namespace":"default", "name": "test", "tableCoder":"Phoenix"},
|"rowkey":"key",
|"columns":{
|"rowkey":{"cf":"rowkey", "col":"key", "type":"string"},
|"resource_id":{"cf":"data", "col":"resource_id", "type":"string"},
|"resource_name":{"cf":"data", "col":"resource_name", "type":"string"},
|"parent_id":{"cf":"data", "col":"parent_id", "type":"string"},
|"parent_name":{"cf":"data", "col":"parent_name", "type":"string"},
|"id":{"cf":"data", "col":"id", "type":"string"},
|"name":{"cf":"data", "col":"name", "type":"string"},
|"timestamp":{"cf":"data", "col":"timestamp", "type":"string"},
|"readable_timestamp":{"cf":"data", "col":"readable_timestamp",
"type":"string"},
|"value":{"cf":"data", "col":"value", "type":"string"},
|"partition":{"cf":"data", "col":"partition", "type":"string"}
|}
|}""".stripMargin
要阅读的目录
def catalog = s"""{
|"table":{"namespace":"default", "name": "test", "tableCoder":"Phoenix"},
|"rowkey":"partition:timestamp:id:parent_id:resource_id",
|"columns":{
|"partition":{"cf":"rowkey", "col":"partition", "type":"string"},
|"timestamp":{"cf":"rowkey", "col":"timestamp", "type":"string"},
|"id":{"cf":"rowkey", "col":"id", "type":"string"},
|"parent_id":{"cf":"rowkey", "col":"parent_id", "type":"string"},
|"resource_id":{"cf":"rowkey", "col":"resource_id", "type":"string"},
|"resource_name":{"cf":"data", "col":"resource_name", "type":"string"},
|"parent_name":{"cf":"data", "col":"parent_name", "type":"string"},
|"name":{"cf":"data", "col":"name", "type":"string"},
|"value":{"cf":"data", "col":"value", "type":"string"},
|"readable_timestamp":{"cf":"data", "col":"readable_timestamp", "type":"string"}
|}
|}""".stripMargin
要进行范围扫描,我将使用所有分区和时间范围:
val endrow = "1540080000000"
val startrow = "1539993600000"
df.filter(($"partition"==="0" && ($"timestamp" >= startrow && $"timestamp" <= endrow)) ||($"partition"==="1" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="2" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="3" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="4" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="5" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="6" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="7" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="8" && ($"timestamp" >= startrow && $"timestamp" <= endrow))||($"partition"==="9" && ($"timestamp" >= startrow && $"timestamp" <= endrow))).count
上面的过滤器将产生830个任务,这些任务等于区域数。花了太多时间。我怎样才能改进它?
暂无答案!
目前还没有任何答案,快来回答吧!