In this simple test, why does Athena fail to prune partitions?

0

I have defined two tables:

CREATE EXTERNAL TABLE `event_data`(
  `systemid` string COMMENT 'from deserializer', 
  `eventtime` string COMMENT 'from deserializer', 
  `eventtype` string COMMENT 'from deserializer', 
  `source` string COMMENT 'from deserializer', 
  `updtdate` string COMMENT 'from deserializer', 
  `rawdata` string COMMENT 'from deserializer', 
  `media` string COMMENT 'from deserializer')
PARTITIONED BY ( 
  `partition_day` string)
ROW FORMAT SERDE 
  'org.openx.data.jsonserde.JsonSerDe' 
STORED AS INPUTFORMAT 
  'org.apache.hadoop.mapred.TextInputFormat' 
OUTPUTFORMAT 
  'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION
  's3://test-bucket/incoming/data/'
TBLPROPERTIES (
  'classification'='json', 
  'projection.enabled'='true', 
  'projection.partition_day.format'='yyyy-MM-dd', 
  'projection.partition_day.range'='2010-01-01,NOW', 
  'projection.partition_day.type'='date', 
  'storage.location.template'='s3://test-bucket/incoming/data/${partition_day}/')

and

CREATE EXTERNAL TABLE `event_index`(
  `systemid` string COMMENT 'from deserializer', 
  `eventtime` string COMMENT 'from deserializer', 
  `eventtype` string COMMENT 'from deserializer', 
  `source` string COMMENT 'from deserializer', 
  `updtdate` string COMMENT 'from deserializer', 
  `partition_day` string COMMENT 'from deserializer')
PARTITIONED BY ( 
  `partition_year` string)
ROW FORMAT SERDE 
  'org.openx.data.jsonserde.JsonSerDe' 
STORED AS INPUTFORMAT 
  'org.apache.hadoop.mapred.TextInputFormat' 
OUTPUTFORMAT 
  'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION
  's3://test-bucket/incoming/index/'
TBLPROPERTIES (
  'classification'='json', 
  'projection.enabled'='true', 
  'projection.partition_year.format'='yyyy', 
  'projection.partition_year.range'='2010,NOW', 
  'projection.partition_year.type'='date', 
  'storage.location.template'='s3://test-bucket/incoming/index/${partition_year}/')

then I inserted some data into partitions 2024-03-09 and 2024-03-11 and ran some simple test queries:

select * from event_data where partition_day = '2024-03-09'

select * from event_index where event_index.updtdate <= '2024-03-10'

SELECT * FROM event_index JOIN event_data USING(partition_day, systemid) where event_index.updtdate <= '2024-03-10'

By looking at the "Data scanned" metric when querying in the console, I can see clearly that partition pruning works fine with the single-table queries, but degrades to full table scan in the JOIN query.

Why is that (not) happening?

EDIT: if I simply swap the order of tables in the JOIN, the partition pruning happens as expected.

EDIT2: If I put the query with the swapped JOIN into a View, then the pruning fails again, so I'm back to where I started.

AlexR
已提問 2 個月前檢視次數 553 次
1 個回答
1

Hey, the problem you're experiencing is that when you join two tables in Athena, the query might scan all the data instead of just the relevant partitions. So, this happens because the query planner, which decides how to execute the query, might not recognize that it can skip some partitions based on your conditions.

When you swap the order of the tables in the join, the query planner can see which partitions to skip, so it scans less data. However, when you put this join in a view, the query planner loses this ability, and it scans all the data again.

To fix this, you can:

  1. Make sure your query clearly specifies which partitions it needs. For example, use WHERE partition_day = '2024-03-09' to tell the query planner to only look at data from March 9, 2024.
  2. Keep your join conditions simple and related to the partition columns, so the query planner can easily understand which partitions are needed.
  3. Check how your query is executed using the EXPLAIN command in Athena, which can give you hints on why it's scanning all the data.

Additional Resource:

profile picture
專家
已回答 2 個月前
profile picture
專家
已審閱 2 個月前

您尚未登入。 登入 去張貼答案。

一個好的回答可以清楚地回答問題並提供建設性的意見回饋,同時有助於提問者的專業成長。

回答問題指南