Details
Major Description
When using a FilterList the scan is much slower compared to a scan with only a single filter (tested SingleColumnValueFilter and PrefixFilter).
The difference is extrem for very small ranges: if the range is only 10 rows the scan is 10 times slower when using the FilterList.
Is the cause just GC or object serialization/deserialization?
For a simple test I used the PerformanceEvaluation tool and created the TestTable with only 10
rows:
$ bin/hbase org.apache.hadoop.hbase.PerformanceEvaluation --rows=10 sequentialWrite 1
The attached test performs 100 scan using diffent filters. The filter is then wrapped into a FilterList (with only a single filter). This FilterList is then nested two more times into another FilterList. For each nested level the scan gets slower and slower.
The test created the following output:
Scan Null Filter (10): 391ms
Scan FilterList with Null Filter (0): 4788ms
Scan Nested FilterList with Null Filter (0): 8303ms
Scan Nested Nested FilterList with Null Filter (0): 11915ms
Scan SingleColumValueFilter Equal (0): 257ms
Scan FilterList with SingleColumValueFilter Equal (0): 4121ms
Scan Nested FilterList with SingleColumValueFilter Equal (0): 7965ms
Scan Nested Nested FilterList with SingleColumValueFilter Equal (0): 11600ms
Scan SingleColumValueFilter Not Equal (10): 912ms
Scan FilterList with SingleColumValueFilter Not Equal (10): 4542ms
Scan Nested FilterList with SingleColumValueFilter Not Equal (10): 8459ms
Scan Nested Nested FilterList with SingleColumValueFilter Not Equal (10): 11513ms
Scan PrefixFilter (10): 306ms
Scan FilterList with PrefixFilter (10): 3695ms
Scan Nested FilterList with PrefixFilter (10): 7762ms
Scan Nested Nested FilterList with PrefixFilter (10): 11721ms
Get: 245ms