[ 
https://issues.apache.org/jira/browse/HBASE-15676?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Matt Warhaftig updated HBASE-15676:
-----------------------------------
    Status: Patch Available  (was: Open)

The FuzzyRowFilter constructor as a side effect modifies passed mask values 
(0->-1 & 1->0).  When a FuzzyRowFilter reuses a previously used mask, the 
FuzzyRowFilter constructor gets a mask that is already updated to -1s & 0s.  
FuzzyRowFilter logic has isPreprocessedMask() to check mask state and only 
update new mask.  However, as [~talktorohi...@gmail.com] found, that check 
fails for a mask of all 0s since that value is ambiguous (both new and 
previously updated masks could contain all 0s).

Attached patch 'hbase-15676-v1.patch' adds an isPreprocessed flag byte to the 
mask to track if previously updated.  Adding a flag byte is a bit inelegant but 
FuzzyRowFilter expects masks to be reusable while retaining state and being 
modified by FuzzyRowFilter's constructor.

> FuzzyRowFilter fails and matches all the rows in the table if the mask 
> consists of all 0s
> -----------------------------------------------------------------------------------------
>
>                 Key: HBASE-15676
>                 URL: https://issues.apache.org/jira/browse/HBASE-15676
>             Project: HBase
>          Issue Type: Bug
>          Components: Filters
>    Affects Versions: 1.1.1, 1.2.0, 1.0.2, 0.98.13, 2.0.0
>            Reporter: Rohit Sinha
>
> While using FuzzyRowFilter we noticed that if the mask array consists of all 
> 0s (fixed) the FuzzyRowFilter matches all the rows in the table. We noticed 
> this on HBase 1.1, 1.2 and higher.
> After some digging we suspect that this is because of isPreprocessedMask() 
> check which is used in preprocessMask() which was added here: 
> https://issues.apache.org/jira/browse/HBASE-13761
> If the mask consists of all 0s then the isPreprocessedMask() returns true and 
> the preprocessing which responsible for changing 0s to -1 doesn't happen and 
> hence all rows are matched in scan.
> This scenario can be tested in TestFuzzyRowFilterEndToEnd#testHBASE14782() If 
> we change the 
> byte[] fuzzyKey = Bytes.toBytesBinary("\\x00\\x00\\x044");
> byte[] mask = new byte[] {1,0,0,0};
> to 
> byte[] fuzzyKey = Bytes.toBytesBinary("\\x9B\\x00\\x044e");
> byte[] mask = new byte[] {0,0,0,0,0};
> We expect one match but this will match all the rows in the table. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to