[HBASE-6509] Implement fast-forwarding FuzzyRowFilter to allow filtering rows e.g. by "???alex?b"

• Type: New Feature
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.95.0
• Component/s: Filters
• Hadoop Flags: Reviewed
• Assignee: Alex Baranau
• Reporter: Alex Baranau
• Created: 03/Aug/12 21:55
• Updated: 23/Sep/13 18:45
• Resolved: 08/Aug/12 15:11

Description

Implement fuzzy row key filter to allow fetching records e.g. by this criteria: "???alex?b".

This seems to be very useful as an alternative to select records by row keys by specifying their part which is not prefix part. Due to fast-forwarding nature of the filter in many situations this helps to avoid heavy full-table scans.

This is especially effective when you have composite row key and (some of) its parts has fixed length. E.g. with the key of format userId_actionId_time, given that userId and actionId length is fixed, one can select user actions of specific type using fuzzy row key by specifying mask "????_myaction". Given fast-forwarding nature of filter, this will usually work much faster than doing whole table scan with any of the existing server-side filters.

In many cases this can work as secondary-indexing alternative.

Many times users implement it as a custom filter and many times they just don' know this is possible. Let's add it to the common codebase.

Issue Links

• breaks: HBASE-7892 FuzzyRowFilter would have wrong behaviors if user gives an arbitary byte for an "unfixed" position instead of byte 0
• is related to: HBASE-6527 Make custom filters plugin
• relates to: HBASE-7628 Port HBASE-6509 fast-forwarding FuzzyRowFilter to 0.94

Sub-Tasks

• Document FuzzyRowFilter: Closed