- 1、本文档共16页,可阅读全部内容。
- 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
HBase的二级索引概要1
;Huawei Hadoop RD – In a Glance;Senior Tech Lead in Huawei RD centre @Bangalore
Active contributor in Apache HBase community
Active member in HBase dev/user mailing list
Working with HBase and Hive;Master, Region servers
Table split into regions
Columnar storage, Column family
Memstore, Hfiles in DFS
Hfiles logically split into smaller blocks, data write/read as blocks;No indexing available in HBase
Scans with conditions on column value on huge sparse data.
Filter usage
Scan throughput very less
Timeouts at client side and
lease expiry;Index table
Persist index data in separate table
Co-processor based design
100% server side solution
Region wise indexing with collocation
1-1 mapping with actual table and index table regions
Region collocation using custom Load balancer;Secondary Index Architecture;Table creation;Put operation;Scan operation;Scan operation;CP reads index data and seek to exact rows in user table
Doing seeks on HFiles based on rowkey
obtained from index data
HFiles reads as block by block
Default block size is 64KB
Skipping block reads from HDFS where data not at all present
Some times skip a full HFile ?
No need to read index details back to
client avoiding network extra usage;Explicit split on index region is avoided
When user table region splits, corresponding index region also splits
Split key for index region same as that of user region
Custom HalfStoreFileReader for index daughter regions;Performance Test Results (Scan);4 Region servers, table with 160 regions
~10% degradation in put performance;Design
Supports multiple index on table and multi column index
Support indexing part of a column value
Support for equals and range condition Scans using index.
Support to dynamically add/drop index
Support bulk loading data to indexed table – Indexing done with bulk load
Application Usage
No change in scan code for the client app.
No need to specify the index(s) to be used. Intelligence to find best index(s) to be used for a Scan by looking at the Filt
文档评论(0)