文档章节

YB-Master

MtrS
 MtrS
发布于 08/17 22:04
字数 568
阅读 2
收藏 0

YB-Master The YB-Master is the keeper of system meta-data/records, such as what tables exist in the system, where their tablets live, what users/roles exist, the permissions associated with them, and so on.

It is also responsible for coordinating background operations (such as load-balancing or initiating re-replication of under-replicated data) and performing a variety of administrative operations such as create/alter/drop of a table.

Note that the YB-Master is highly available as it forms a Raft group with its peers, and it is not in the critical path of IO against user tables.

master_overview

Functions of YB-Master Coordination of universe-wide admin operations Examples of such operations are user-issued create/alter/drop table requests, as well as a creating a backup of a table. The YB-Master performs these operations with a guarantee that the operation is propagated to all tablets irrespective of the state of the YB-TServers hosting these tablets. This is essential because a YB-TServer failure while one of these universe-wide operations is in progress cannot affect the outcome of the operation by failing to apply it on some tablets.

Storage of Ssystem metadata The master store system metadata such as the information about all the keyspaces, tables, roles, permissions, and assignment of tablets to YB-TServers. These system records are replicated across the YB-Masters for redundancy using Raft as well. The system metadata is also stored as a DocDB table by the YB-Master(s).

Authoritative source of tablet ssignments to YB-TServers The YB-Master stores all tablets and the corresponding YB-TServers that currently host them. This map of tablets to the hosting YB-TServers is queried by clients (such as the YQL layer). Applications using the YB smart clients for various languages (such as Cassandra, Redis, or PostgreSQL(beta)) are very efficient in retrieving data. The smart clients query the YB-Master for the tablet to YB-TServer map and cache it. By doing so, the smart clients can talk directly to the correct YB-TServer to serve various queries without incurring additional network hops.

Background Operations These operations performed throughout the lifetime of the universe in the background without impacting foreground read/write performance.

Data Placement & Load Balancing The YB-Master leader does the initial placement (at CREATE table time) of tablets across YB-TServers to enforce any user-defined data placement constraints and ensure uniform load. In addition, during the lifetime of the universe, as nodes are added, fail or decommissioned, it continues to balance the load and enforce data placement constraints automatically.

Leader Balancing Aside from ensuring that the number of tablets served by each YB-TServer is balanced across the universe, the YB-Masters also ensures that each node has a symmetric number of tablet-peer leaders across eligible nodes.

Re-replication of Data on Extended YB-TServer Failure The YB-Master receives heartbeats from all the YB-TServers, and tracks their liveness. It detects if any YB-TServers has failed, and keeps track of the time interval for which the YB-TServer remains in a failed state. If the time duration of the failure extends beyond a threshold, it finds replacement YB-TServers to which the tablet data of the failed YB-TServer is re-replicated. Re-replication is initiated in a throttled fashion by the YB-Master leader so as to not impact the foreground operations of the universe.

© 著作权归作者所有

上一篇: YugaByte TServer
下一篇: Yuga
MtrS
粉丝 35
博文 671
码字总数 410400
作品 0
榆林
私信 提问
前方交会计算

在三角形ABP中,已知A、B的坐标为(Xa, Ya)和(Xb,Yb),在A、B两站设点,测得角PAB,角PBA,通过计算出未知点P的坐标坐标: Xp = (XacotB + XbcotA - Ya + Yb) / (cotA + cotB); Yp = (YacotB...

乙知
2016/11/08
11
0
对话 | 如何建立 AI 研究实验室?Yoshua Bengio 有话要说

  选自CIFAR   作者:Graham Taylor   机器之心编译   参与:翁俊坚、王淑婷、张倩      作为机器学习领域的大牛,从带领三个人的团队到现在大规模的实验室,Bengio 有很多话要说...

机器之心
2018/08/11
0
0
专访Yoshua Bengio:如何高效建立和运营一家机器学习实验室?

     大数据文摘出品   编译:瓜瓜、艾玛、蒋宝尚   人工智能和机器学习风靡全球,高校作为人才培养基地正在顶风而上,如何建立相关的实验室,“当家人”如何管理实验室才能最高效?...

大数据文摘
2018/09/03
0
0
极坐标作点

选择已有线段AB,以已有线段为极轴,输入角度α和长度d,求点P坐标。 第一步:计算AB的长度L = 开根((Xb-Xa)² + (Yb-Ya)²) 第二步:根据有向线段AB坐标计算 dx = Xb - Xa dy = Yb - Ya 第三...

乙知
2016/11/08
8
0
理解 Deffie-Hellman 密钥交换算法

再拾此书时,怀念 4 年前阳光明媚的午后,我在浙大紫金港图书馆的三楼,梦回了中学奥数的小伙伴,和那时曾想证明哥德巴尔猜想的情怀! Deffie-Hellman 算法简介 Deffie-Hellman(简称 DH) 密钥...

koala bear
2016/02/04
0
0

没有更多内容

加载失败,请刷新页面

加载更多

用盛金公式解三次方程(ansi c版)

/* cc cubic.c -lm gcc cubic.c -lm Shengjin's Formulas Univariate cubic equation aX ^ 3 + bX ^ 2 + cX + d = 0, (a, b, c, d < R, and a!= 0). Multiple root disc......

wangxuwei
33分钟前
4
0
JBolt开发平台入门(16)- 弹出Dialog显示指定区域的html内容

在JBolt里,有个主从表Demo。 子表中除了普通的table显示,还有其它从表显示形式 比如下方案例:是针对一个升级管理子表中存放版本的changelog富文本信息。 需求是点击左上角的弹出查看按钮,...

山东-小木
38分钟前
29
0
TensorFlow .NET 队列操作

ThensorFlow能够并行处理多个任务,而队列是一种强大的异步计算机制。 如果我们拥有大型数据集,则可以大大加快模型的训练过程。 当以小批量读取,预处理和提取我们的训练数据时,此功能特别...

HaipingChen
38分钟前
8
0
Redis持久化

 这节介绍Redis的持久化,包括RDB和AOF两种方式。 1.RDB持久化  Redis能够将内存中的数据持久化到RDB文件中,避免数据丢失。RDB文件的格式如下示:  第一部分是开头的5个字节,值为REDIS...

程序员啊驼
39分钟前
5
0
黄海广:那些年做的学术公益——你不是一个人在战斗

来源:机器学习初学者 我叫黄海广,这篇文章主要写了我的学术公益之路。我希望初学者少踩一些坑,我也希望我能做个“雪中送炭”的人,希望越来越多的人加入到学术公益的道路上来。 2014年9月...

ApacheCN_飞龙
53分钟前
6
0

没有更多内容

加载失败,请刷新页面

加载更多

返回顶部
顶部