[300 + selected interview questions from big companies continued to share] big data operation and maintenance sharp knife interview question column (VII)

Continuous sharing is useful 、 valuable 、 Selected high-quality big data interview questions

We are committed to building the most comprehensive big data interview topic database in the whole network

61、 Why do distributed clusters have Master？

Refer to the answer ：

         In a distributed environment , Some business logic only needs one machine in the cluster to execute , Other machines can share this result , This can greatly reduce the repeated calculation , Improve performance , So we need to do leader The election .

62、zk How are node outages handled ？

Refer to the answer ：

        Zookeeper It's also a cluster , The recommended configuration is no less than 3 Servers .Zookeeper We also need to ensure that when a node goes down , Other nodes will continue to provide services .

         If it's a Follower Downtime , also 2 Servers provide access , because Zookeeper There are multiple copies of the data on , Data is not lost ;

         If it's a Leader Downtime ,Zookeeper New ones will be elected Leader.

        ZK The mechanism of clustering is that as long as more than half of the nodes are normal , The cluster can provide services normally . Only in ZK Too many nodes , Only half or less of the nodes can work , Clusters fail .

         therefore 3 Of nodes cluster You can hang up 1 Nodes (leader You can get 2 ticket >1.5),2 Of nodes cluster You can't hang up on anything 1 There are nodes (leader You can get 1 ticket <=1).

63、Zookeeper What are several deployment modes ？

Refer to the answer ：

         Deployment mode ： standalone mode 、 Pseudo cluster mode 、 Cluster pattern .

64、 Cluster at least a few machines , What are the rules of clustering ?

Refer to the answer ：

         The cluster rule is 2N+1 platform ,N>0, namely 3 platform .

65、Zookeeper Does the cluster support dynamic addition of machines ？

Refer to the answer ：

         In fact, it is the horizontal expansion , There are two ways ：

         Restart all ： Close all Zookeeper service , Change the configuration and start . Does not affect previous client sessions .

         Restart one by one ： Under the principle that more than half of life is available , The restart of a machine does not affect the external service provided by the whole cluster . This is a common way .3.5 The version starts to support dynamic expansion .

66、Zookeeper For nodes watch Are monitoring notifications permanent ？ Why not forever ?

  Refer to the answer ：

          No .

          official declaration ： One Watch Event is a one-time trigger , When it's set Watch When the data has changed , The server sends the change to the settings Watch The client of , In order to inform them .

         Why not forever , for instance , If the server changes frequently , In many cases, the listening client , Every change should be notified to all clients , Put a lot of pressure on the network and servers .

         It's usually client execution getData(“/ node A”,true), If node A There has been a change or deletion , The client will get it watch event , But after that node A Another change has taken place , And the client has no settings watch event , No more sending to the client .

         in application , In many cases , Our client does not need to know every change of the server , I just need the latest data .

67、Zookeeper Of java What does the client have ？

Refer to the answer ：

        java client ：zk Self contained zkclient And Apache Open source Curator.

68、chubby What is it? , and zookeeper What do you think ？

Refer to the answer ：

        chubby yes google Of , Complete realization paxos Algorithm , Not open source .zookeeper yes chubby Open source implementation , Use zab agreement ,paxos Variations of the algorithm .

69、 Say a few zookeeper Common commands .

Refer to the answer ：

         Common commands ：ls get set create delete etc. .

70、ZAB and Paxos The relation and difference of algorithm ？

Refer to the answer ：

The same thing ：

        （1） There is a similarity between the two Leader The role of the process , It is responsible for coordinating multiple Follower Running process .

        （2）Leader The process will wait for more than half of Follower After giving the right feedback , To submit a proposal .

        （3）ZAB Agreement , Every Proposal All of them contain a epoch Value to represent the current Leader cycle ,Paxos The Chinese name is Ballot.

Difference ：

        ZAB It is used to build a highly available distributed data primary and secondary system （Zookeeper）,Paxos It is used to build a distributed consistency state machine system .

Continuous sharing is useful 、 valuable 、 Selected high-quality big data interview questions

We are committed to building the most comprehensive big data interview topic database in the whole network