Shulou(Shulou.com)06/03 Report--

Zookeeper has been widely used in both real projects and distributed open source projects. Starting with this blog, I will bring you my understanding of Zookeeper. This series will cover Zookeeper introduction, environment construction, configuration instructions, Java operation Zookeeper (native API mode), zkclient operation Zookeeper mode, Zookeeper typical application scenario analysis and Curator framework.

Hello,Zookeeper

Zookeeper is a highly available distributed management and coordination framework based on the ZAB algorithm (Atomic Message Broadcasting Protocol). ZK can guarantee data consistency in distributed environment, which makes ZK a weapon to deal with distributed consistency problems! What are the characteristics of this sharp weapon? Take a look below and you will understand.

Reliability: Once Zookeeper has successfully applied a transaction and completed its response to the client, the state of all servers in Zookeeper's internal cluster is consistently preserved.

Single view: Due to the reliability guarantee above, the data seen by the client is consistent regardless of which server in the ZK cluster it is connected to.

Sequential consistency: Requests originating from a client are eventually applied to Zookeeper strictly in the order in which they originated. "In essence, ZK will number each request of each client. To put it bluntly, it assigns a globally unique incremental number, which reflects the sequence of all transaction operations."

Real-time: Real-time in the general sense means that once the transaction is successfully applied, the client will immediately obtain the changed new data from the server. ZK can only guarantee that the client will eventually get the latest data from the server within a certain time.

High availability: Within a ZK cluster, there will be one Leader and multiple followers. Once the Leader hangs, ZK will elect a new Leader through the Paxos algorithm. As long as more than half of the servers in the ZK cluster are working normally, ZK will be able to provide services to the outside world normally!

Simple data structure: tree structure similar to Linux file system, simple and practical! (Tree Hierarchy)

High performance: How high is the performance? For example, we often deal with distributed locks by creating temporary nodes. We should know that temporary nodes are stored in memory. Under the stress test of reading scenarios, QPS is as high as 10W+! In other words, ZK's performance in reading scenes is very outstanding!

A preliminary understanding of Zookeeper's data model

Each node is called a znode, a znode can have child directories, and each znode can store data (note in particular that temporary nodes cannot have child nodes).

If a znode is a temporary node, it means that the temporary znode will be automatically deleted once the client that created it loses contact with the ZK cluster. (In fact, the client communicates with ZK using a long connection and maintains the connection through heartbeat. This state is session. Once the session fails, the connection is disconnected and the temporary node will be deleted.)

Znode can be monitored, whether it is the data change of znode itself or the change of child nodes under this znode, which is also the core feature of ZK. (Many application scenarios are based on this feature, which will be described in detail later)

A preliminary understanding of Zookeeper's role composition

Here, let's first understand the identity characteristics of ZK Server:

Leader: Responsible for client write type requests

Follower: Responsible for client read type requests and can participate in Leader elections

Watcher: introduced later.

install Zookeeper

Take zookeeper-3.4.5 as an example, build a ZK cluster (at least 3 nodes are required, and the system time between each ZK SERVER is consistent). The list of machines used is: 192.168.99.121, 192.168.99.122, 192.168.99.123.

Take 192.168.99.121 as an example:

Configuration Description:

tickTime: heartbeat interval between ZK cluster and clients and servers within ZK cluster. Default is 2S.

initLimit: How many heartbeats a client can tolerate when connecting to a ZK cluster. The above configuration shows that if the client initially connects to the ZK cluster, the ZK cluster does not return a successful connection within 10*2S= 20S, which means that the connection fails.

syncLimit: Indicates how many heartbeats can be tolerated for a request response between a Leader and a Follower within the ZK cluster.

dataDir: Directory where ZK stores data and logs.

clientPort: The exposed interface of the ZK cluster, i.e. the port (2181) through which the client accesses the ZK cluster.

server.x = IP:port-a:port-b

X indicates which server it is, numbered from 0 and corresponding to myid file under dataDir.

port-a indicates the port (2888) through which the Leader and Follower communicate information.

port-b indicates port (3888) on which the Follower elects.

Basic ZK commands

Enter the client through zkCli.sh to operate:

Find ls, create znode create (note that each node has a value), get, set.

Let's look at the following attributes:

ctime and cZxid are a pair that represents the time and event number at which ZNODE was created;

mtime and mZxid are a pair, indicating the time and event number to modify the ZNODE data content;(these two attributes will be changed by the set command, but changes in the child nodes of this node will not affect this node)

dataVersion: indicates the version of ZNODE data. Note that when deleting ZNODE using JAVA native API, version is required to delete ZNODE. (Of course, we can provide-1 to bypass version checking)

DataLength and numChildren are easy to understand.

Also, note that the rmr directive recursively deletes ZNODEs; delete only deletes the specified ZNODE.

ZooInspector tool and IntelliJ IDEA integration with ZK

ZooInspector is a runnable JAR that can be directly connected to any SERVER in the ZK cluster after running.

OK, here, everything is ready ~

Let's look at JAVA operation ZK (based on Zookeeper's native API), distributed lock discussion, Watch features, etc.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.