This article brings you relevant knowledge about Redis, which mainly introduces issues related to the sentinel mode, including the concept of sentinels, the role of sentinels, enabling sentinels, the principle of sentinels, etc. I hope Helpful to everyone.
Recommended learning: Redis Tutorial
Sentinel Mode
1. Sentinel Concept
In a master-multi-slave structure, if the master goes down, you need to select one from multiple slaves as the new master. To complete such a function, you need to Using Redis's sentinel mechanism
-
Sentinel (sentinel) is a distributed system used to monitor each server in the master-slave structure, when a failure occurs, the voting mechanism selects the new master and connects all slaves to the new master, as shown in the following figure:
2. Sentinel function
-
Monitoring: Monitor master and slave
- Continuously check whether master and slave Normal operation
- Master survival detection, master and slave operation status detection
Notification (reminder): When there is a problem with the monitored server, send notifications to other sentinels and Redis servers
Failover: Disconnect the downed master and slave, select One slave serves as the master, connects other slaves to the new master, and informs the client of the new server address
Note: Sentinel is also a redis server and only provides Monitoring service, does not provide data management services; usually the number of sentinels is configured as an odd number (to facilitate voting)
3. Enable sentinels
-
Modify sentinels The configuration file sentinel.conf
is located in the Redis directory, as shown in the figure below:
-
The meaning of some configurations is as shown in the figure below Display:
-
Start Sentinel
redis-sentinel filename
#filename指的是配置文件名
#每个哨兵都要配置自己的配置文件
Copy after login
After configuring and starting Sentinel, after the main server goes down, it willautomatically Execution of voting, master-slave switching and other processes
4. Sentinel principle
The sentinel goes through three processes during the master-slave switching process Stage:
- Monitoring
- Notification
- Failover
4.1 Monitoring
Used to synchronize the status information of each node. These nodes include master, slave, sentinel, that is, the information of all servers in the entire system.
Mainly completed functions:
Get the status of each sentinel (whether online or not): newly added The sentinels need to interact with the existing sentinels in the system
-
Get the status of the master (through the info command)
-
Get the status of all slaves (send info instructions to all slaves based on the slave information in the master)
The specific internal working principle is as follows:
- After Sentinel 1 sends the info command to the master, a cmd connection will be established. The created connection is used to send commands.
- After the cmd connection is created, the current situation will be saved on the Sentinel 1 side. For all the information he obtains, the master at the other end will also save the information it holds.
- Then Sentinel 1 sends the info command to the slave based on the information about the salve obtained from the master to obtain the salve information and enrich this end. Saved information
- When a new Sentinel 2 is added, Sentinel 2 sends an info command to the master to establish a cmd connection. According to the information in the master, the previously existing Sentinel 1 can be obtained on its own end. Save the information you have obtained. Then determine whether Sentinel 1 is online and establish a connection with Sentinel 1. The two exchange information with each other, and both parties will continue to ping to ensure that there is smooth communication between them.
- Sentinel 2 is based on the slave information obtained from the master , and then obtain information from the slave to enrich the information saved by itself
- When adding a Sentinel 3, it is similar to the previous process, and finally the three Sentinels established a relationship network
- Relationship network The three of them will exchange and send information to each other. This working model in the relationship network is called publish and subscribe model
4.2 Notification
During the notification phase, the sentinel must continuously obtain the information of the master and slave, and then share the obtained information among the various sentinels.
The specific process is as follows:
The sentinel sends the hello
command to the master and slave through the established cmd connection to obtain their information, and then Share in the relationship network:
##4.3 Failover
Disconnect the downed master from the slave and select a slave As the master, connect other slaves to the new master and inform the client of the new server address.
The specific process is as follows:
- # Sentinel 1 never gets a response from the master. It is subjectively judged that the master is down and marked
sdown (Only one sentinel thinks the master is down)
- Sentinel 1 delivers messages in the relationship network and sends the message in the upper left corner of the picture to other sentinels
- After other sentinels get the message, they all send messages to the master to determine whether the master is down. They will also send their results to other sentinels in the network. After confirming that the master is down, they will The mark is changed to
odown (after all sentinel detections, more than half of the sentinels believe that the master is down)
- When it is determined that the master is down, it is necessary at this time To elect a new master, which sentinel will complete this task must be decided through the election among the sentinels. The specific process is as follows:
During the election, each sentinel has one vote. Each sentinel will issue an instruction to tell other sentinels on the intranet that they will be electors. For example, if sentinel1 and sentinel2 issue this election instruction, then after sentinel3 receives their applications, sentinel3 will vote for one of the parties. , vote according to the order in which - arrives, the party that gets the votes will have one more electoral vote. According to this form, there will eventually be an election result, and the sentinel with the most votes will be elected.
There may be failures in this process. For example, if the election is not successful in the first round, then the second and third rounds will continue until the election is completed-
- After the election between sentinels is completed, the winning sentinel will then select one of the slaves and turn it into a new master. The specific process is as follows:
The principle of Sentinel selecting alternative masters from the server list: - Exclude first, then select
Outsides that are not online- Outsides that respond slowly
- Outsides that have been disconnected from the original master for a long time
- Exclusion ends and selection begins
-
The ones with higher priority will be selected - The ones with larger offset will be selected (larger means that the original master has the most data synchronized)
- If it has not been selected yet, it will be selected based on runid.
-
- After selecting the new master, the sentry sends instructions to the server:
- Also tell other sentinels who the new master is
5. Summary
Monitoring phase: synchronization information
Notification phase: maintain connectivity
Failover phase:
The sentinel detects the problem and marks subjective offline and objective offline in turn- Campaign responsible Sentinel
- Prefer the new master
- The new master takes office, and other slaves connect to the new master.
- After the original master recovers, it connects to the new master as a slave
Recommended learning:
Redis video tutorial
The above is the detailed content of Let's analyze the Redis sentry mode together. For more information, please follow other related articles on the PHP Chinese website!