Project practice, redis cluster technology learning (10)

2. Forget nodes

Because the nodes in the cluster keep passing Gossip Message exchange node status , Therefore, a robust mechanism is needed to make all nodes in the cluster forget the offline nodes . In other words, let other nodes no longer communicate with the nodes to be offline Gossip The message exchange .Redis Provides cluster forget{downNodeId} life

Enable this function .

When the node receives cluster forget{down NodeId} After the command , Will be able to nodeId The specified node is added to the disabled list , Nodes in the disabled list will no longer send Gossip news . The disabled list is valid for 60 second , exceed 60 Second, the node will participate in the message exchange again . That is, when the first time forget When the order is given , We have 60 In seconds, all nodes in the cluster forget to go offline .

Direct use of... Is not recommended for online operation cluster forget Command offline node , Need to interact with a large number of node commands , The actual operation is too cumbersome and easy to miss forget node . It is recommended to use redis-trib.rbdel-node{host：port}{downNodeId} command ,

When the offline master node has a slave node, you need to point the slave node to other master nodes , Therefore, for master-slave

Nodes are offline , It is recommended to offline the slave node first and then the master node , Prevent unnecessary full-scale repetition

system . about 6381 and 6384 Node offline operation , The order is as follows ：

redis-trib.rb del-node 127.0.0.1:6379

4fa7eac4080f0b667ffeab9b87841da49b84a6e4 # From the node 6384 id

redis-trib.rb del-node 127.0.0.1:6379

40b8d09d44294d2e23c7c768efc8fcd153446746 # Master node 6381 id

Confirm the node status after the node goes offline ：

127.0.0.1:6379> cluster nodes

Cluster node status no longer contains 6384 and 6381 node , up to now , We have completed the safe offline of the node , The new cluster structure is shown in the figure below ：

10.5 Request routing

At present, we have built Redis Cluster and understand the communication and scaling details , But the client has not been used to operate the cluster .Redis The cluster has made great changes to the client communication protocol , In order to maximize performance , Instead of using the proxy method, it uses the method of client directly connecting nodes . Therefore, for applications that want to switch from a single machine to a cluster environment, the client code needs to be modified .

2.Smart client ——JedisCluster

（1）JedisCluster The definition of

Jedis by Redis Cluster Provides Smart client , The corresponding class is JedisCluster, Its initialization method

as follows ：

public JedisCluster(Set<HostAndPort> jedisClusterNode, int connectionTimeout, int 
 soTimeout, int maxAttempts, final GenericObjectPoolConfig poolConfig) {
 ...
}

Which includes 5 Parameters ：

·Set<HostAndPort>jedisClusterNode： all Redis Cluster Node information （ It can also be part of , Because the client can use cluster slots Auto discovery ）.

·int connectionTimeout： Connection timeout .

·int soTimeout： Read write timeout .

·int maxAttempts： Retry count .

·GenericObjectPoolConfig poolConfig： Connection pool parameters ,JedisCluster Would be Redis Cluster Create a connection pool for each node .

For example, the following code shows a JedisCluster The initialization process

1. //  Initialize all nodes ( for example  6  Nodes )
2. Set<HostAndPort> jedisClusterNode = new HashSet<HostAndPort>();
3. jedisClusterNode.add(new HostAndPort("10.10.xx.1", 6379));
4. jedisClusterNode.add(new HostAndPort("10.10.xx.2", 6379));
5. jedisClusterNode.add(new HostAndPort("10.10.xx.3", 6379));
6. jedisClusterNode.add(new HostAndPort("10.10.xx.4", 6379));
7. jedisClusterNode.add(new HostAndPort("10.10.xx.5", 6379));
8. jedisClusterNode.add(new HostAndPort("10.10.xx.6", 6379));
9. //  initialization  commnon-pool  Connection pool ,  And set relevant parameters 
10. GenericObjectPoolConfig poolConfig = new GenericObjectPoolConfig();
11. //  initialization  JedisCluster
12. JedisCluster jedisCluster = new JedisCluster(jedisClusterNode, 1000,
1000, 5, poolConfig);

JedisCluster You can call commands , As shown below .

jedisCluster.set("hello", "world");
jedisCluster.get("key");

（2） Multi node commands and operations

Redis Cluster Although it provides distributed features , But some commands or operations , Such as keys、

flushall、 Delete the key of the specified mode , You need to traverse all nodes to complete . The following code implements from Redis Cluster Delete the function of the specified mode key ：

1. //  from  RedisCluster  Batch delete specified  pattern  The data of 
2. public void delRedisClusterByPattern(JedisCluster jedisCluster, String patte
rn,
3. int scanCounter) {
4. //  Get... Of all nodes  JedisPool
5. Map<String, JedisPool> jedisPoolMap = jedisCluster.getClusterNodes();
6. for (Entry<String, JedisPool> entry : jedisPoolMap.entrySet()) {
7. //  Get the  Jedis  Connect 
8. Jedis jedis = entry.getValue().getResource();
9. //  Delete only the primary node data 
10. if (!isMaster(jedis)) {
11. continue;
12. }
13. //  Use  Pipeline  Delete the data with the specified prefix every time 
14. Pipeline pipeline = jedis.pipelined();
15. //  Use  scan  Scan data with the specified prefix 
16. String cursor = "0";
17. //  Specify scan parameters ： Number of scans per time and  pattern
18. ScanParams params = new ScanParams().count(scanCounter).match(patter
n);
19. while (true) {
20. //  Perform a scan 
21. ScanResult<String> scanResult = jedis.scan(cursor, params);
22. //  Delete the  key  list 
23. List<String> keyList = scanResult.getResult();
24. if (keyList != null && keyList.size() > 0) {
25. for (String key : keyList) {
26. pipeline.del(key);
27. }
28. //  Batch deletion 
29. pipeline.syncAndReturnAll();
30. }
31. cursor = scanResult.getStringCursor();
32. //  If the cursor changes to  0,  Indicates that the scan is complete 
33. if ("0".equals(cursor)) { more  IT  For certification courses, please visit   beautiful 
break;
35. }
36. }
37. }
38. }
39. //  Judge the present  Redis  Is it  master  node 
40. private boolean isMaster(Jedis jedis) {
41. String[] data = jedis.info("Replication").split("\r\n");
42. for (String line : data) {
43. if ("role:master".equals(line.trim())) {
44. return true;
45. }
46. }
47. return false;
48. }

It is divided into the following steps ：

1） adopt jedisCluster.getClusterNodes（） Get the connection pool of all nodes .

2） Use info replication Screening 1） The master node in .

3） Traverse the master node , Use scan Command to find the specified pattern key, Use Pipeline Mechanism delete .

For example, the following operations traverse each time 1000 individual key, take Redis Cluster China and Israel user At the beginning key whole

Department delete .

String pattern = "user*";

int scanCounter = 1000;

delRedisClusterByPattern(jedisCluster, pattern, scanCounter);

So for keys、flushall And other commands that need to traverse all nodes , You can also refer to the above method to realize the corresponding functions .

（3） There are more methods of batch operation IT For certification courses, please visit Meihe learning online ww

Redis Cluster in , because key Distributed to each node , It will make it impossible to achieve mget、mset And so on . But it can be used CRC16 The algorithm calculates key Corresponding slot, as well as Smart The client saved slot Characteristics corresponding to nodes , Will belong to the same Redis Node key Filing , Then, the corresponding sub nodes of each node are analyzed respectively key List execution mget perhaps pipeline operation .

（4） Use Lua、 Methods of transaction and other characteristics

Lua And transactions need to be operated key, Must be on a node , however Redis Cluster Provides hashtag, If developers really want to use Lua Or business , You can set what you want to do key Use one hashtag, As shown below ：

1. // hashtag
2. String hastag = "{user}";
3. //  user  A  Your attention list 
4. String userAFollowKey = hastag + ":a:follow";
5. //  user  B  My fan list 
6. String userBFanKey = hastag + ":b:fans";
7. //  Calculation  hashtag  Corresponding  slot
8. int slot = JedisClusterCRC16.getSlot(hastag);
9. //  Get specified  slot  Of  JedisPool
10. JedisPool jedisPool = jedisCluster.getConnectionHandler().getJedisPoolFromSl
ot(slot);
11. //  Execute transactions on the same node 
12. Jedis jedis = null;
13. try {
14. jedis = jedisPool.getResource();
15. //  user  A  Join the user's attention table  B,  user  B  Add users to your fan list  A
16. Transaction transaction = jedis.multi();
17. transaction.sadd(userAFollowKey, "user:b");
18. transaction.sadd(userBFanKey, "user:a");
19. transaction.exec();
20. } catch (Exception e) {
21. logger.error(e.getMessage(), e);
22. } finally {
23. if (jedis!= null)
24. jedis.close();
25. }

All the in the transaction key add to hashtag.

2） Use CRC16 Calculation hashtag Corresponding slot.

3） Get specified slot The corresponding node connection pool JedisPool.

4） stay JedisPool Execute transactions on .