redis主从之间的复制分为两部分：

1. 全量复制
2. 增量复制

全量复制

redis之间在第一次实现主从关系时会进行全量复制，通过观察第一次主从关系的建立时的主从日志文件，主从关系首次建立时复制过程如下：

master 18

1401:M 19 Sep 2021 23:49:47.566 * Replica 10.0.0.82:6379 asks for synchronization
1401:M 19 Sep 2021 23:49:47.566 * Full resync requested by replica 10.0.0.82:6379      收到请求全量同步
1401:M 19 Sep 2021 23:49:47.566 * Starting BGSAVE for SYNC with target: disk          
1401:M 19 Sep 2021 23:49:47.567 * Background saving started by pid 1466                后台fork子进程执行bgsave
1466:C 19 Sep 2021 23:49:47.568 * DB saved on disk
1466:C 19 Sep 2021 23:49:47.568 * RDB: 0 MB of memory used by copy-on-write
1401:M 19 Sep 2021 23:49:47.598 * Background saving terminated with success
1401:M 19 Sep 2021 23:49:47.598 * Synchronization with replica 10.0.0.82:6379 succeeded

slave 28

1269:S 19 Sep 2021 23:49:47.469 * Ready to accept connections
1269:S 19 Sep 2021 23:49:47.469 * Connecting to MASTER 10.0.0.81:6379
1269:S 19 Sep 2021 23:49:47.469 * MASTER <-> REPLICA sync started
1269:S 19 Sep 2021 23:49:47.470 * Non blocking connect for SYNC fired the event.
1269:S 19 Sep 2021 23:49:47.470 * Master replied to PING, replication can continue...              主从关系检测
1269:S 19 Sep 2021 23:49:47.472 * Partial resynchronization not possible (no cached master)         no部分复制
1269:S 19 Sep 2021 23:49:47.473 * Full resync from master: e0e82f16a47b8955d76937b317bf2f7925c9b856:0    从master全量复制：master_replid
1269:S 19 Sep 2021 23:49:47.504 * MASTER <-> REPLICA sync: receiving 187 bytes from master
1269:S 19 Sep 2021 23:49:47.504 * MASTER <-> REPLICA sync: Flushing old data
1269:S 19 Sep 2021 23:49:47.504 * MASTER <-> REPLICA sync: Loading DB in memory
1269:S 19 Sep 2021 23:49:47.513 * MASTER <-> REPLICA sync: Finished with success

全量复制过程如下：

1. slave向master发送psync指令 MASTER <-> REPLICA sync started
2. master将数据同步的相关消息（runid、offset）回应给slave Full resync requested by replica 10.0.0.82:6379
3. slave保存同步消息
4. master的redis主进程fork出子进程执行bgsave（非阻塞）将内存中数据快照保存为rdb文件；在bgsave的同时将新写入的数据保存到buffer中
5. master发送RDB
6. master发送buffer
7. slave清楚原有旧数据 Flushing old data
8. load RDB Loading DB in memory
9. load buffer

master根据slave的runid来判断是否进行全量同步，因为redis每次重启的runid都不相同，且每次redis由master改为slave状态时都会清楚就数据。

在bgsave时，可能由于buffer过小，写入数据将buffer写满后冲刷buffer中的旧数据，容易造成主从数据不同不；从而造成复制失败或者循环的全量复制；所以buffer 需要根据不同环境合理设置其大小；
repl-timeout x 允许从节点的最大断开时间
repl-backlog-size y
其大小设置一般为大于 master的每秒写入量*允许从节点的最大断开时间

增量复制

如果由于网络原因造成原因造成主从断开，期间有数据写入master，再次形成主从时则会形成增量复制。

[root@81 etc]# iptables -A INPUT -s 10.0.0.82 -j REJECT    断开主从
127.0.0.1:6379> set haha hehe                              断开期间写入数据到master
OK
[root@81 etc]# iptables -F                                 恢复主从

观察master、slave日志如下

master 18

1401:M 20 Sep 2021 00:42:23.762 # Connection with replica 10.0.0.82:6379 lost.          失联
1401:M 20 Sep 2021 00:44:46.245 * Replica 10.0.0.82:6379 asks for synchronization
1401:M 20 Sep 2021 00:44:46.246 * Partial resynchronization request from 10.0.0.82:6379 accepted. Sending 127 bytes of backlog starting from offset 4408.                                                        接受从的增量同步

slave 28

1269:S 20 Sep 2021 00:44:45.139 * MASTER <-> REPLICA sync started
1269:S 20 Sep 2021 00:44:45.139 # Error condition on socket for SYNC: Connection refused
1269:S 20 Sep 2021 00:44:46.154 * Connecting to MASTER 10.0.0.81:6379
1269:S 20 Sep 2021 00:44:46.154 * MASTER <-> REPLICA sync started
1269:S 20 Sep 2021 00:44:46.155 * Non blocking connect for SYNC fired the event.
1269:S 20 Sep 2021 00:44:46.156 * Master replied to PING, replication can continue...
1269:S 20 Sep 2021 00:44:46.158 * Trying a partial resynchronization (request e0e82f16a47b8955d76937b317bf2f7925c9b856:4408).  尝试部分同步（请求 master_replid）
1269:S 20 Sep 2021 00:44:46.159 * Successful partial resynchronization with master.
1269:S 20 Sep 2021 00:44:46.159 * MASTER <-> REPLICA sync: Master accepted a Partial Resynchronization.

增量复制过程如下

1. connection lost
2. slave connection to master
3. master发送同步信息到slave
4. 进行部分同步，master传输主从断开期间的积压数据backlog