主从复制
从服务器执行 SLAVEOF IP PORT 命令即可实现对主服务器的复制。
复制又分为完整同步和部分同步（2.8之后）
完整同步：与混合持久化过程类似，先以rdb的方式保存rdb文件然后发送给从服务器。再将期间的命令保存到复制缓存区（每个从节点都有一份），最后以aof形式发给从服务器实现同步。
部分同步：因为存在这么一种情况，从服务器中间只断开了几秒钟的时间，然后又立马重连上主服务器了。这个时候如果无脑完整同步无疑是没有必要的，psync命令只有当需要的时候才会执行完整同步，否则只需将断线期间的命令同步过去就好了。其现原理依靠：
1：主/从服务器各自的复制偏移量（已发送/复制的偏移量）
2：主服务器的复制积压缓存区（一段时间内写命令的保存，只有一份）
3：服务器的运行id（唯一标识）
通过对比主从服务器的复制偏移量找到在主服务器的复制积压缓存区找到已发送但从服务器未同步的命令。当然缓存区的大小是有限制的（默认1M），当超过了这个限制还是只能完整同步。
SLAVEOF命令源码流程如下
replicaofCommand
源码解析第一步当然是找到slaveof的命令执行处理器replicaofCommand
可以发现这个处理器保存了主服务器的ip和port并将server.repl_state设为REPL_STATE_CONNECT（以便定时器触发下一步操作）就返回ok了。
数据结构：
connectWithMaster
而后由定时器建立主从套接字连接并创建文件事件，最后同步状态为 REPL_STATE_CONNECTING
syncWithMaster
接下来的步骤就在syncWithMaster方法了（由上一步创建的ae事件触发）
void syncWithMaster(aeEventLoop *el, int fd, void *privdata, int mask) {
char tmpfile[256], *err = NULL;
int dfd = -1, maxtries = 5;
int sockerr = 0, psync_result;
socklen_t errlen = sizeof(sockerr);
UNUSED(el);
UNUSED(privdata);
UNUSED(mask);
//REPL_STATE_NONE直接返回
if (server.repl_state == REPL_STATE_NONE) {
close(fd);
return;
}
if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &sockerr, &errlen) == -1)
sockerr = errno;
if (sockerr) {
serverLog(LL_WARNING,"Error condition on socket for SYNC: %s",
strerror(sockerr));
goto error;
}
//发送ping命令，检查网络。然后下一步REPL_STATE_RECEIVE_PONG
if (server.repl_state == REPL_STATE_CONNECTING) {
serverLog(LL_NOTICE,"Non blocking connect for SYNC fired the event.");
aeDeleteFileEvent(server.el,fd,AE_WRITABLE);
server.repl_state = REPL_STATE_RECEIVE_PONG;
err = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"PING",NULL);
if (err) goto write_error;
return;
}
//接收ping回复。然后下一步REPL_STATE_SEND_AUTH
if (server.repl_state == REPL_STATE_RECEIVE_PONG) {
err = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
if (err[0] != '+' &&
strncmp(err,"-NOAUTH",7) != 0 &&
strncmp(err,"-ERR operation not permitted",28) != 0)
{
serverLog(LL_WARNING,"Error reply to PING from master: '%s'",err);
sdsfree(err);
goto error;
} else {
serverLog(LL_NOTICE,
"Master replied to PING, replication can continue...");
}
sdsfree(err);
server.repl_state = REPL_STATE_SEND_AUTH;
}
//发送AUTH命令，检查身份信息。然后下一步REPL_STATE_RECEIVE_AUTH（不需要身份校验直接跳过进入为REPL_STATE_SEND_PORT）
if (server.repl_state == REPL_STATE_SEND_AUTH) {
if (server.masterauth) {
err = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"AUTH",server.masterauth,NULL);
if (err) goto write_error;
server.repl_state = REPL_STATE_RECEIVE_AUTH;
return;
} else {
server.repl_state = REPL_STATE_SEND_PORT;
}
}
//接收AUTH回复。然后下一步REPL_STATE_SEND_PORT
if (server.repl_state == REPL_STATE_RECEIVE_AUTH) {
err = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
if (err[0] == '-') {
serverLog(LL_WARNING,"Unable to AUTH to MASTER: %s",err);
sdsfree(err);
goto error;
}
sdsfree(err);
server.repl_state = REPL_STATE_SEND_PORT;
}
//发送端口号给主节点。然后下一步REPL_STATE_RECEIVE_PORT
if (server.repl_state == REPL_STATE_SEND_PORT) {
sds port = sdsfromlonglong(server.slave_announce_port ?
server.slave_announce_port : server.port);
err = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"REPLCONF",
"listening-port",port, NULL);
sdsfree(port);
if (err) goto write_error;
sdsfree(err);
server.repl_state = REPL_STATE_RECEIVE_PORT;
return;
}
//接收端口号发送的回复。然后下一步REPL_STATE_SEND_IP
if (server.repl_state == REPL_STATE_RECEIVE_PORT) {
err = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
/* Ignore the error if any, not all the Redis versions support
* REPLCONF listening-port. */
if (err[0] == '-') {
serverLog(LL_NOTICE,"(Non critical) Master does not understand "
"REPLCONF listening-port: %s", err);
}
sdsfree(err);
server.repl_state = REPL_STATE_SEND_IP;
}
//发送ip地址给主节点，然后下一步REPL_STATE_SEND_CAPA
if (server.repl_state == REPL_STATE_SEND_IP &&
server.slave_announce_ip == NULL)
{
server.repl_state = REPL_STATE_SEND_CAPA;
}
if (server.repl_state == REPL_STATE_SEND_IP) {
err = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"REPLCONF",
"ip-address",server.slave_announce_ip, NULL);
if (err) goto write_error;
sdsfree(err);
server.repl_state = REPL_STATE_RECEIVE_IP;
return;
}
//接收ip发送的回复。然后下一步REPL_STATE_SEND_CAPA
if (server.repl_state == REPL_STATE_RECEIVE_IP) {
err = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
if (err[0] == '-') {
serverLog(LL_NOTICE,"(Non critical) Master does not understand "
"REPLCONF ip-address: %s", err);
}
sdsfree(err);
server.repl_state = REPL_STATE_SEND_CAPA;
}
//告知主服务器从节点的一些信息，如：以eof结尾和支持psync。然后下一步REPL_STATE_RECEIVE_CAPA
if (server.repl_state == REPL_STATE_SEND_CAPA) {
err = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"REPLCONF",
"capa","eof","capa","psync2",NULL);
if (err) goto write_error;
sdsfree(err);
server.repl_state = REPL_STATE_RECEIVE_CAPA;
return;
}
//接收信息通知的回复。然后下一步REPL_STATE_SEND_PSYNC
if (server.repl_state == REPL_STATE_RECEIVE_CAPA) {
err = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
if (err[0] == '-') {
serverLog(LL_NOTICE,"(Non critical) Master does not understand "
"REPLCONF capa: %s", err);
}
sdsfree(err);
server.repl_state = REPL_STATE_SEND_PSYNC;
}
//发送PSYNC命令，并尽可能部分同步
if (server.repl_state == REPL_STATE_SEND_PSYNC) {
if (slaveTryPartialResynchronization(fd,0) == PSYNC_WRITE_ERROR) {
err = sdsnew("Write error sending the PSYNC command.");
goto write_error;
}
server.repl_state = REPL_STATE_RECEIVE_PSYNC;
return;
}
//接收PSYNC回复
if (server.repl_state != REPL_STATE_RECEIVE_PSYNC) {
serverLog(LL_WARNING,"syncWithMaster(): state machine error, "
"state should be RECEIVE_PSYNC but is %d",
server.repl_state);
goto error;
}
psync_result = slaveTryPartialResynchronization(fd,1);
if (psync_result == PSYNC_WAIT_REPLY) return; /* Try again later... */
if (psync_result == PSYNC_TRY_LATER) goto error;
if (psync_result == PSYNC_CONTINUE) {
serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Master accepted a Partial Resynchronization.");
return;
}
//回收套接字资源
disconnectSlaves(); /* Force our slaves to resync with us as well. */
freeReplicationBacklog(); /* Don't allow our chained slaves to PSYNC. */
//PSYNC不支持就改用SYNC
if (psync_result == PSYNC_NOT_SUPPORTED) {
serverLog(LL_NOTICE,"Retrying with SYNC...");
if (syncWrite(fd,"SYNC\r\n",6,server.repl_syncio_timeout*1000) == -1) {
serverLog(LL_WARNING,"I/O error writing to MASTER: %s",
strerror(errno));
goto error;
}
}
//到这一步说明是全量同步。因为比较耗时所以创建时间，再由事件处理器readSyncBulkPayload异步接收主节点传过来的rdb文件
while(maxtries--) {
snprintf(tmpfile,256,
"temp-%d.%ld.rdb",(int)server.unixtime,(long int)getpid());
dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644);
if (dfd != -1) break;
sleep(1);
}
if (dfd == -1) {
serverLog(LL_WARNING,"Opening the temp file needed for MASTER <-> REPLICA synchronization: %s",strerror(errno));
goto error;
}
if (aeCreateFileEvent(server.el,fd, AE_READABLE,readSyncBulkPayload,NULL)
== AE_ERR)
{
serverLog(LL_WARNING,
"Can't create readable event for SYNC: %s (fd=%d)",
strerror(errno),fd);
goto error;
}
//最后同步状态为 REPL_STATE_TRANSFER
server.repl_state = REPL_STATE_TRANSFER;
server.repl_transfer_size = -1;
server.repl_transfer_read = 0;
server.repl_transfer_last_fsync_off = 0;
server.repl_transfer_fd = dfd;
server.repl_transfer_lastio = server.unixtime;
server.repl_transfer_tmpfile = zstrdup(tmpfile);
return;
error:
aeDeleteFileEvent(server.el,fd,AE_READABLE|AE_WRITABLE);
if (dfd != -1) close(dfd);
close(fd);
server.repl_transfer_s = -1;
server.repl_state = REPL_STATE_CONNECT;
return;
write_error: /* Handle sendSynchronousCommand(SYNC_CMD_WRITE) errors. */
serverLog(LL_WARNING,"Sending command to master in replication handshake: %s", err);
sdsfree(err);
goto error;
}
PSYNC
syncWithMaster里最重要的步骤还得是从服务器psync命令的发送/回复处理函数以及主服务器的命令执行器，下面就从主/从两个角度着重看一下代码。
从服务器
PSYNC命令的发送及对主服务器不同回复的处理逻辑主要在slaveTryPartialResynchronization函数
int slaveTryPartialResynchronization(int fd, int read_reply) {
char *psync_replid;
char psync_offset[32];
sds reply;
//发送PSYNC
if (!read_reply) {
/* Initially set master_initial_offset to -1 to mark the current
* master run_id and offset as not valid. Later if we'll be able to do
* a FULL resync using the PSYNC command we'll set the offset at the
* right value, so that this information will be propagated to the
* client structure representing the master into server.master. */
server.master_initial_offset = -1;
if (server.cached_master) {
psync_replid = server.cached_master->replid;
snprintf(psync_offset,sizeof(psync_offset),"%lld", server.cached_master->reploff+1);
serverLog(LL_NOTICE,"Trying a partial resynchronization (request %s:%s).", psync_replid, psync_offset);
} else {
serverLog(LL_NOTICE,"Partial resynchronization not possible (no cached master)");
psync_replid = "?";
memcpy(psync_offset,"-1",3);
}
/* Issue the PSYNC command */
reply = sendSynchronousCommand(SYNC_CMD_WRITE,fd,"PSYNC",psync_replid,psync_offset,NULL);
if (reply != NULL) {
serverLog(LL_WARNING,"Unable to send PSYNC to master: %s",reply);
sdsfree(reply);
aeDeleteFileEvent(server.el,fd,AE_READABLE);
return PSYNC_WRITE_ERROR;
}
return PSYNC_WAIT_REPLY;
}
//接收回复
reply = sendSynchronousCommand(SYNC_CMD_READ,fd,NULL);
if (sdslen(reply) == 0) {
sdsfree(reply);
return PSYNC_WAIT_REPLY;
}
aeDeleteFileEvent(server.el,fd,AE_READABLE);
//FULLRESYNC-全量同步（逻辑返回，后续readSyncBulkPayload异步执行）
if (!strncmp(reply,"+FULLRESYNC",11)) {
char *replid = NULL, *offset = NULL;
replid = strchr(reply,' ');
if (replid) {
replid++;
offset = strchr(replid,' ');
if (offset) offset++;
}
if (!replid || !offset || (offset-replid-1) != CONFIG_RUN_ID_SIZE) {
serverLog(LL_WARNING,
"Master replied with wrong +FULLRESYNC syntax.");
memset(server.master_replid,0,CONFIG_RUN_ID_SIZE+1);
} else {
memcpy(server.master_replid, replid, offset-replid-1);
server.master_replid[CONFIG_RUN_ID_SIZE] = '\0';
server.master_initial_offset = strtoll(offset,NULL,10);
serverLog(LL_NOTICE,"Full resync from master: %s:%lld",
server.master_replid,
server.master_initial_offset);
}
replicationDiscardCachedMaster();
sdsfree(reply);
return PSYNC_FULLRESYNC;
}
//+CONTINUE部分同步（同步执行）
if (!strncmp(reply,"+CONTINUE",9)) {
serverLog(LL_NOTICE,
"Successful partial resynchronization with master.");
char *start = reply+10;
char *end = reply+9;
while(end[0] != '\r' && end[0] != '\n' && end[0] != '\0') end++;
if (end-start == CONFIG_RUN_ID_SIZE) {
char new[CONFIG_RUN_ID_SIZE+1];
memcpy(new,start,CONFIG_RUN_ID_SIZE);
new[CONFIG_RUN_ID_SIZE] = '\0';
if (strcmp(new,server.cached_master->replid)) {
serverLog(LL_WARNING,"Master replication ID changed to %s",new);
memcpy(server.replid2,server.cached_master->replid,
sizeof(server.replid2));
server.second_replid_offset = server.master_repl_offset+1;
memcpy(server.replid,new,sizeof(server.replid));
memcpy(server.cached_master->replid,new,sizeof(server.replid));
disconnectSlaves();
}
}
sdsfree(reply);
//部分同步主逻辑方法
replicationResurrectCachedMaster(fd);
if (server.repl_backlog == NULL) createReplicationBacklog();
return PSYNC_CONTINUE;
}
//其他报错情况，如命令不支持
if (!strncmp(reply,"-NOMASTERLINK",13) ||
!strncmp(reply,"-LOADING",8))
{
serverLog(LL_NOTICE,
"Master is currently unable to PSYNC "
"but should be in the future: %s", reply);
sdsfree(reply);
return PSYNC_TRY_LATER;
}
if (strncmp(reply,"-ERR",4)) {
serverLog(LL_WARNING,
"Unexpected reply to PSYNC from master: %s", reply);
} else {
serverLog(LL_NOTICE,
"Master does not support PSYNC or is in "
"error state (reply: %s)", reply);
}
sdsfree(reply);
replicationDiscardCachedMaster();
return PSYNC_NOT_SUPPORTED;
}
针对主服务器部分同步的回复直接由replicationResurrectCachedMaster函数同步处理
void replicationResurrectCachedMaster(int newfd) {
server.master = server.cached_master;
server.cached_master = NULL;
server.master->fd = newfd;
server.master->flags &= ~(CLIENT_CLOSE_AFTER_REPLY|CLIENT_CLOSE_ASAP);
server.master->authenticated = 1;
server.master->lastinteraction = server.unixtime;
server.repl_state = REPL_STATE_CONNECTED;
server.repl_down_since = 0;
/* Re-add to the list of clients. */
linkClient(server.master);
if (aeCreateFileEvent(server.el, newfd, AE_READABLE,
readQueryFromClient, server.master)) {
serverLog(LL_WARNING,"Error resurrecting the cached master, impossible to add the readable handler: %s", strerror(errno));
freeClientAsync(server.master); /* Close ASAP. */
}
/* We may also need to install the write handler as well if there is
* pending data in the write buffers. */
if (clientHasPendingReplies(server.master)) {
if (aeCreateFileEvent(server.el, newfd, AE_WRITABLE,
sendReplyToClient, server.master)) {
serverLog(LL_WARNING,"Error resurrecting the cached master, impossible to add the writable handler: %s", strerror(errno));
freeClientAsync(server.master); /* Close ASAP. */
}
}
}
而由于全量同步比较耗时，所有这里(slaveTryPartialResynchronization)只做逻辑返回，由后面的readSyncBulkPayload事件处理器异步处理（异步接收主服务器发来的rdb文件）。
void readSyncBulkPayload(aeEventLoop *el, int fd, void *privdata, int mask) {
char buf[4096];
ssize_t nread, readlen, nwritten;
off_t left;
UNUSED(el);
UNUSED(privdata);
UNUSED(mask);
/* Static vars used to hold the EOF mark, and the last bytes received
* form the server: when they match, we reached the end of the transfer. */
static char eofmark[CONFIG_RUN_ID_SIZE];
static char lastbytes[CONFIG_RUN_ID_SIZE];
static int usemark = 0;
/* If repl_transfer_size == -1 we still have to read the bulk length
* from the master reply. */
if (server.repl_transfer_size == -1) {
if (syncReadLine(fd,buf,1024,server.repl_syncio_timeout*1000) == -1) {
serverLog(LL_WARNING,
"I/O error reading bulk count from MASTER: %s",
strerror(errno));
goto error;
}
if (buf[0] == '-') {
serverLog(LL_WARNING,
"MASTER aborted replication with an error: %s",
buf+1);
goto error;
} else if (buf[0] == '\0') {
/* At this stage just a newline works as a PING in order to take
* the connection live. So we refresh our last interaction
* timestamp. */
server.repl_transfer_lastio = server.unixtime;
return;
} else if (buf[0] != '$') {
serverLog(LL_WARNING,"Bad protocol from MASTER, the first byte is not '$' (we received '%s'), are you sure the host and port are right?", buf);
goto error;
}
/* There are two possible forms for the bulk payload. One is the
* usual $<count> bulk format. The other is used for diskless transfers
* when the master does not know beforehand the size of the file to
* transfer. In the latter case, the following format is used:
*
* $EOF:<40 bytes delimiter>
*
* At the end of the file the announced delimiter is transmitted. The
* delimiter is long and random enough that the probability of a
* collision with the actual file content can be ignored. */
if (strncmp(buf+1,"EOF:",4) == 0 && strlen(buf+5) >= CONFIG_RUN_ID_SIZE) {
usemark = 1;
memcpy(eofmark,buf+5,CONFIG_RUN_ID_SIZE);
memset(lastbytes,0,CONFIG_RUN_ID_SIZE);
/* Set any repl_transfer_size to avoid entering this code path
* at the next call. */
server.repl_transfer_size = 0;
serverLog(LL_NOTICE,
"MASTER <-> REPLICA sync: receiving streamed RDB from master");
} else {
usemark = 0;
server.repl_transfer_size = strtol(buf+1,NULL,10);
serverLog(LL_NOTICE,
"MASTER <-> REPLICA sync: receiving %lld bytes from master",
(long long) server.repl_transfer_size);
}
return;
}
/* Read bulk data */
if (usemark) {
readlen = sizeof(buf);
} else {
left = server.repl_transfer_size - server.repl_transfer_read;
readlen = (left < (signed)sizeof(buf)) ? left : (signed)sizeof(buf);
}
nread = read(fd,buf,readlen);
if (nread <= 0) {
serverLog(LL_WARNING,"I/O error trying to sync with MASTER: %s",
(nread == -1) ? strerror(errno) : "connection lost");
cancelReplicationHandshake();
return;
}
server.stat_net_input_bytes += nread;
/* When a mark is used, we want to detect EOF asap in order to avoid
* writing the EOF mark into the file... */
int eof_reached = 0;
if (usemark) {
/* Update the last bytes array, and check if it matches our delimiter.*/
if (nread >= CONFIG_RUN_ID_SIZE) {
memcpy(lastbytes,buf+nread-CONFIG_RUN_ID_SIZE,CONFIG_RUN_ID_SIZE);
} else {
int rem = CONFIG_RUN_ID_SIZE-nread;
memmove(lastbytes,lastbytes+nread,rem);
memcpy(lastbytes+rem,buf,nread);
}
if (memcmp(lastbytes,eofmark,CONFIG_RUN_ID_SIZE) == 0) eof_reached = 1;
}
server.repl_transfer_lastio = server.unixtime;
if ((nwritten = write(server.repl_transfer_fd,buf,nread)) != nread) {
serverLog(LL_WARNING,"Write error or short write writing to the DB dump file needed for MASTER <-> REPLICA synchronization: %s",
(nwritten == -1) ? strerror(errno) : "short write");
goto error;
}
server.repl_transfer_read += nread;
/* Delete the last 40 bytes from the file if we reached EOF. */
if (usemark && eof_reached) {
if (ftruncate(server.repl_transfer_fd,
server.repl_transfer_read - CONFIG_RUN_ID_SIZE) == -1)
{
serverLog(LL_WARNING,"Error truncating the RDB file received from the master for SYNC: %s", strerror(errno));
goto error;
}
}
/* Sync data on disk from time to time, otherwise at the end of the transfer
* we may suffer a big delay as the memory buffers are copied into the
* actual disk. */
if (server.repl_transfer_read >=
server.repl_transfer_last_fsync_off + REPL_MAX_WRITTEN_BEFORE_FSYNC)
{
off_t sync_size = server.repl_transfer_read -
server.repl_transfer_last_fsync_off;
rdb_fsync_range(server.repl_transfer_fd,
server.repl_transfer_last_fsync_off, sync_size);
server.repl_transfer_last_fsync_off += sync_size;
}
/* Check if the transfer is now complete */
if (!usemark) {
if (server.repl_transfer_read == server.repl_transfer_size)
eof_reached = 1;
}
if (eof_reached) {
int aof_is_enabled = server.aof_state != AOF_OFF;
/* Ensure background save doesn't overwrite synced data */
if (server.rdb_child_pid != -1) {
serverLog(LL_NOTICE,
"Replica is about to load the RDB file received from the "
"master, but there is a pending RDB child running. "
"Killing process %ld and removing its temp file to avoid "
"any race",
(long) server.rdb_child_pid);
kill(server.rdb_child_pid,SIGUSR1);
rdbRemoveTempFile(server.rdb_child_pid);
}
/* Make sure the new file (also used for persistence) is fully synced
* (not covered by earlier calls to rdb_fsync_range). */
if (fsync(server.repl_transfer_fd) == -1) {
serverLog(LL_WARNING,
"Failed trying to sync the temp DB to disk in "
"MASTER <-> REPLICA synchronization: %s",
strerror(errno));
cancelReplicationHandshake();
return;
}
if (rename(server.repl_transfer_tmpfile,server.rdb_filename) == -1) {
serverLog(LL_WARNING,"Failed trying to rename the temp DB into dump.rdb in MASTER <-> REPLICA synchronization: %s", strerror(errno));
cancelReplicationHandshake();
return;
}
serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Flushing old data");
/* We need to stop any AOFRW fork before flusing and parsing
* RDB, otherwise we'll create a copy-on-write disaster. */
if(aof_is_enabled) stopAppendOnly();
signalFlushedDb(-1);
emptyDb(
-1,
server.repl_slave_lazy_flush ? EMPTYDB_ASYNC : EMPTYDB_NO_FLAGS,
replicationEmptyDbCallback);
/* Before loading the DB into memory we need to delete the readable
* handler, otherwise it will get called recursively since
* rdbLoad() will call the event loop to process events from time to
* time for non blocking loading. */
aeDeleteFileEvent(server.el,server.repl_transfer_s,AE_READABLE);
serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Loading DB in memory");
rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
if (rdbLoad(server.rdb_filename,&rsi) != C_OK) {
serverLog(LL_WARNING,"Failed trying to load the MASTER synchronization DB from disk");
cancelReplicationHandshake();
/* Re-enable the AOF if we disabled it earlier, in order to restore
* the original configuration. */
if (aof_is_enabled) restartAOFAfterSYNC();
return;
}
/* Final setup of the connected slave <- master link */
zfree(server.repl_transfer_tmpfile);
close(server.repl_transfer_fd);
replicationCreateMasterClient(server.repl_transfer_s,rsi.repl_stream_db);
server.repl_state = REPL_STATE_CONNECTED;
server.repl_down_since = 0;
/* After a full resynchroniziation we use the replication ID and
* offset of the master. The secondary ID / offset are cleared since
* we are starting a new history. */
memcpy(server.replid,server.master->replid,sizeof(server.replid));
server.master_repl_offset = server.master->reploff;
clearReplicationId2();
/* Let's create the replication backlog if needed. Slaves need to
* accumulate the backlog regardless of the fact they have sub-slaves
* or not, in order to behave correctly if they are promoted to
* masters after a failover. */
if (server.repl_backlog == NULL) createReplicationBacklog();
serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Finished with success");
/* Restart the AOF subsystem now that we finished the sync. This
* will trigger an AOF rewrite, and when done will start appending
* to the new file. */
if (aof_is_enabled) restartAOFAfterSYNC();
}
return;
error:
cancelReplicationHandshake();
return;
}
主服务器
主服务器则是对psync命令的执行和回复
/* SYNC and PSYNC command implemenation. */
void syncCommand(client *c) {
/* ignore SYNC if already slave or in monitor mode */
if (c->flags & CLIENT_SLAVE) return;
/* Refuse SYNC requests if we are a slave but the link with our master
* is not ok... */
if (server.masterhost && server.repl_state != REPL_STATE_CONNECTED) {
addReplySds(c,sdsnew("-NOMASTERLINK Can't SYNC while not connected with my master\r\n"));
return;
}
//当有数据要返回给客户端时，无法执行sync/psync
if (clientHasPendingReplies(c)) {
addReplyError(c,"SYNC and PSYNC are invalid with pending output");
return;
}
serverLog(LL_NOTICE,"Replica %s asks for synchronization",
replicationGetSlaveName(c));
//如果是psync，先尝试部分同步
if (!strcasecmp(c->argv[0]->ptr,"psync")) {
if (masterTryPartialResynchronization(c) == C_OK) {
server.stat_sync_partial_ok++;
return; /* No full resync needed, return. */
} else {
char *master_replid = c->argv[1]->ptr;
/* Increment stats for failed PSYNCs, but only if the
* replid is not "?", as this is used by slaves to force a full
* resync on purpose when they are not albe to partially
* resync. */
if (master_replid[0] != '?') server.stat_sync_partial_err++;
}
} else {
/* If a slave uses SYNC, we are dealing with an old implementation
* of the replication protocol (like redis-cli --slave). Flag the client
* so that we don't expect to receive REPLCONF ACK feedbacks. */
c->flags |= CLIENT_PRE_PSYNC;
}
//无法部分同步，只能全量同步
server.stat_sync_full++;
/* Setup the slave as one waiting for BGSAVE to start. The following code
* paths will change the state if we handle the slave differently. */
c->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
if (server.repl_disable_tcp_nodelay)
anetDisableTcpNoDelay(NULL, c->fd); /* Non critical if it fails. */
c->repldbfd = -1;
c->flags |= CLIENT_SLAVE;
listAddNodeTail(server.slaves,c);
//这里创建一个复制积压缓冲区，用于部分同步
if (listLength(server.slaves) == 1 && server.repl_backlog == NULL) {
/* When we create the backlog from scratch, we always use a new
* replication ID and clear the ID2, since there is no valid
* past history. */
changeReplicationId();
clearReplicationId2();
createReplicationBacklog();
}
// case1：当前有bgsave，其实RDB_CHILD_TYPE_DISK类型，尝试共用落盘RDB
if (server.rdb_child_pid != -1 &&
server.rdb_child_type == RDB_CHILD_TYPE_DISK)
{
/* Ok a background save is in progress. Let's check if it is a good
* one for replication, i.e. if there is another slave that is
* registering differences since the server forked to save. */
client *slave;
listNode *ln;
listIter li;
listRewind(server.slaves,&li);
while((ln = listNext(&li))) {
slave = ln->value;
if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END) break;
}
/* To attach this slave, we check that it has at least all the
* capabilities of the slave that triggered the current BGSAVE. */
if (ln && ((c->slave_capa & slave->slave_capa) == slave->slave_capa)) {
/* Perfect, the server is already registering differences for
* another slave. Set the right state, and copy the buffer. */
copyClientOutputBuffer(c,slave);
replicationSetupSlaveForFullResync(c,slave->psync_initial_offset);
serverLog(LL_NOTICE,"Waiting for end of BGSAVE for SYNC");
} else {
/* No way, we need to wait for the next BGSAVE in order to
* register differences. */
serverLog(LL_NOTICE,"Can't attach the replica to the current BGSAVE. Waiting for next BGSAVE for SYNC");
}
// case2：当前有bgsave，但是是RDB_CHILD_TYPE_SOCKET类型，等待下次bgsave
} else if (server.rdb_child_pid != -1 &&
server.rdb_child_type == RDB_CHILD_TYPE_SOCKET)
{
/* There is an RDB child process but it is writing directly to
* children sockets. We need to wait for the next BGSAVE
* in order to synchronize. */
serverLog(LL_NOTICE,"Current BGSAVE has socket target. Waiting for next BGSAVE for SYNC");
// case3：当前无bgsave
} else {
//无磁盘复制
if (server.repl_diskless_sync && (c->slave_capa & SLAVE_CAPA_EOF)) {
/* Diskless replication RDB child is created inside
* replicationCron() since we want to delay its start a
* few seconds to wait for more slaves to arrive. */
if (server.repl_diskless_sync_delay)
serverLog(LL_NOTICE,"Delay next BGSAVE for diskless SYNC");
}
//有磁盘复制
else {
/* Target is disk (or the slave is not capable of supporting
* diskless replication) and we don't have a BGSAVE in progress,
* let's start one. */
if (server.aof_child_pid == -1) {
startBgsaveForReplication(c->slave_capa);
} else {
serverLog(LL_NOTICE,
"No BGSAVE in progress, but an AOF rewrite is active. "
"BGSAVE for replication delayed");
}
}
}
return;
}
部分同步回复：masterTryPartialResynchronization
只要数据还在复制积压缓冲区就不用走全量同步，从机等着接收数据更新Offset即可。
int masterTryPartialResynchronization(client *c) {
long long psync_offset, psync_len;
char *master_replid = c->argv[1]->ptr;
char buf[128];
int buflen;
/* Parse the replication offset asked by the slave. Go to full sync
* on parse error: this should never happen but we try to handle
* it in a robust way compared to aborting. */
if (getLongLongFromObjectOrReply(c,c->argv[2],&psync_offset,NULL) !=
C_OK) goto need_full_resync;
/* Is the replication ID of this master the same advertised by the wannabe
* slave via PSYNC? If the replication ID changed this master has a
* different replication history, and there is no way to continue.
*
* Note that there are two potentially valid replication IDs: the ID1
* and the ID2. The ID2 however is only valid up to a specific offset. */
if (strcasecmp(master_replid, server.replid) &&
(strcasecmp(master_replid, server.replid2) ||
psync_offset > server.second_replid_offset))
{
/* Run id "?" is used by slaves that want to force a full resync. */
if (master_replid[0] != '?') {
if (strcasecmp(master_replid, server.replid) &&
strcasecmp(master_replid, server.replid2))
{
serverLog(LL_NOTICE,"Partial resynchronization not accepted: "
"Replication ID mismatch (Replica asked for '%s', my "
"replication IDs are '%s' and '%s')",
master_replid, server.replid, server.replid2);
} else {
serverLog(LL_NOTICE,"Partial resynchronization not accepted: "
"Requested offset for second ID was %lld, but I can reply "
"up to %lld", psync_offset, server.second_replid_offset);
}
} else {
serverLog(LL_NOTICE,"Full resync requested by replica %s",
replicationGetSlaveName(c));
}
goto need_full_resync;
}
//数据不在复制积压缓存区则全量同步
if (!server.repl_backlog ||
psync_offset < server.repl_backlog_off ||
psync_offset > (server.repl_backlog_off + server.repl_backlog_histlen))
{
serverLog(LL_NOTICE,
"Unable to partial resync with replica %s for lack of backlog (Replica request was: %lld).", replicationGetSlaveName(c), psync_offset);
if (psync_offset > server.master_repl_offset) {
serverLog(LL_WARNING,
"Warning: replica %s tried to PSYNC with an offset that is greater than the master replication offset.", replicationGetSlaveName(c));
}
goto need_full_resync;
}
//到这里表示可以部分同步
/* If we reached this point, we are able to perform a partial resync:
* 1) Set client state to make it a slave.
* 2) Inform the client we can continue with +CONTINUE
* 3) Send the backlog data (from the offset to the end) to the slave. */
c->flags |= CLIENT_SLAVE;
c->replstate = SLAVE_STATE_ONLINE;
c->repl_ack_time = server.unixtime;
c->repl_put_online_on_ack = 0;
listAddNodeTail(server.slaves,c);
/* We can't use the connection buffers since they are used to accumulate
* new commands at this stage. But we are sure the socket send buffer is
* empty so this write will never fail actually. */
//回复 +CONTINUE
if (c->slave_capa & SLAVE_CAPA_PSYNC2) {
buflen = snprintf(buf,sizeof(buf),"+CONTINUE %s\r\n", server.replid);
} else {
buflen = snprintf(buf,sizeof(buf),"+CONTINUE\r\n");
}
if (write(c->fd,buf,buflen) != buflen) {
freeClientAsync(c);
return C_OK;
}
//发送复制积压缓存区数据
psync_len = addReplyReplicationBacklog(c,psync_offset);
serverLog(LL_NOTICE,
"Partial resynchronization request from %s accepted. Sending %lld bytes of backlog starting from offset %lld.",
replicationGetSlaveName(c),
psync_len, psync_offset);
/* Note that we don't need to set the selected DB at server.slaveseldb
* to -1 to force the master to emit SELECT, since the slave already
* has this state from the previous connection with the master. */
refreshGoodSlavesCount();
return C_OK; /* The caller can return, no full resync needed. */
need_full_resync:
/* We need a full resync for some reason... Note that we can't
* reply to PSYNC right now if a full SYNC is needed. The reply
* must include the master offset at the time the RDB file we transfer
* is generated, so we need to delay the reply to that moment. */
return C_ERR;
}
注：每次命令的执行，都会将命令写入到复制积压缓存区server.repl_backlog和传播给子节点
（图中命令传播给子节点后面会用到，这里先圈出来）
全量同步回复：startBgsaveForReplication
int startBgsaveForReplication(int mincapa) {
int retval;
int socket_target = server.repl_diskless_sync && (mincapa & SLAVE_CAPA_EOF);
listIter li;
listNode *ln;
serverLog(LL_NOTICE,"Starting BGSAVE for SYNC with target: %s",
socket_target ? "replicas sockets" : "disk");
//准备rdb文件
rdbSaveInfo rsi, *rsiptr;
rsiptr = rdbPopulateSaveInfo(&rsi);
/* Only do rdbSave* when rsiptr is not NULL,
* otherwise slave will miss repl-stream-db. */
if (rsiptr) {
if (socket_target)
retval = rdbSaveToSlavesSockets(rsiptr);
else
retval = rdbSaveBackground(server.rdb_filename,rsiptr);
} else {
serverLog(LL_WARNING,"BGSAVE for replication: replication information not available, can't generate the RDB file right now. Try later.");
retval = C_ERR;
}
//出错情况
if (retval == C_ERR) {
serverLog(LL_WARNING,"BGSAVE for replication failed");
listRewind(server.slaves,&li);
while((ln = listNext(&li))) {
client *slave = ln->value;
if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
slave->replstate = REPL_STATE_NONE;
slave->flags &= ~CLIENT_SLAVE;
listDelNode(server.slaves,ln);
addReplyError(slave,
"BGSAVE failed, replication can't continue");
slave->flags |= CLIENT_CLOSE_AFTER_REPLY;
}
}
return retval;
}
//如果是有磁盘复制
if (!socket_target) {
listRewind(server.slaves,&li);
while((ln = listNext(&li))) {
client *slave = ln->value;
//找正在等待同步开始的从机回复+FULLRESYNC
if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
replicationSetupSlaveForFullResync(slave,
getPsyncInitialOffset());
}
}
}
if (retval == C_OK) replicationScriptCacheFlush();
return retval;
}
到这里，其实全量同步只回复了一个 +FULLRESYNC，那么rdb文件是怎么发送的呢？答案在serverCron定时任务（1242-1275行）。
全量同步rdb文件发送：sendBulkToSlave
一路跟踪
serverCron->backgroundSaveDoneHandler->backgroundSaveDoneHandlerDisk->updateSlavesWaitingBgsave->sendBulkToSlave
sendBulkToSlave
void sendBulkToSlave(aeEventLoop *el, int fd, void *privdata, int mask) {
client *slave = privdata;
UNUSED(el);
UNUSED(mask);
char buf[PROTO_IOBUF_LEN]; //每次发送16K
ssize_t nwritten, buflen;
//在发送rdb文件之前先发送文件长度
if (slave->replpreamble) {
nwritten = write(fd,slave->replpreamble,sdslen(slave->replpreamble));
if (nwritten == -1) {
serverLog(LL_VERBOSE,"Write error sending RDB preamble to replica: %s",
strerror(errno));
freeClient(slave);
return;
}
server.stat_net_output_bytes += nwritten;
sdsrange(slave->replpreamble,nwritten,-1);
if (sdslen(slave->replpreamble) == 0) {
sdsfree(slave->replpreamble);
slave->replpreamble = NULL;
/* fall through sending data. */
} else {
return;
}
}
//每次发送16K数据
lseek(slave->repldbfd,slave->repldboff,SEEK_SET);
buflen = read(slave->repldbfd,buf,PROTO_IOBUF_LEN);
if (buflen <= 0) {
serverLog(LL_WARNING,"Read error sending DB to replica: %s",
(buflen == 0) ? "premature EOF" : strerror(errno));
freeClient(slave);
return;
}
if ((nwritten = write(fd,buf,buflen)) == -1) {
if (errno != EAGAIN) {
serverLog(LL_WARNING,"Write error sending DB to replica: %s",
strerror(errno));
freeClient(slave);
}
return;
}
slave->repldboff += nwritten;
server.stat_net_output_bytes += nwritten;
if (slave->repldboff == slave->repldbsize) {
close(slave->repldbfd);
slave->repldbfd = -1;
aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE);
putSlaveOnline(slave);
}
}
至此全量同步的rdb文件就发送了。
那么问题来了，主服务器从生成rdb文件到从服务器接收并加载rdb文件肯定也是一个比较耗时的操作，在这段时间主服务器新执行的命令是如何同步给从服务器的呢？
其实Redis和客户端通信也好，和从库通信也好，都会给分配一个内存buffer进行数据交互。客户端是一个client，从库也是一个client，所有数据交互都通过这个buffer进行。先把数据写到这个buffer中，然后再把buffer中的数据发到client socket中通过网络发送出去。而在主从复制中这个buffer叫做replication buffer（复制缓存区）
与命令写入复制积压缓存区一样（分析复制积压缓存区时一起圈出来的地方），在call方法执行的时候，（当需要时）也会传播给子节点，即写入复制缓存区。
一直跟踪addReplyBulk
可以发现由于此时replicas还未处于SLAVE_STATE_ONLINE状态以及ack不等于0，未加入待处理列表clients_pending_write，这些命令只是先积攒在replicas的缓存区中。
那么何时发送呢？自然是在rdb文件发送完后再发送。回到sendBulkToSlave函数，可以看到在函数的最后会判断rdb文件是否发送完，如果发送完了就会执行putSlaveOnline将从节点逻辑上线，并创建对应套接字的文件事件发送积攒的命令。
分析到这里其实就不难猜出，无论是全量复制还是部分复制，在复制完成后都会将从节点ONLINE，之后的命令才得以发出去。全量复制刚已经分析过了，现在来验证部分复制完成后是否会将从节点状态置为ONLINE以及将ONLINE_ON_ACK置为0呢？
果然在masterTryPartialResynchronization函数的最后可以发现
c->replstate = SLAVE_STATE_ONLINE;
c->repl_put_online_on_ack = 0;
至此，slaveof命令流程就已经看完了。最后来一波总结
slaveof流程：
1、从服务器保存主服务器ip和port
3、ping 命令检查网络是否联通
4、AUTH 命令身份检验
5、从服务器发送自身ip和port给主服务器
6、PSYNC 开始同步
7、命令传播
增量/全量的决定逻辑：
增量同步：
全量同步：
主从复制除上面的复制流程外，还有一个心跳检测主要用于检查主从节点之间的连接状况。
心跳检测
从服务器默认每秒发从一次replconf ack命令进行心跳检测。其目的在于
1：检测主从服务连接状态
2：检测命令的丢失
3：辅助实现min-slaves
参考文献：《Redis设计与实现》黄健宏著、redis-5.0.14源码
![[论文评析]基于人体姿态识别的立定跳远 动作智能评估系统](https://img-blog.csdnimg.cn/3ed885f5f3f64d0b9940146f5a58a515.png)
