Redis强大的备份同步工具Redis-Shake
redis-shake是阿里云Redis&MongoDB团队开源的用于redis数据同步的工具。下载地址：这里。
基本功能
redis-shake是我们基于redis-port基础上进行改进的一款产品。它支持解析、恢复、备份、同步四个功能。以下主要介绍同步sync。
- 恢复restore：将RDB文件恢复到目的redis数据库。
- 备份dump：将源redis的全量数据通过RDB文件备份起来。
- 解析decode：对RDB文件进行读取，并以json格式解析存储。
- 同步sync：支持源redis和目的redis的数据同步，支持全量和增量数据的迁移，支持从云下到阿里云云上的同步，也支持云下到云下不同环境的同步，支持单节点、主从版、集群版之间的互相同步。需要注意的是，如果源端是集群版，可以启动一个RedisShake，从不同的db结点进行拉取，同时源端不能开启move slot功能；对于目的端，如果是集群版，写入可以是1个或者多个db结点。
- 同步rump：支持源redis和目的redis的数据同步，仅支持全量的迁移。采用scan和restore命令进行迁移，支持不同云厂商不同redis版本的迁移。
- 指定Key恢复:filter.key.whitelist可以过滤指定前辍的key，但是只适用于
restore,syncandrump三种模式。
基本原理
redis-shake的基本原理就是模拟一个从节点加入源redis集群，首先进行全量拉取并回放，然后进行增量的拉取（通过psync命令）。如下图所示：
如果源端是集群模式，只需要启动一个redis-shake进行拉取，同时不能开启源端的move slot操作。如果目的端是集群模式，可以写入到一个结点，然后再进行slot的迁移，当然也可以多对多写入。
目前，redis-shake到目的端采用单链路实现，对于正常情况下，这不会成为瓶颈，但对于极端情况，qps比较大的时候，此部分性能可能成为瓶颈，后续我们可能会计划对此进行优化。另外，redis-shake到目的端的数据同步采用异步的方式，读写分离在2个线程操作，降低因为网络时延带来的同步性能下降。
高效性
全量同步阶段并发执行，增量同步阶段异步执行，能够达到毫秒级别延迟（取决于网络延迟）。同时，我们还对大key同步进行分批拉取，优化同步性能。
监控
用户可以通过我们提供的restful拉取metric来对redis-shake进行实时监控：curl 127.0.0.1:9320/metric。
校验
如何校验同步的正确性？可以采用我们开源的redis-full-check，具体原理可以参考这篇博客。
支持
支持2.8-5.0版本的同步。
支持codis。
支持云下到云上，云上到云上，云上到云下（阿里云目前支持主从版），其他云到阿里云等链路，帮助用户灵活构建混合云场景。
开发中的功能
断点续传。支持断开后按offset恢复，降低因主备切换、网络抖动造成链路断开重新同步拉取全量的性能影响。
说明
多活支持的支持，需要依赖内核的能力，目前单单依赖通道层面无法解决该问题。redis内核层面需要提供gtid的概念，以及CRDT的解决冲突的方式才可以。
参数解释：
启动示例（以sync举例）：
vinllen@ ~/redis-shake/bin$ ./redis-shake -type=sync -conf=../conf/redis-shake.conf
conf路径下的redis-shake.conf存放的是配置文件，其内容及部分中文注释如下
# this is the configuration of redis-shake.
# if you have any problem, please visit https://github.com/alibaba/RedisShake/wiki/FAQ
# current configuration version, do not modify.
# 当前配置文件的版本号，请不要修改该值。
conf.version = 1
# ------------------------------------------------------
# id
id = redis-shake
# log file，日志文件，不配置将打印到stdout (e.g. /var/log/redis-shake.log )
log.file =
# log level: "none", "error", "warn", "info", "debug". default is "info".
log.level = info
# pid path，进程文件存储地址（e.g. /var/run/)，不配置将默认输出到执行下面,
# 注意这个是目录，真正的pid是`{pid_path}/{id}.pid`
pid_path =
# pprof port.
system_profile = 9310
# restful port, set -1 means disable, in `restore` mode RedisShake will exit once finish restoring RDB only if this value
# is -1, otherwise, it'll wait forever.
# restful port，查看metric端口, -1表示不启用，如果是`restore`模式，只有设置为-1才会在完成RDB恢复后退出，否则会一直block。
http_profile = 9320
# parallel routines number used in RDB file syncing. default is 64.
# 启动多少个并发线程同步一个RDB文件。
parallel = 32
# source redis configuration.
# used in `dump`, `sync` and `rump`.
# source redis type, e.g. "standalone" (default), "sentinel" or "cluster".
# 1. "standalone": standalone db mode.
# 2. "sentinel": the redis address is read from sentinel.
# 3. "cluster": the source redis has several db.
# 4. "proxy": the proxy address, currently, only used in "rump" mode.
# 源端redis的类型，支持standalone，sentinel，cluster和proxy四种模式，注意：目前proxy只用于rump模式。
source.type = standalone
# ip:port
# the source address can be the following:
# 1. single db address. for "standalone" type.
# 2. ${sentinel_master_name}:${master or slave}@sentinel single/cluster address, e.g., mymaster:master@127.0.0.1:26379;127.0.0.1:26380, or @127.0.0.1:26379;127.0.0.1:26380. for "sentinel" type.
# 3. cluster that has several db nodes split by semicolon(;). for "cluster" type. e.g., 10.1.1.1:20331;10.1.1.2:20441.
# 4. proxy address(used in "rump" mode only). for "proxy" type.
# 源redis地址。对于sentinel或者开源cluster模式，输入格式为"master名字:拉取角色为master或者slave@sentinel的地址"，别的cluster
# 架构，比如codis, twemproxy, aliyun proxy等需要配置所有master或者slave的db地址。
source.address = 127.0.0.1:20441
# password of db/proxy. even if type is sentinel.
source.password_raw = 123456
# auth type, don't modify it
source.auth_type = auth
# tls enable, true or false. Currently, only support standalone.
# open source redis does NOT support tls so far, but some cloud versions do.
source.tls_enable = false
# input RDB file.
# used in `decode` and `restore`.
# if the input is list split by semicolon(;), redis-shake will restore the list one by one.
# 如果是decode或者restore，这个参数表示读取的rdb文件。支持输入列表，例如：rdb.0;rdb.1;rdb.2
# redis-shake将会挨个进行恢复。
source.rdb.input = local
# the concurrence of RDB syncing, default is len(source.address) or len(source.rdb.input).
# used in `dump`, `sync` and `restore`. 0 means default.
# This is useless when source.type isn't cluster or only input is only one RDB.
# 拉取的并发度，如果是`dump`或者`sync`，默认是source.address中db的个数，`restore`模式默认len(source.rdb.input)。
# 假如db节点/输入的rdb有5个，但rdb.parallel=3，那么一次只会
# 并发拉取3个db的全量数据，直到某个db的rdb拉取完毕并进入增量，才会拉取第4个db节点的rdb，
# 以此类推，最后会有len(source.address)或者len(rdb.input)个增量线程同时存在。
source.rdb.parallel = 0
# for special cloud vendor: ucloud
# used in `decode` and `restore`.
# ucloud集群版的rdb文件添加了slot前缀，进行特判剥离: ucloud_cluster。
source.rdb.special_cloud =
# target redis configuration. used in `restore`, `sync` and `rump`.
# the type of target redis can be "standalone", "proxy" or "cluster".
# 1. "standalone": standalone db mode.
# 2. "sentinel": the redis address is read from sentinel.
# 3. "cluster": open source cluster (not supported currently).
# 4. "proxy": proxy layer ahead redis. Data will be inserted in a round-robin way if more than 1 proxy given.
# 目的redis的类型，支持standalone，sentinel，cluster和proxy四种模式。
target.type = standalone
# ip:port
# the target address can be the following:
# 1. single db address. for "standalone" type.
# 2. ${sentinel_master_name}:${master or slave}@sentinel single/cluster address, e.g., mymaster:master@127.0.0.1:26379;127.0.0.1:26380, or @127.0.0.1:26379;127.0.0.1:26380. for "sentinel" type.
# 3. cluster that has several db nodes split by semicolon(;). for "cluster" type.
# 4. proxy address. for "proxy" type.
target.address = 127.0.0.1:20551
# password of db/proxy. even if type is sentinel.
target.password_raw =
# auth type, don't modify it
target.auth_type = auth
# all the data will be written into this db. < 0 means disable.
target.db = -1
# tls enable, true or false. Currently, only support standalone.
# open source redis does NOT support tls so far, but some cloud versions do.
target.tls_enable = false
# output RDB file prefix.
# used in `decode` and `dump`.
# 如果是decode或者dump，这个参数表示输出的rdb前缀，比如输入有3个db，那么dump分别是:
# ${output_rdb}.0, ${output_rdb}.1, ${output_rdb}.2
target.rdb.output = local_dump
# some redis proxy like twemproxy doesn't support to fetch version, so please set it here.
# e.g., target.version = 4.0
target.version =
# use for expire key, set the time gap when source and target timestamp are not the same.
# 用于处理过期的键值，当迁移两端不一致的时候，目的端需要加上这个值
fake_time =
# how to solve when destination restore has the same key.
# rewrite: overwrite.
# none: panic directly.
# ignore: skip this key. not used in rump mode.
# used in `restore`, `sync` and `rump`.
# 当源目的有重复key，是否进行覆写
# rewrite表示源端覆盖目的端。
# none表示一旦发生进程直接退出。
# ignore表示保留目的端key，忽略源端的同步key。该值在rump模式下没有用。
key_exists = none
# filter db, key, slot, lua.
# filter db.
# used in `restore`, `sync` and `rump`.
# e.g., "0;5;10" means match db0, db5 and db10.
# at most one of `filter.db.whitelist` and `filter.db.blacklist` parameters can be given.
# if the filter.db.whitelist is not empty, the given db list will be passed while others filtered.
# if the filter.db.blacklist is not empty, the given db list will be filtered while others passed.
# all dbs will be passed if no condition given.
# 指定的db被通过，比如0;5;10将会使db0, db5, db10通过, 其他的被过滤
filter.db.whitelist =
# 指定的db被过滤，比如0;5;10将会使db0, db5, db10过滤，其他的被通过
filter.db.blacklist =
# filter key with prefix string. multiple keys are separated by ';'.
# e.g., "abc;bzz" match let "abc", "abc1", "abcxxx", "bzz" and "bzzwww".
# used in `restore`, `sync` and `rump`.
# at most one of `filter.key.whitelist` and `filter.key.blacklist` parameters can be given.
# if the filter.key.whitelist is not empty, the given keys will be passed while others filtered.
# if the filter.key.blacklist is not empty, the given keys will be filtered while others passed.
# all the namespace will be passed if no condition given.
# 支持按前缀过滤key，只让指定前缀的key通过，分号分隔。比如指定abc，将会通过abc, abc1, abcxxx
filter.key.whitelist =
# 支持按前缀过滤key，不让指定前缀的key通过，分号分隔。比如指定abc，将会阻塞abc, abc1, abcxxx
filter.key.blacklist =
# filter given slot, multiple slots are separated by ';'.
# e.g., 1;2;3
# used in `sync`.
# 指定过滤slot，只让指定的slot通过
filter.slot =
# filter lua script. true means not pass. However, in redis 5.0, the lua
# converts to transaction(multi+{commands}+exec) which will be passed.
# 控制不让lua脚本通过，true表示不通过
filter.lua = false
# big key threshold, the default is 500 * 1024 * 1024 bytes. If the value is bigger than
# this given value, all the field will be spilt and write into the target in order. If
# the target Redis type is Codis, this should be set to 1, please checkout FAQ to find
# the reason.
# 正常key如果不大，那么都是直接调用restore写入到目的端，如果key对应的value字节超过了给定
# 的值，那么会分批依次一个一个写入。如果目的端是Codis，这个需要置为1，具体原因请查看FAQ。
# 如果目的端大版本小于源端，也建议设置为1。
big_key_threshold = 524288000
# enable metric
# used in `sync`.
# 是否启用metric
metric = true
# print in log
# 是否将metric打印到log中
metric.print_log = false
# sender information.
# sender flush buffer size of byte.
# used in `sync`.
# 发送缓存的字节长度，超过这个阈值将会强行刷缓存发送
sender.size = 104857600
# sender flush buffer size of oplog number.
# used in `sync`. flush sender buffer when bigger than this threshold.
# 发送缓存的报文个数，超过这个阈值将会强行刷缓存发送，对于目的端是cluster的情况，这个值
# 的调大将会占用部分内存。
sender.count = 4095
# delay channel size. once one oplog is sent to target redis, the oplog id and timestamp will also
# stored in this delay queue. this timestamp will be used to calculate the time delay when receiving
# ack from target redis.
# used in `sync`.
# 用于metric统计时延的队列
sender.delay_channel_size = 65535
# enable keep_alive option in TCP when connecting redis.
# the unit is second.
# 0 means disable.
# TCP keep-alive保活参数，单位秒，0表示不启用。
keep_alive = 0
# used in `rump`.
# number of keys captured each time. default is 100.
# 每次scan的个数，不配置则默认100.
scan.key_number = 50
# used in `rump`.
# we support some special redis types that don't use default `scan` command like alibaba cloud and tencent cloud.
# 有些版本具有特殊的格式，与普通的scan命令有所不同，我们进行了特殊的适配。目前支持腾讯云的集群版"tencent_cluster"
# 和阿里云的集群版"aliyun_cluster"，注释主从版不需要配置，只针对集群版。
scan.special_cloud =
# used in `rump`.
# we support to fetching data from given file which marks the key list.
# 有些云版本，既不支持sync/psync，也不支持scan，我们支持从文件中进行读取所有key列表并进行抓取：一行一个key。
scan.key_file =
# limit the rate of transmission. Only used in `rump` currently.
# e.g., qps = 1000 means pass 1000 keys per second. default is 500,000(0 means default)
qps = 200000
# enable resume from break point, please visit xxx to see more details.
# 断点续传开关
resume_from_break_point = false
# ----------------splitter----------------
# below variables are useless for current open source version so don't set.
# replace hash tag.
# used in `sync`.
replace_hash_tag = false
开源地址
redis-shake: https://github.com/aliyun/redis-shake
