[BUG] It is wrong to rewrite config `rename-command` and lead kvrocks to fail to restart

[iushas]

Describe

If your kvrocks config file has multiple rename-command, kvrocks would generate duplicate configuration items after executing config rewrite, whatever automatically or manually, and leads kvrocks to fail to restart.

It is introduced in #272

one example

rename-command debug                               290033_dont_run_debug_rudely
rename-command bgsave                              290033_dont_run_bgsave_rudely
rename-command shutdown                            290033_dont_run_shutdown_rudely
rename-command keys                                290033_dont_run_keys_rudely
rename-command debug                               290033_dont_run_debug_rudely

there are two debug rename-commands, so there will be the following error logs.

* Failed to load config, err: at line: #L19, err: No such command in rename-command

How to fix

• you can find the duplicate configuration items, and remove duplicate items, only keep one remame-command for one command.
• of course, you can cherry-pick this commit 6c8534a (Fix config rewrite overwrite the first rename-command with the last one #429 )
• use 2.0.6 version (coming soon)

[iushas]

复现过程：主从切换后，新的master会发生上述问题

环境：version 2.0.4 ，一主一从 ， 接了sentinel监控

1、查看配置正常
root@redis-kvsenior02-290041:/home/redis/data/conf# cat kvrocks.conf

# server
bind                           7.33.107.212 10.91.38.110 127.0.0.1

# rename


# autofailover
slaveof 7.33.106.212 6379
slave-serve-stale-data         yes
slave-read-only                yes



# rename
rename-command flushall                            290041_dont_run_flushall_rudely
rename-command flushdb                             290041_dont_run_flushdb_rudely
rename-command bgsave                              290041_dont_run_bgsave_rudely
rename-command shutdown                            290041_dont_run_shutdown_rudely
rename-command keys                                290041_dont_run_keys_rudely
rename-command debug                               290041_dont_run_debug_rudely
rename-command compact                             290041_dont_run_compact_rudely

# server
port                                               6379
timeout                                            0
workers                                            1
...

2、手动执行slaveof no one 和切换 。

• 这里有两个问题， slaveof no one后发现角色还是slave。
• 再次执行slaveof no one ，变为了master

root@redis-kvsenior02-290041:/home/redis/data/conf# redis-cli
cann't open conf:/home/redis/etc/redis.conf
127.0.0.1:6379> slaveof no one
OK
127.0.0.1:6379> info replication
# Replication
role:slave
master_host:7.33.106.212
master_port:6379
master_link_status:up
master_sync_unrecoverable_error:no
master_sync_in_progress:0
master_last_io_seconds_ago:2
slave_repl_offset:252419
slave_priority:100
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252419,lag=0
master_repl_offset:252419
127.0.0.1:6379> info replication
# Replication
role:slave
master_host:7.33.106.212
master_port:6379
master_link_status:up
master_sync_unrecoverable_error:no
master_sync_in_progress:0
master_last_io_seconds_ago:0
slave_repl_offset:252419
slave_priority:100
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252419,lag=0
master_repl_offset:252419
127.0.0.1:6379> info replication
# Replication
role:slave
master_host:7.33.106.212
master_port:6379
master_link_status:up
master_sync_unrecoverable_error:no
master_sync_in_progress:0
master_last_io_seconds_ago:1
slave_repl_offset:252421
slave_priority:100
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252421,lag=0
master_repl_offset:252421
127.0.0.1:6379>
127.0.0.1:6379> info replication
# Replication
role:slave
master_host:7.33.106.212
master_port:6379
master_link_status:up
master_sync_unrecoverable_error:no
master_sync_in_progress:0
master_last_io_seconds_ago:2
slave_repl_offset:252421
slave_priority:100
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252421,lag=0
master_repl_offset:252421
127.0.0.1:6379> info replication
# Replication
role:slave
master_host:7.33.106.212
master_port:6379
master_link_status:up
master_sync_unrecoverable_error:no
master_sync_in_progress:0
master_last_io_seconds_ago:2
slave_repl_offset:252421
slave_priority:100
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252421,lag=0
master_repl_offset:252421
127.0.0.1:6379> slaveof no one
OK
127.0.0.1:6379>
127.0.0.1:6379>
127.0.0.1:6379>
127.0.0.1:6379> info replication
# Replication
role:master
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252423,lag=0
master_repl_offset:252423
127.0.0.1:6379> info replication
# Replication
role:master
connected_slaves:1
slave0:ip=7.33.106.212,port=6379,offset=252425,lag=0
master_repl_offset:252425
127.0.0.1:6379> set xtk sd
OK
127.0.0.1:6379> get xtk
"sd"
127.0.0.1:6379>

3、再次看配置，发现rename-command flushall 290041_dont_run_flushall_rudely变了，原来的master的配置文件没有变
root@redis-kvsenior02-290041:/home/redis/data/conf# cat kvrocks.conf

# server
bind 7.33.107.212 10.91.38.110 127.0.0.1

# rename


# autofailover
slaveof no one
slave-serve-stale-data yes
slave-read-only yes



# rename
rename-command compact                             290041_dont_run_compact_rudely
rename-command flushdb                             290041_dont_run_flushdb_rudely
rename-command bgsave                              290041_dont_run_bgsave_rudely
rename-command shutdown                            290041_dont_run_shutdown_rudely
rename-command keys                                290041_dont_run_keys_rudely
rename-command debug                               290041_dont_run_debug_rudely
rename-command compact                             290041_dont_run_compact_rudely

# server
port 6379
timeout 0
w...

@git-hulk @ShooterIT

[ShooterIT]

two debug rename-commands

Failed to load config, err: at line: #L19, err: No such command in rename-command
Version: 2.0.4 @8374af2

It is not a bug, redis have similar behavior, since the command is already renamed, you can rename renamed command again, i.e. rename-command 290033_dont_run_debug_rudely debug2

[iushas]

rename-command

I use redis to test， the issue dsnt happens

[iushas]

two debug rename-commands
Failed to load config, err: at line: #L19, err: No such command in rename-command
Version: 2.0.4 @8374af2

It is not a bug, redis have similar behavior, since the command is already renamed, you can rename renamed command again, i.e. rename-command 290033_dont_run_debug_rudely debug2

which redis version you said!

[ShooterIT]

rename-command debug                               290041_dont_run_debug_rudely
rename-command debug                               290041_dont_run_debug_rudely

use this config, redis could start?

[iushas]

you have not understand the question， after switch master， the new master's config will auto change to that， isn‘t changed by manul.

use that file, certainly cont start!

[ShooterIT]

I just answer your first question instead of the second

[iushas]

I just answer your first question instead of the second

这两个说的是一个问题，第二个是复现过程，切换后，配置文件会自动改变。 麻烦确认下吧

[ShooterIT]

2、手动执行slaveof no one 和切换 。

这里有两个问题， slaveof no one后发现角色还是slave。
再次执行slaveof no one ，变为了master

第一次执行slaveof no one后，日志中有没有出现 [replication] Stopped
以及第二次执行呢

[git-hulk]

是两个问题:

1. config rewrite 之后不对导致 rename command 问题重启有问题，这个已经找到原因
2. slaveof no one 失败是另外一个问题

[iushas]

每次执行slaveof no one，应该是有出现[replication] stopped

流程应该是这样的：
初始 A -> B

1、 在B执行slaveof no one
2、在A执行slaveof B， 理想情况下应该是变成 B-> A. 但是可能由于sentinel 主动又把 B变成了A的slave。 所以在A执行slaveof B后， 通过info replcation 居然看到了 A <=> B的情况
3、然后再次在B执行 slaveof no one ，架构改成了B -> A