mysql8 gtid #434 test intermittent rejoin failure

[nyxneuf]

SRM - v2.2.24
MySQL8 - M-S

Test scenario

1. M - S -> master db stop and slave is new master!
2. old master db start -> S - M
3. S - M -> master db stop and slave is new master!
4. old master db start -> M - S

repeat the above process!
intermittent rejoin failure!

[nyxneuf]

Similarly, I use the code
there was no problem!

replication-manager/cluster/srv_rejoin.go

Lines 675 to 682 in 932673c

	ss, errss := server.GetSlaveStatus(server.ReplicationSourceName)
	if errss != nil {
	server.ClusterGroup.LogPrintf(LvlInfo, "Failed to check if server was using GTID %s", errss)
	return false
	}
	server.ClusterGroup.LogPrintf(LvlInfo, "Rejoin server using GTID %s", ss.UsingGtid.String)

errss -> err after modification

	ss, err := server.GetSlaveStatus(server.ReplicationSourceName)
	if err != nil {
		server.ClusterGroup.LogPrintf(LvlInfo, "Failed to check if server was using GTID %s", err)
		return false
	}
	server.ClusterGroup.LogPrintf(LvlInfo, "Rejoin server using GTID %s", ss.UsingGtid.String)

[nyxneuf]

I found it!

https://github.com/signal18/replication-manager/blob/develop/cluster/srv_rejoin.go#L675-L682

not the code above!

https://github.com/signal18/replication-manager/blob/develop/cluster/srv_rejoin.go#L684

[Before]
if crash.FailoverIOGtid == nil {

[After]
if crash.FailoverIOGtid != nil {

=======================

I applied and tested it and there was no problem.

What does this code mean?

And

https://github.com/signal18/replication-manager/blob/develop/cluster/srv_rejoin.go#L376

You need to check that the code is also correct!

[svaroqui]

Re, I think you are missing one point here . There are time when you can not rejoin the old master , and replication-manager is detecting this scenario if it's not possible to rejoin you need a full state transfert via mysqldump , restoring a backup or a snapshot to put the old master back in time before the crash of the old leader. you need to install the same server package on the replication-manager server and provide the path to the tools needed for state transfert.

autorejoin-mysqldump = true
backup-mysqlbinlog-path = "/Users/apple/mysql/bin/mysqlbinlog"
backup-mysqldump-path = "/Users/apple/mysql/bin/mysqldump"
backup-mysqlclient-path = "/Users/apple/mysql/bin/mysql"

[nyxneuf]

Rejoin fails even if there is no change in DB data.
Do I have to use mysqldump for all DB data?

[svaroqui]

Ok i'll give it try you say multiple times failover rejoin

[svaroqui]

I'm testing always with 3 nodes may be it make a difference

[svaroqui]

To explain the code when the leader is crashed , replication-manager record the position of the crash by looking show slave status on the candidate master that is still a replica at this time . It record gtid_executed and translate uuid to hash to get similar to mariadb GTID an record it in property FailoverIOGtid that is the name of the MariaDB counter part .

[nyxneuf]

yes! multiple times failover rejoin.
I'm sorry to bother you.
Please check!

[svaroqui]

You don't bother me at all , you're help is very valuable, would prefer still to get issues without images but with log files it's more easy to dig into it

[nyxneuf]

Thank you.
I'm not a developer. I'm an infra engineer.
It's hard to understand Golang. It is difficult to use English. So I use a translator.
I think replication-manager is great.
I'm glad it helped. :)

[svaroqui]

Found it MySQL can print carriage return in GTID output of show slave status i 've pushed a fix and will make a new release soon , thanks again for founding this !