When the slave sets maintenance, there is a problem with the status change of proxysql.

[peterpan104]

This is the state when repman parameters, repman and proxysql are first start:

## repman parameter
proxy-servers-read-on-master = false
proxy-servers-read-on-master-no-slave = true

I set 172.24.44.182 as maintenance,
the status of repman and proxysql is as follows:

Then set 172.24.47.188 as maintenance,
the status of repman and proxysql is as follows:

Now set 172.24.47.188 as online,
the status of repman and proxysql is as follows:

then set 172.24.43.182 as online,
the status of repman and proxysql is as follows:

proxy-servers-read-on-master-no-slave seems to have no effect, but proxysql status is normal

Now I trigger the switchover,
the status of repman and proxysql is as follows:

I set 172.24.43.182 as maintenance,
the status of repman and proxysql is as follows:

Then set 172.24.47.188 maintenance,
the status of repman and proxysql is as follows:

You can see that 172.24.43.182 is online in the proxysql status!

Now I restore the status to after switchover:

I set 172.24.47.188 as maintenance,
the status of repman and proxysql is as follows:

Then set 172.24.43.183 as maintenance,
the status of repman and proxysql is as follows:

At this time, proxysql status is still normal.

Now set 172.24.47.188 as online,
the status of repman and proxysql is as follows:

You can see that 172.24.43.182 is online in the proxysql status!

I tried setting proxysql-bootstrap=true,
It just allows proxy-servers-read-on-master-no-slave to work,
However, the status of proxysql will still be the same as the above test. Can this problem be fixed?

[svaroqui]

Hi peterpan104 We have to think on it a lot, the solution was never made to allow to put master in maintenance it is just made to avoid that situation to do that and to to preserve the writer traffic . A maintenance would be done on slave one by one , once all slaves have finish maintenance one is promote as master and the old master now a slave or a broken slave will receive the maintenance and get reintroduce as the master

[svaroqui]

Sorry please correct I set 172.24.43.182 as maintenance on the first lines as it lead to miss understand of the issue , we need to reproduce as what you are trying to do make sens, still i would never stop all my slaves to put all read traffic on a single node

[svaroqui]

Is this scripting on your side or is it reproductible manually by using the same steps from the gui ?

[peterpan104]

@svaroqui
Sorry, I described it wrong. The above maintenance is all for slave actions.

Under normal conditions, all slaves will not be maintenance.
I am simulating the processing after all slaves are failed.
The above tests don't use any monitor scripts. The operation can be reproduced through the gui.
I will find that it is because my script has set maintenance, But I found that proxysql's changes were not what I expected.

thanks for your reply.

[caffeinated92]

Do you enable proxysql bootstrap?

[peterpan104]

HI @caffeinated92
I didn't enable proxysql bootstrap for the above test, but I also enabled this parameter test and the result was the same. The only difference is that when there is no readable machine, master will be used as readable.

[caffeinated92]

Added bug fix in v2.3.44 but it will need bootstrap to ON. Since we don't want the script to add master in reader list except with acknowledgement

[peterpan104]

Hi @caffeinated92
I'm using v2.3.44 and setting bootstrap to ON but the problem is not solved.
When I set maintenance on both slaves, one will always show online in proxysql.
Can this issue be opened and fixed?

[svaroqui]

Hi we are investigating and found some issues in the new logic that maintenance state could differ from proxy to proxy