Orchestrator relocation with gtid_mode = ON should be more flexible

[sjmudd]

I turned off pseudo-gtid on a cluster where gtid_mode = ON. Orchestrator sees this and was happy. However, when I tried to relocate some slaves under a different intermediate master it complained as being too complex. This is I think wrong.

Enclosed is a santised screen image of the error message I saw.

I was trying to drag the 2 slaves under server 2001 under server 1004.
With gtid this should be easy and I should be able to pretty much move any box under any box.

The only exception I see to this is if the slaves are missing gtid events which have been purged on the new master: that won't work so probably needs checking, but if the gtid event list is consistent this should work.

The example shown is using the latest code from github (or at last it's less than a week old).

Allowing this more flexible relocation would be good if GTID is being used.

Feel free to poke me if you need my configuration or further information.

[shlomi-noach]

I see this frequently where AUTO_POSITION=0 in SHOW SLAVE STATUS. Which means yours servers are GTID-enabled, but are not actually using GTID for replication.

orchestrator sees this as not-using-GTID and will not assume it is allowed to turn AUTO_POSITION on.

Can you please check if that is the case on your cluster?

I think this might be an interesting warning orchestrator can provide you with: "hey, you said you want to use GTID but actually your are not".