CentOS 7.4 x86_64

please provide output of

SHOW CREATE TABLE Tickets;

SHOW CREATE TABLE Groups;

SHOW CREATE TABLE CachedGroupMembers;

It looks like difference in this execution plans is not between key vs unique key,
but in using apid_status key vs apid_lastupdated, so whether optimizer choose index on column, used in ORDER BY clause or not.
Please see Sergei Petrunia's explanation of this problem here: MDEV-14621

Okay. Is there a reasonably-clean way to influence the optimizer to get the desired order? Unfortunately, I probably can't change the query directly (RT's query builder doesn't expose the whole query).

Thanks.

Hi Chris,
I do not have the exact solution for this case, but idea is to add some coverage index, like ALTER TABLE Tickets ADD INDEX newkey (Status, Lastupdated), so the optimizer will use it instead of the others.
Btw you have 2 identical indexes on column (`Owner`) in table Tickets (is it by design or just accidentally)

Sorry, it took me a bit to get back to this... yes, that one index is purely an accident. Thanks for noting it! I did add the suggested index, and that did measurably help my query. Thanks!

Really happy to see your message!)

Name and EmailAddress differ in both NULLness and Uniqueness. Plus a CROSS JOIN is involved.

Is this the order of actions being taken by the Optimizer to decide which index to use?

Step 1: Does the optimizer see UNIQUE together with an "=" test and realize that there can be no more than one row from Users by Name? But when analyzing the Email case, it assumes there could be more than 1 row?

Step 2: Does the Optimizer now decide whether to use the ORDER BY indexing versus some filtering for WHERE.

Step 3: It probes the table and realized there is only one Email row. I see that the table Users first (in EXPLAIN); this tells me that it did discover only one row.

@Chris – I tell users "If an index can entirely handle the WHERE + GROUP BY + ORDER BY, then the LIMIT may be usefully be cut short." If not, then there is the risk that some WHERE clauses can lead to a full table scan before discovering that there are not enough rows to satisfy the LIMIT.

Because of that, I don't like to even suggest an index on the ORDER BY column(s) unless the INDEX is good enough. (I am not talking about "covering", but, for example, all '=' and no ranges in the WHERE.)