The Main Thing
Make sure you have a compound index on first_name
and last_name
. Otherwise, it's really easy to end up doing a full table scan regardless of how you approach this. So if you don't already have one, create one:
CREATE INDEX users_firstandlast ON users(first_name, last_name);
Syntax Options
Once that index is in place, you have some options:
Option 1: As Willis Blackburn said, repeat the CONCAT
in your WHERE
clause (because AS
doesn't create a name you can use in the WHERE
clause):
SELECT CONCAT(first_name, ' ', last_name) as fullname
FROM users
WHERE CONCAT(first_name, ' ', last_name) LIKE '%doe%';
Use EXPLAIN
to check in your specific situation, but in my tests it says it uses the compound index, even though you're using a function in the WHERE
clause.
Option 2: In this particular case, you can always just use two LIKE
s in your WHERE
clause:
SELECT CONCAT(first_name, ' ', last_name) as fullname
FROM users
WHERE first_name LIKE '%doe%' or last_name LIKE '%doe%';
Again this can make use of the compound index (whereas it won't make use of individual indexes on the first_name
and last_name
columns -- it would if you weren't leading with a wildcard, but according to EXPLAIN
[and your mileage may vary, always check], in that case it goes with the table scan).
Option 3 In his answer, Andy says you can use HAVING
for this. My read of the MySQL manual suggests it will first build the result set, and only then apply HAVING
at the very end before sending it to the client, and so I'd be leery of this. But, in my quick and dirty tests, EXPLAIN
tells me that if you have the compound index I mentioned above, the HAVING
version does an index search, not a table scan. If your tests with real data bear that out, that may be a good option for you. This use of HAVING
in this way is a MySQL extension (not standard), but then again, so is CONCAT
so we're already into MySQL-specific stuff. :-) But again, double-check in your real life environment.
Conclusion
Create the index if you don't already have it, then I'd go with Option 2 if it's remotely a possibility; otherwise, option 1 unless you can find (or Andy can provide) a reference for the HAVING
thing not building a massive interim result set (it would be really cool, if non-standard, if it didn't). Regardless, check with EXPLAIN
, and test, in your specific environment.