Question

I have the following SQL code:

select val.PersonNo,
       val.event_time,
       clg.number_dialed
  from vicidial_agent_log val
         join
       call_log           clg on date_add('1970-01-01 02:00:00', interval clg.uniqueid second) = val.event_time
 order by val.event_time desc
 limit 100;

which executes and returns rows in less than 1 second. However, if I change the straight join to a left outer:

select val.PersonNo,
       val.event_time,
       clg.number_dialed
  from vicidial_agent_log val
         left outer join
       call_log           clg on date_add('1970-01-01 02:00:00', interval clg.uniqueid second) = val.event_time
 order by val.event_time desc
 limit 100;

the query runs forever and uses ~100% of the server’s CPU.

I ran explain on both queries and the first hits the event_time index on vicidial_agent_log, while the second ignores all indexes. There is an index on call_log.uniqueid.

vicidial_agent_log contains ~41,000 rows, call_log contains ~43,000.

So my question is – why is MySQL not hitting the indexes I’ve defined, is there any way to force it to do so, and if not, how can I make this query run at an acceptable speed?

edit

Full solution:

select val.PersonNo,
       val.event_time,
       cl.number_dialed
  from vicidial_agent_log val
         left outer join
       (select date_add('1970-01-01 02:00:00', interval clg.uniqueid second) as 'converted_date',
               number_dialed
          from call_log clg) cl ON cl.converted_date = val.event_time
 order by val.event_time desc
 limit 100;

Answer 1

Using functions in JOIN or WHERE clauses will always cause havoc with indexes. Example:

DATE_ADD('1970-01-01 02:00:00', INTERVAL clg.uniqueid SECOND)

The database is using the uniqueid index to lookup the values for the conversion, not the comparison to the event_time column in your case. If this were Oracle with PLW errors turned on, you’d be notified about potentially converting away from the data type.

Such situations should always be handled before the comparison, which means using an inline view to perform the conversion and then join to that result column. IE:

JOIN (SELECT DATE_ADD('1970-01-01 02:00:00', INTERVAL clg.uniqueid SECOND) 'converted_date'
        FROM CALL_LOG clg) cl ON cl.converted_date = val.event_time