Question

In SQL (specifically MySQL, but the question is generic enough), what is the most efficient way to query time-series data when I have multiple tables across disjoint time ranges? For example, if my tables are as follows:

router1_20090330( unixtime integer unsigned,                    iface1_in integer unsigned,                   iface1_out integer unsigned ) router1_20090331( unixtime integer unsigned,                    iface1_in integer unsigned,                   iface1_out integer unsigned ) 

and so on, with (say) 1 minute samples of the in and out traffic on interface1, giving 86400 records per day table.

I want the query to represent something like

SELECT CAST(unixtime/3600 as unsigned) AS hour,         sum(iface1_in), sum(iface1_out)    from router1_20090330 *then* router1_20090331   group by hour order by hour; 

i.e. I would like both tables to be queried and aggregated, but one after the other instead of being joined.

Answer 1

You want to UNION the two (or more) tables:

SELECT CAST(unixtime/3600 as unsigned) AS hour,     SUM(iface1_in), SUM(iface1_out) FROM (     SELECT * FROM router1_20090330     UNION ALL     SELECT * FROM router1_20090331 ) x GROUP BY hour ORDER BY hour 

Note that the ‘ALL’ after UNION tells the query to keep duplicate rows (if the same values appear in both tables – unlikely in your scenario). I usually use UNION ALL because I think it avoids the overhead of testing the data for duplicate rows.