Complex SQL query suggestions please - mysql

Question

I have three tables with schema as below:
Table: Apps
| ID (bigint) | USERID (Bigint)| START_TIME (datetime) |
-------------------------------------------------------------
| 1 | 13 | 2013-05-03 04:42:55 |
| 2 | 13 | 2013-05-12 06:22:45 |
| 3 | 13 | 2013-06-12 08:44:24 |
| 4 | 13 | 2013-06-24 04:20:56 |
| 5 | 13 | 2013-06-26 08:20:26 |
| 6 | 13 | 2013-09-12 05:48:27 |
Table: Hosts
| ID (bigint) | APPID (Bigint)| DEVICE_ID (Bigint) |
-------------------------------------------------------------
| 1 | 1 | 1 |
| 2 | 2 | 1 |
| 3 | 1 | 1 |
| 4 | 3 | 3 |
| 5 | 1 | 4 |
| 6 | 2 | 3 |
Table: Usage
| ID (bigint) | APPID (Bigint)| HOSTID (Bigint) | Factor (varchar) |
-------------------------------------------------------------------------------------
| 1 | 1 | 1 | Low |
| 2 | 1 | 3 | High |
| 3 | 2 | 2 | Low |
| 4 | 3 | 4 | Medium |
| 5 | 1 | 5 | Low |
| 6 | 2 | 2 | Medium |
Now if put is userid, i want to get the count of rows of table rows for each month (of all app) for each "Factor" month wise for the last 6 months.
If a DEVICE_ID appears more than once in a month (based on START_TIME, based on joining Apps and Hosts), only the latest rows of Usage (based on combination of Apps, Hosts and Usage) be considered for calculating count.
Example output of the query for the above example should be: (for input user id=13)
| MONTH | USAGE_COUNT | FACTOR |
-------------------------------------------------------------
| 5 | 0 | High |
| 6 | 0 | High |
| 7 | 0 | High |
| 8 | 0 | High |
| 9 | 0 | High |
| 10 | 0 | High |
| 5 | 2 | Low |
| 6 | 0 | Low |
| 7 | 0 | Low |
| 8 | 0 | Low |
| 9 | 0 | Low |
| 10 | 0 | Low |
| 5 | 1 | Medium |
| 6 | 1 | Medium |
| 7 | 0 | Medium |
| 8 | 0 | Medium |
| 9 | 0 | Medium |
| 10 | 0 | Medium |
How is this calculated?
For Month May 2013 (05-2013), there are two Apps from table Apps
In table Hosts , these apps are associated with device_id's 1,1,1,4,3
For this month (05-2013) for device_id=1, the latest value of start_time is: 2013-05-12 06:22:45 (from tables hosts,apps), so in table Usage, look for combination of appid=2&hostid=2 for which there are two rows one with factor Low and other Medium,
For this month (05-2013) for device_id=4, by following same procedure we get one entry i.e 0 Low
Similarly all the values are calculated.
To get the last 6 months via query i'm trying to get it with the following:
SELECT MONTH(DATE_ADD(NOW(), INTERVAL aInt MONTH)) AS aMonth
FROM
(
SELECT 0 AS aInt UNION SELECT -1 UNION SELECT -2 UNION SELECT -3 UNION SELECT -4 UNION SELECT -5
)
Please check sqlfiddle: http://sqlfiddle.com/#!2/55fc2

Answer 1

Because the calculation you're doing involves the same join multiple times, I started by creating a view.
CREATE VIEW `app_host_usage`
AS
SELECT a.id "appid", h.id "hostid", u.id "usageid",
a.userid, a.start_time, h.device_id, u.factor
FROM apps a
LEFT OUTER JOIN hosts h ON h.appid = a.id
LEFT OUTER JOIN `usage` u ON u.appid = a.id AND u.hostid = h.id
WHERE a.start_time > DATE_ADD(NOW(), INTERVAL -7 MONTH)
The WHERE condition is there because I made the assumption that you don't want July 2005 and July 2006 to be grouped together in the same count.
With that view in place, the query becomes
SELECT months.Month, COUNT(DISTINCT device_id), factors.factor
FROM
(
-- Get the last six months
SELECT (MONTH(NOW()) + aInt + 11) % 12 + 1 "Month" FROM
(SELECT 0 AS aInt UNION SELECT -1 UNION SELECT -2 UNION SELECT -3 UNION SELECT -4 UNION SELECT -5) LastSix
) months
JOIN
(
-- Get all known factors
SELECT DISTINCT factor FROM `usage`
) factors
LEFT OUTER JOIN
(
-- Get factors for each device...
SELECT
MONTH(start_time) "Month",
device_id,
factor
FROM app_host_usage a
WHERE userid=13
AND start_time IN (
-- ...where the corresponding usage row is connected
-- to an app row with the highest start time of the
-- month for that device.
SELECT MAX(start_time)
FROM app_host_usage a2
WHERE a2.device_id = a.device_id
GROUP BY MONTH(start_time)
)
GROUP BY MONTH(start_time), device_id, factor
) usageids ON usageids.Month = months.Month
AND usageids.factor = factors.factor
GROUP BY factors.factor, months.Month
ORDER BY factors.factor, months.Month
which is insanely complicated, but I've tried to comment explaining what each part does. See this sqlfiddle: http://sqlfiddle.com/#!2/5c871/1/0