Query Overlapping Periods from Usage Information stored in SQL Database

Question

I have a table in a PostgreSQL database which tracks usage of various resources. The (simplified) Schema of the table is that each row has a ResourceID, StartTime Timestamp and EndTime Timestamp. Each row in the table represents a timespan in which the resource was in use, so the table might look like: (Note, timestamps also include dates, removed below for clarity)

ResourceID  StartTime   EndTime
---------------------------------------
1           12:30:00    12:45:00
1           12:48:25    12:50:22
2           12:32:50    12:33:44

The database would have perhaps a thousand different resources tracked and a few million rows in the table. I've recently received a feature request for a new report which details time periods in that a group of resources are all in use, so the query might be "Between 12:00 and 15:00, display all the time periods when resources 1,2,5,8 and 12 were all in use". In addition,the query should have a "Minimum Idle" period, which a resource needs to be idle for before being considered idle, (example: If Minimum Idle is 2 seconds, a resource in use 12:00:00-12:01:00 and 12:01:01 to 12:02:00 would not be considered to have any idle time,even though technically it was not in use for 1 second).

The output of the query would be a list of starttime/endtimes of all times when all the queried resources were in use. From that point, I need to compute some statistics on that dataset, which won't be a problem for me, but I'm at a loss on how to efficiently create that dataset from the original table. If necessary I can log additional information to the database at insert time, and if not for the arbitary resource subset requirement, I could just create a table of all the idle times then, but with 1000 different resources and any possible combination of 1-1000 resources in a query, that seems excessive as only a very small number of combinations will ever be reported on.

Thanks in advance for any help or insights.

Answer 1

For usage periods

Use a range type from PostgreSQL 9.2 and check for overlap across whatever periods you have. You can take multiple overlapping segments so you can whittle down ranges progressively.

This is not quite trivial so I am afraid I don't have a simple example.

For idle periods:

I think you'd want to do this with some sort of interval type (the new types in 9.2 would be helpful here) or create a similar type you could use for query purposes. Note that where I have done this, it has not been trivial.

The second thing you'd want to do is create a custom aggregate to compare and add intervals. It would need to return an array of these types. Finally you will need to be able to iteratively run through differences.

There isn't a simple solution here. The code involved is more complex than you'd probably like, and it will be more than one would typically get from an answer here. There's a significant amount of logic involved and design effort involved. It is quite possible, but it isn't extremely simple.