Hi Kirk,
We have a (near) real time data database for instrument observations from our research vessels. All observations (summarised to one minute intervals - the actual raw data is in netCDF, this database makes for easier access & meets most users needs) go into a single table, with other tables for metadata about platforms, instruments, etc. Now approaching 350m records, so reasonably substantial.
Underlying optimisations include
partitioned readings table, with a separate partition for each year (now 23 years)
clustered index on timestamp for the previous years partitions.
largeish filesystem block size - tested to work well with the clustered index & small size records)
These generally make a big difference to performance. To address one issue, much like yours, where some users want hourly data for a year, some want daily data for 10 years & some want 1 minute data for the last month (& some, no doubt, want one minute data for 20+ years!) I introduced an integer column called timer. This value is set according to the time (not date) of each record.
Along the lines of (from memory) :an even no of minutes after the hour is 2, 5 minutes is 4, 10 minutes is 8, 15 minute is 16, 30 minutes is 32, 60 minutes is 64, 6 hourly is 128, 12:00 AM is 256 & 12:00PM is 512. When any timestamp is in more than one category (eg: 12:00 is all of even, 5, 15m 30m 60 minutes), the timer value is set to the largest appropriate one.
So a request for:
1 minute data is select from table;
2 minute data is select from table where timer >=2 and timer !=15 and timer !=4;
hourly data is select from table where timer >=64 and timer != 15 and timer != 4;
etc
5 & 15 minute add a bit of complexity, but we gave the users what they wanted. This has worked well for years now, & we have an internal web (mapserver/openlayers based) application allowing users to visualise & download their selected data - they choose from an interval pick list & the SQL is hidden. Some extra enhancements are the automatic collation of lat & lon gps readings into a Postgis point for each reading record, & the automatic aggregation of daily points into daily track lines, so the track for any selected set of dates can easily be displayed on a map (the platforms are mobile vessels - not fixed sites)
You might adapt some of these ideas for your use case?
Cheers
Brent Wood
Programme leader: Environmental Information Delivery
NIWA
DDI: +64 (4) 3860529
________________________________________
From: pgsql-general-owner@xxxxxxxxxxxxxx [pgsql-general-owner@xxxxxxxxxxxxxx] on behalf of Kirk Wythers [wythe001@xxxxxxx]
Sent: Tuesday, February 05, 2013 5:58 AM
To: pgsql-general@xxxxxxxxxxxxxx
Subject: Fwd: partial time stamp query
Thanks. That worked great! Now I am trying to aggregate these same fifteen minute to hourly. I have tried using date_trunk:
date_trunc('hour', derived_tsoil_fifteen_min_stacked.time2),
but date_truck only seems to aggriage the timestamp. I thought I could use
AVG(derived_tsoil_fifteen_min_stacked.value)
in combination with date_trunk, but I still get 15 minute values, not the hourly average from the four 15 minute records.
rowid date_truck time2 site canopy plot variable name value avg
2010-07-07_00:00:00_b4warm_a 2010-07-07 00:00:00 1 2010-07-07 00:00:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 21.06 21.0599994659424
2010-07-07_00:15:00_b4warm_a 2010-07-07 00:00:00 1 2010-07-07 00:15:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.96 20.9599990844727
2010-07-07_00:30:00_b4warm_a 2010-07-07 00:00:00 1 2010-07-07 00:30:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.88 20.8799991607666
2010-07-07_00:45:00_b4warm_a 2010-07-07 00:00:00 1 2010-07-07 00:45:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.8 20.7999992370605
2010-07-07_01:00:00_b4warm_a 2010-07-07 01:00:00 1 2010-07-07 01:00:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.72 20.7199993133545
2010-07-07_01:15:00_b4warm_a 2010-07-07 01:00:00 1 2010-07-07 01:15:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.64 20.6399993896484
2010-07-07_01:30:00_b4warm_a 2010-07-07 01:00:00 1 2010-07-07 01:30:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.55 20.5499992370605
2010-07-07_01:45:00_b4warm_a 2010-07-07 01:00:00 1 2010-07-07 01:45:00 cfc closed a2 tsoil_sc tsoil_avg1_sc 20.47 20.4699993133545
I was tying to get two records out of this set, with the 'avg" column representing the mean of the first and last four of each 15 minute records.
Perhaps date_trunk only works for the timestamp?
On Feb 4, 2013, at 8:50 AM, Misa Simic <misa.simic@xxxxxxxxx<mailto:misa.simic@xxxxxxxxx>> wrote:
WHERE derived_tsoil_fifteen_min_stacked.time2::date = '2010-07-07'::date
On Monday, February 4, 2013, Kirk Wythers wrote:
I am trying to write a query that grabs one particular day from a timestamp column. The data are ordered in 15 minute chunks like this:
2010-07-07 12:45:00
2010-07-07 13:00:00
2010-07-07 13:15:00
2010-07-07 13:30:00
etc…
If I wanted all records from july 7th 2010, I would expect 4 x 24 = 96 records per day.
I have tried the '=' operator, like this
WHERE derived_tsoil_fifteen_min_stacked.time2 = '2010-07-07*'
but that grabs nothing, and using the '~' operator grabs everything with a 2010 or 07 in it… in other words all days from July of 2010.
Any suggestions would be much appreciated.
--
Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
--
Please consider the environment before printing this email.
NIWA is the trading name of the National Institute of Water & Atmospheric Research Ltd.
--
Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
