pastbedti.me » PostgreSQL

Grouping a timestamp field by date in Ruby On Rails / PostgreSQL

Mathias Stjernström — Sun, 08 Nov 2009 00:55:09 +0000

Recently I’ve been working a lot with dates and tonight I had a hard time finding information about how to group a model by date when the model only have a timestamp column.

Lets pretend we have a user model with a created_at attribute which is stored as a timestamp.

If we want to plot a simple graph showing signups per date this is the way to extract them:

1	User.count(:group => "DATE(created_at)")

Thats all there is to it. From this you will get an OrderedHash containing the date and count of users for each date.

One small thing to remember if you are using PostgreSQL. You will need the DATE() function in any ORDER/SELECT statements because PostgreSQL will only select/order by fields thats in the GROUP BY statement and this is a pretty healthy behavior that many databases (no names here) don’t care about. Think about it. If you are grouping all users by date and use * as selector how would the database know which username or email address to display?

This is what happens if you try to group by created_at without DATE() ->

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

>> User.count(:order => 'created_at DESC', :group => ["DATE(created_at)"])
ActiveRecord::StatementInvalid: PGError: ERROR: column "users.created_at" must appear in the GROUP BY clause or be used in an aggregate function
: SELECT count(*) AS count_all, DATE(created_at) AS date_created_at FROM "users" GROUP BY DATE(created_at) ORDER BY created_at DESC
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/abstract_adapter.rb:212:in `log'
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/postgresql_adapter.rb:507:in `execute'
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/postgresql_adapter.rb:985:in `select_raw'
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/postgresql_adapter.rb:972:in `select'
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/abstract/database_statements.rb:7:in `select_all_without_query_cache'
from /tmp/app/vendor/rails/activerecord/lib/active_record/connection_adapters/abstract/query_cache.rb:62:in `select_all'
from /tmp/app/vendor/rails/activerecord/lib/active_record/calculations.rb:255:in `execute_grouped_calculation'
from /tmp/app/vendor/rails/activerecord/lib/active_record/calculations.rb:132:in `calculate'
from /tmp/app/vendor/rails/activerecord/lib/active_record/calculations.rb:130:in `catch'
from /tmp/app/vendor/rails/activerecord/lib/active_record/calculations.rb:130:in `calculate'
from /tmp/app/vendor/rails/activerecord/lib/active_record/calculations.rb:48:in `count'
from (irb):18

And this is the right way to do it:

1
2

>> User.count(:order => 'DATE(created_at) DESC', :group => ["DATE(created_at)"])
=> #22, "2009-10-25"=>4, "2009-11-04"=>8, "2009-10-15"=>9, "2009-10-26"=>16, "2009-11-05"=>9, "2009-10-16"=>193, "2009-10-27"=>14, "2009-11-06"=>9, "2009-10-17"=>49, "2009-10-28"=>15, "2009-11-07"=>6, "2009-10-18"=>36, "2009-10-29"=>8, "2009-10-19"=>116, "2009-10-30"=>15>

As you may notice the Hash does not look ordered. But if you loop over it you will get them in order:

1
2
3
4
5
6
7
8
9
10
11
12

>> User.count(:order => 'DATE(created_at) DESC', :group => ["DATE(created_at)"]).each {|u| puts "#{u[0]} -> #{u[1]}" }
2009-11-07 -> 6
2009-11-06 -> 9
2009-11-05 -> 9
2009-11-04 -> 8
2009-11-03 -> 14
2009-11-02 -> 20
2009-11-01 -> 10
2009-10-31 -> 6
2009-10-30 -> 15
2009-10-29 -> 8
..

And thats how you group a timestamp field with just its date part.

Cheers!

PostgreSQL logging with syslog-ng

Mathias Stjernström — Mon, 15 Dec 2008 22:25:16 +0000

Today I where about to analyze a PostgreSQL database with the wonderful pgFouine. On the machine syslog-ng where installed as logging daemon and I have never worked with PostgreSQL logging and syslog-ng. This post is about how to get them to work together.

pgFouine is pretty simple to setup and the process is where well documented at their website (http://pgfouine.projects.postgresql.org/tutorial.html).

However syslog-ng is very different if you are used to “normal” syslog daemon.

The first thing you have to do is setup PostgreSQL as the instructions says on the website. The next step is to tell syslog-ng to log you postgres data to a separate file.

The first thing to create in syslog-ng.conf is a destination. Destination blocks is used to send logs somewhere.

By adding the following row we tell syslog-ng that there is a destination called postgres and that we want this data in /var/log/pgsql.

1	destination postgres { file("/var/log/pgsql"); };

Now we are ready to add a filter to caputre our PostgreSQL data. Filters can both be used to include or exclude data from you logs. The following row in your syslog-conf captures all data sent to local0 or whatever you called it in postgresql.conf

1	filter f_postgres { facility(local0); };

It’s time to join our destination with our filter and the following line tells syslog-ng to join our source, destination and filter together.

1	log { source(src); filter(f_postgres); destination(postgres); };

Put these lines in your syslog-ng.conf and restart syslog-ng and also PostgreSQL and you will have logging in a separate file!

Next step is to write a little guide to pgFouine

Cheers!

Interesting use of SSD drives with PostgreSQL

Mathias Stjernström — Wed, 06 Aug 2008 10:47:19 +0000

These are two interesting articles/benchmarks about how you can use SSD drives for storing postgres indexes on them.

The low low seek times on a SSD drives make it a perfect candidate for the indexes.

Postgresql table sizes

Mathias Stjernström — Sun, 16 Mar 2008 18:40:10 +0000

PostgreSQL has lots and lots of sweet internals but its not always easy to know how to find them. Here is an simple way of getting the disk size of a PostgreSQL table.

From within the database of choice.

Run the following query to get some useful table statistics.

1	SELECT relname, reltuples, relpages * 8 / 1024 AS "MB" FROM pg_class ORDER BY relpages DESC;

Block size

If you use another block size than 8kb you have do multiply relpages with your own block size. To find your block size you can use pg_controldata which is a command-line utility to display control information of a PostgreSQL database cluster.

Run it with your datadir as first parameter pg_controldata ./data/ and look for Database block size in the output.

Database query result

The outout gives you:

relname: The name of the table.
reltuples: Estimated numbers of rows.
MB: Table size in Mbytes.

1
2
3
4
5
6
7
8
9
10
11
12

relname | reltuples | MB
-----------------------------------+-------------+-------
history_uint | 1.685e+08 | 15396
history_uint_1 | 1.685e+08 | 12279
history | 1.40719e+08 | 8405
history_1 | 1.40719e+08 | 7153
trends | 1.03077e+07 | 929
trends_pkey | 1.03077e+07 | 423
items | 4698 | 83
history_str | 536599 | 53
history_str_1 | 536599 | 27
events | 166602 | 15

We can now see that history_uint is the biggest table and holds an estimate of 168,572,082 rows and takes about 15GB on disk.

Be aware that these numbers only gets updated when you run vacuum.

I have more PostgreSQL internals to come!