postgresql/contrib/tips

HOW TO get Apache to log to PostgreSQL
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 Note: contain of files 'httpconf.txt' and 'apachelog.sql' are below this
       text. 


First, this is intended mostly as a starting point, an example of how to do it.

The file 'httpconf.txt' is commented and contains two example lines to make
this work, a custom log format, and a line that sends the log data to psql.
I think that the comments in this file should be sufficient.

The file 'apachelog.sql' is a little SQL to create the table and grant 
permissions to it.

You must:

1. Already have 'nobody' (or what ever your web server runs as) as a valid 
   PostgreSQL user.

2. Create the database to hold the log, (example 'createdb www_log')

3. Edit the file 'apachelog.sql' and change the name of the table to what
   ever you used in step 2.  ALSO if need be, change the name 'nobody' in
	the grant statement.

4. As an appropriate user (postgres is ok), do 'psql www_log < apachelog.sql'.
   This should have created the table and granted access to it.

5. SAVE A COPY OF YOUR httpd.conf !!!

6. Edit httpd.conf, add the two lines in the example file as appropriate, 
   IN THE ORDER IN WHICH THEY APPEAR.  This is simple for a single server, 
	but a little more complex for virtual hosts, but if you set up virtual 
	hosts, then you should know were to put these lines.

7. Down and restart your httpd.  I do it on Red Hat 4.1 like this:
   /etc/rc.d/init.d/httpd.init stop
	then
   /etc/rc.d/init.d/httpd.init start
   OR I understand you can send it a signal 16 like 'kill -16 <pid>' and do it.

8. I should be working, query the web server about 30 or more times then look 
   in the db and see what you have, if nothing then query the web server 
	30 or 50 more time and then check.  If still nothing, look in the server's 
	error log to see what is going on.  But you should have data.

NOTES:
The log data is cached some where, and so will not appear INSTANTLY in the 
database!  I found that it took around 30 queries of the web server, then 
many rows are written to the db at once.

ALSO, I leave it up to you to create any indexes on the table that you want.

The error log can (*I think*) also be sent to PostgreSQL in the same fashion.

At some point in the future, I will be writing some PHP to interface to this
and generate statistical type reports, so check my site once and a while if
you are interested it this.

Terry Mackintosh <terry@terrym.com>
http://www.terrym.com

Have fun ... and remember, this is mostly just intended as a stating point, 
not as a finished idea.

--- apachelog.sql : ---

drop table access;
CREATE TABLE access (host char(200), ident char(200), authuser char(200), accdate timestamp, request char(500), ttime int2, status int2, bytes int4) archive = none;
grant all on access to nobody;

--- httpconf.txt: ---

# This is mostly the same as the default, except for no square brakets around
# the time or the extra timezone info, also added the download time, 3rd from
# the end, number of seconds.

LogFormat "insert into access values ( '%h', '%l', '%u', '%{%d/%b/%Y:%H:%M:%S}t', '%r', %T, %s, %b );"

# The above format ALMOST eleminates the need to use sed, except that I noticed
# that when a frameset page is called, then the bytes transfered is '-', which
# will choke the insert, so replaced it with '-1'.

TransferLog '| su -c "sed \"s/, - );$/, -1 );/\" | /usr/local/pgsql/bin/psql www_log" nobody'