System configuration
====================

Apache configuration
--------------------

Verify that mod_python has been configured: in your apache2.conf you should have a line like 

.. sourcecode:: apache

    LoadModule python_module modules/mod_python.so

Get the needed apache configuration with the command

.. sourcecode:: bash

	python $SANET_HOME/manage.py deploy apache

check the configuration (in particular pathnames) and insert it into your Apache configuration file.
It should be something like

.. sourcecode:: apache

	DocumentRoot /var/www/

	# Uncomment if you want to redirect the root of your webserver
	# RedirectMatch ^/$ /sanet/

	# Uncomment the following if you have mod_deflate
	# SetOutputFilter DEFLATE 
	# BrowserMatch ^Mozilla/4 gzip-only-text/html
	# BrowserMatch ^Mozilla/4\.0[678] no-gzip
	# BrowserMatchMSIE !no-gzip !gzip-only-text/html 

	SetEnv SANET_HOME /usr/local/sanet

	<LocationMatch "(/static/\|/media/)">
		SetHandler None
	</LocationMatch>

	<LocationMatch "/sanet/">
		SetHandler python-program
		PythonHandler django.core.handlers.modpython
		SetEnv DJANGO_SETTINGS_MODULE settings
		SetEnv PYTHON_EGG_CACHE /var/python-eggs/
		PythonDebug Off
		PythonPath "['/usr/local/sanet'] + sys.path"
	</LocationMatch>

You must then link your /media in DocumentRoot to Django media files,
and /static to $SANET_HOME/static

.. sourcecode:: bash

    # ln -s /usr/local/sanet/django/contrib/admin/media/ /var/www/
    # ln -s /usr/local/sanet/static /var/www/


Create `/var/python-eggs` directory and let it be writable by uid that executes apache daemon process.


Postgres tuning
---------------

SANET has been developed on PostgreSQL, so MySQL compatibility is not tested.
Here we discuss some tuning actions in order to make PostgreSQL behave smarter.

General considerations
^^^^^^^^^^^^^^^^^^^^^^

There are some parameters in `postgresql.conf` that we suggest
to adapt in order to suite the application needs.

**USE THE FOLLOWING INFORMATION AT YOUR OWN RISK !**
They are just some considerations and you have to evaluate them by each installation,
depending on how much you take care of ACID properties and your ability to recover
from off-line backups and tolerate lost of some history info.

After changing this file you have to stop the poller via `etc/rc.sanet stop`,
restart postgres and then launch again the poller with `etc/rc.sanet start`.

+ Max connections: if you have a lot of threads you would like to raise the default value to at least `thread num + httpd thread daemon (usually apache) + some other`. If you have no strict resources limit, We suggest to set it to `(thread num + httpd thread) *2`. Not dangerous.::

    max_connections: 200 or what you need
    shared_buffers: max_connections*16kB (p.e. 30MB)
    max_fsm_pages: min max_fsm_relations*16, 6 bytes each (p.e. 350000)

+ Autovacuum: you need it to perform some garbage collections. Not dangerous.::

    vacuum_cost_delay = 100
    track_counts = on
    autovacuum = on
    log_autovacuum_min_duration = 10 # (check postgresql logfile that autovacuum is actually doing some work)

+ Transaction parameters: tune sync writing to disk. If you disable
  synchronous commmit, the risk is to lose some transactions if postgres
  crashes, but you save a bunch of I/O operations since operations are aggregated
  and delayed. DANGEROUS.::

    synchronous_commit = off #immediate fsync at commit
    wal_writer_delay = 10000ms # 1-10000 milliseconds
    commit_delay = 100000# range 0-100000, in microseconds
    commit_siblings = 5# range 1-1000

+ Disable fsync: if you have **REALLY NO HOPE ANYMORE** you can disable fsync, and raise your risk to have a corrupted database. **DANGEROUS**. If you use this options pay attention to save CLI configuration each time you update it. If you disable fsync, previous settings of transaction parameters have no effect anymore:::

    fsync = off 


If you have a lot of RAM
^^^^^^^^^^^^^^^^^^^^^^^^

If you have at least 2GB RAM (better more) and over 6000 targets
and measures, you can try the following settings.
After you have restarted it all check that swap is not used:
it not useful to give plenty of RAM to postgres if you use swap.
Remember to add `echo 250000000 > /proc/sys/kernel/shmmax` to your `rc.postgres`::

    # SANET SETTINGS:
    # Put in rc.postgres something like: echo 250000000 > /proc/sys/kernel/shmmax
    # Remember to set noatime and dirnoatime
    shared_buffers = 20000
    vacuum_cost_delay = 100
    track_counts = on
    autovacuum = on
    log_autovacuum_min_duration = 10
    checkpoint_segments = 20
    checkpoint_timeout = 900
    wal_buffers = 64
    autovacuum = on
    max_fsm_pages = 2048000
    max_fsm_relations = 10000
    wal_buffers = 256
    # work_mem = 32768
    # maintenance_work_mem = 524288
    work_mem = 100MB
    maintenance_work_mem = 128MB
    synchronous_commit = off
    wal_writer_delay = 10000ms
    commit_delay = 100000
    commit_siblings = 5


System considerations
---------------------

RAM and swap
^^^^^^^^^^^^

In order to monitor successfully (not too much slowly) a medium network
which has almost 3000 checks, your machine should be provided with around
2GB RAM. A swap partition is also suggested.

If you do not have a swap dedicated partition, you can create a swapfile
with the procedure:

.. sourcecode:: bash

	helicopter:~# dd if=/dev/zero of=/swapfile count=1024 bs=1M
	helicopter:~# mkswap /swapfile
	helicopter:~# swapon /swapfile