Editing Traffic accounting with iptables

Suppose you need to know how much traffic your [[container]]s eat. It can be easily done
using iptables.

== Situation description ==
Let's consider the very simple situation: one container with one IP address on the [[Hardware Node]]
with only one network interface. To be more exact, assume that [[container]] ID is <tt>200</tt>, the IP address of the [[HN]]
is <tt>192.168.0.56</tt>, the network interface name is <tt>eth0</tt>, and the IP address of the [[container]] is <tt>192.168.0.117</tt>.

You wish to know how many bytes container 200 eats. One more assumption is that there are no iptables rules
on HN now. All these assumption are only for clarity!

== Solution ==
Almost any traffic that goes to and from a container can be catched by FORWARD chain of iptables module in [[container0]],
thus we add such rules:
<pre>
# iptables -A FORWARD -s 192.168.0.117
# iptables -A FORWARD -d 192.168.0.117
</pre>
It means that all traffic forwarded to IP 192.168.0.117 and from IP 192.168.0.117 will be accounted.
To obtain current traffic usage of container you can issue the command:
<pre>
# iptables -nv -L FORWARD
Chain FORWARD (policy ACCEPT 243 packets, 28089 bytes)
 pkts bytes target     prot opt in     out     source               destination
    8   832            all  --  *      *       192.168.0.117        0.0.0.0/0
   15  1052            all  --  *      *       0.0.0.0/0            192.168.0.117
</pre>
'''Bytes''' column is the column we need. It's worth saying, that restarting a container doesn't affect accounting,
it remains right. But if you restart your [[hardware node]], all the rules and consequently statistics are dropped.
So it is recommended to
* run some cron job that dumps statistics to some file
* add init script that creates iptables rules on [[HN]] start.

If you want to process the results with a script it is useful to use the "-x" or "--exact" option of iptables
<pre>
# iptables -nvx -L FORWARD
</pre>
You will get the exact value of the packet and byte counters, instead of only the rounded number in K’s (multiples of 1000) M’s (multiples of 1000K) or G’s (multiples  of  1000M).

As is easy to see, it's not per-container statistic, but rather per-IP statistic. Thus you must be careful
then changing container IP addresses, otherwise you'll get mess of results.

By saying ''almost any traffic'' I mean that traffic between a [[container]] and [[container0]] is not accounted by rules above.
Not sure if it can be useful for anybody, but to account such traffic these rules are needed:
<pre>
iptables -I INPUT 1 -i venet0 -d 192.168.0.117
iptables -I OUTPUT 1 -o venet0 -s 192.168.0.117
</pre>

To observe results:
<pre>
# iptables -nvx -L INPUT
Chain INPUT (policy ACCEPT 542 packets, 63745 bytes)
 pkts bytes target     prot opt in     out     source               destination
   35  4533            all  --  venet0 *       0.0.0.0/0            192.168.0.117
# iptables -nvx -L OUTPUT
Chain OUTPUT (policy ACCEPT 247 packets, 27847 bytes)
 pkts bytes target     prot opt in     out     source               destination
   48  4724            all  --  *      venet0  192.168.0.117        0.0.0.0/0
</pre>
If you need to zero counters this command works:
<pre>
# iptables -Z
</pre>
The disadvantage is that by doingit this way you zero all counters in all rules. If it is not what you need,
you can just replace the rule with the same rule:
<pre>
# iptables -nvx -L FORWARD
Chain FORWARD (policy ACCEPT 101 packets, 10715 bytes)
 pkts bytes target     prot opt in     out     source               destination
   44  5151            all  --  *      *       192.168.0.117        0.0.0.0/0
   57  5564            all  --  *      *       0.0.0.0/0            192.168.0.117
# iptables -R FORWARD 1 -s 192.168.0.117
# iptables -nvx -L FORWARD
Chain FORWARD (policy ACCEPT 101 packets, 10715 bytes)
 pkts bytes target     prot opt in     out     source               destination
    0     0            all  --  *      *       192.168.0.117        0.0.0.0/0
   57  5564            all  --  *      *       0.0.0.0/0            192.168.0.117
</pre>

== More complicated cases ==
Well, now, when we know how to work in the easiest case, we'll try to understand what to do in
more complicated situations.

; More than one container on the node
: Just add the rules like above for each container IP.

; More than one IP per container.
: For each IP add the rules like above. When counting the complete traffic of a container you have to summarize over all IPs that this container owns.

; More interfaces on the HN.
: Nothing to do! :)

== Scripting ==
Here are some scripting ideas

=== Get CTIDs of all running containers ===
<pre>
host2:~/bin# cat vz-all-running
vzlist -H -oveid | sed 's/ //g;'
</pre>

=== Get all IPs of running containers ===
<pre>
host2:~/bin# cat vz-all-running-ip
vzlist -H -o ip
</pre>

=== Set up all needed iptables rules ===
<pre>
host2:~/bin# cat vz-iptables-create-rules
for i in `./vz-all-running-ip`;  do iptables -D FORWARD -s $i; iptables -D FORWARD -d $i; done >/dev/null 2>&1
for i in `./vz-all-running-ip`;  do iptables -A FORWARD -s $i; iptables -A FORWARD -d $i; done >/dev/null 2>&1
</pre>

=== Generate a traffic.log ===
Please use crontab to run this script once per hour or day to collect your traffic statistics.

(Warning, the counters can overflow if there is too much traffic within that period. Would recommend 15 minute intervals if you expect a lot of traffic)
<pre>
host2:~/bin# cat vz-generate-traffic-log
trafficlog="/var/log/vz-traffic.log"
for i in `./vz-all-running-ip` ;
 do
  echo -n `date "+%Y-%m-%d %H:%M:%S"` >> $trafficlog
  echo -n " $i " >> $trafficlog
  echo `iptables -nvx -L FORWARD | grep " $i " | tr -s [:blank:] |cut -d' ' -f3| awk '{sum+=$1} END {print sum;}'` >> $trafficlog
 done
 # reset the counter
 iptables -Z
 # update the iptables rules if there is a any change in containers
 ./vz-iptables-create-rules

 
 # copy the trafficlog file to a webserver where users can see their traffic

 # please mind to use
 # ssh-keygen -t rsa
 # to generate ssh keys
 # and append the new public key from your hardware node (~/.ssh/id_rsa.pub)
 # to ~/.ssh/authorized_keys2 on the HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS
 # in order for the below scp command to not ask for root password
 scp $trafficlog USER@HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS:/var/www/OPENVZ-CONTROL-WEB-SITE/tmp/$HOSTNAME-traffic
 
 # clear the copied trafficlog
 cp /dev/null $trafficlog
 # start a php script to store the traffic in a MySQL Database on the HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS
 # please mind to use .htaccess to secure this 
 wget -q http://HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS/login/traffic-read.php?HN=$HOSTNAME -O /dev/null
</pre>

=== Sample php script to store the trafficlog in a database ===

Below script will process traffic.log and store the data into a MySQL Database on the HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS 
<source lang="text">
HOST-TO-SHOW-THE-TRAFFIC-TO-THE-USERS:/var/www/OPENVZ-CONTROL-WEB-SITE# cat traffic-read.php
</source>
<source lang="php">
<?
 $MySQL_Host="INSERT-YOUR-MYSQL-HOST-HERE";
 $MySQL_User="INSERT-YOUR-MYSQL-USER-HERE";
 $MySQL_Passw="INSERT-YOUR-MYSQL-PASSWORD-HERE";
 
 mysql_connect($MySQL_Host,$MySQL_User,$MySQL_Passw);

 $HN=trim(addslashes($_GET["HN"])); // Hardware Node

 $handle = fopen ("tmp/{$HN}-traffic","r");
 while (!feof($handle)) {
   $line = fgets($handle, 4096);
   list($date,$time,$ip,$traffic)=explode(" ",$line);
   if($traffic>0) {mysql($db,"insert into Traffic (ip,measuringtime,bytes) values('{$ip}','{$date} {$time}','{$traffic}')");}
 } 
 fclose($handle);
?>
</source>

=== A SQL query to get the traffic for the last 30 days ===
<source lang="mysql">
SELECT sum(bytes)
FROM Traffic
WHERE ip = 'INSERT-YOUR-IP-HERE'
AND measuringtime > ( now() - INTERVAL 1 MONTH)
GROUP BY ip
</source>

=== Notes ===

As you see this way can be time-consuming in case of a big number of containers. 

So if anybody has scripts that automate all the process — you are welcome!

== See also ==
* [[Traffic accounting through proc]]

[[Category: Networking]]
[[Category: Monitoring]]