ganglia

Another interesting example of Genshi templating is to automatically generate gmond/gmetad configuration files. The idea is that each cluster is headless: it communicates with the rest of the cluster members on an isolated multicast IP address and port. Any of the cluster members is therefore isolated on that particular ip/port pair. Additionally, each gmond instance also listens on UDP. This allows for any of the cluster members to be polled for information on the entire cluster!

The second part of the trick is in gmetad.conf. Here, we dynamically generate a list of clusters (based on profiles names) and a list of members to poll (based on the clients in said profiles). As the number of profiles and client grows, this list will grow automatically as well. When a new host is added, gmetad will receive an updated configuration and act accordingly.

There is one caveat though. The gmetad.conf parser is hard coded to read 16 arguments per data_source line. If you have more than 15 nodes in a cluster, you will see a warning in the logs. You can either ignore it, or truncate the list to the first 15 members.

In our environment, a profile is a one to one match with the role of that particular host. You can also do this based on groups, or any other client property.

Bundler/ganglia.xml

<Bundle name='ganglia'>
    <Package name='ganglia-gmond' />
    <Package name='ganglia-gmond-modules-python' />
    <Path name='/etc/ganglia/gmond.conf' />
    <Service name='gmond' />
    <Action name='gmond-reload' />

    <Group name='gmetad-server'>
      <Package name='ganglia-gmetad'/>
      <Package name='ganglia-web'/>
      <Package name='rrdtool'/>
      <Path name='/etc/ganglia/gmetad.conf' />
      <Service name='gmetad' />
    </Group>
</Bundle>

Rules/services-ganglia.xml

<Rules priority='10'>
    <Service name='gmond' type='chkconfig' status='on' />
    <Group name='gmetad-server'>
        <Service name='gmetad' type='chkconfig' status='on' />
    </Group>
</Rules>

Cfg/etc/ganglia/gmetad.conf/gmetad.conf.genshi

{% python
   client_metadata =  metadata.query.all()
   profile_array = {}
   seen = []
   for item in client_metadata:
       if item.profile not in seen:
           seen.append(item.profile)
           profile_array[item.profile]=[]
       profile_array[item.profile].append(item.hostname)
   seen.sort()
%}\

gridname "Our Grid"

{% for profile in seen %}
data_source "${profile}" \
{% for host in profile_array[profile] %}\
${host} \
{% end %}\
{% end %}

rrd_rootdir "/var/lib/ganglia/rrds"

Cfg/etc/ganglia/gmond.conf/gmod.conf.genshi

{% python
import random
random.seed(metadata.profile)
last_octet=random.randint(2,254)
%}\
/*
 $$Id$$
 $$HeadURL$$
*/

/* This configuration is as close to 2.5.x default behavior as possible
   The values closely match ./gmond/metric.h definitions in 2.5.x */
globals {
  daemonize = yes
  setuid = yes
  user = nobody
  debug_level = 0
  max_udp_msg_len = 1472
  mute = no
  deaf = no
  host_dmax = 1800 /* 30 minutes */
  cleanup_threshold = 604800 /*secs=1 week */
  gexec = no
  send_metadata_interval = 0
}

/* If a cluster attribute is specified, then all gmond hosts are wrapped inside
 * of a <CLUSTER> tag.  If you do not specify a cluster tag, then all <HOSTS> will
 * NOT be wrapped inside of a <CLUSTER> tag. */
cluster {
  name = "${metadata.profile}"
  owner = "user@company.net"
  latlong = "unspecified"
  url = "unspecified"
}

/* The host section describes attributes of the host, like the location */
host {
  location = "unspecified"
}

/* Feel free to specify as many udp_send_channels as you like.  Gmond
   used to only support having a single channel */
udp_send_channel {
  host = ${metadata.hostname}
  port = 8649
}
udp_send_channel {
  mcast_join = 239.2.11.${last_octet}
  port = 8649
  ttl = 1
}

/* You can specify as many udp_recv_channels as you like as well. */
udp_recv_channel {
  port = 8649
  bind = ${metadata.hostname}
}
udp_recv_channel {
  mcast_join = 239.2.11.${last_octet}
  bind       = 239.2.11.${last_octet}
  port = 8649
}

/* You can specify as many tcp_accept_channels as you like to share
   an xml description of the state of the cluster */
tcp_accept_channel {
  port = 8649
}

/* Each metrics module that is referenced by gmond must be specified and
   loaded. If the module has been statically linked with gmond, it does not
   require a load path. However all dynamically loadable modules must include
   a load path. */
modules {
/* [snip] */