Graphite Format

Patterns and graphite.conf

The graphite.conf file in the /opt/atsd/atsd/conf/graphite.conf directory controls how graphite commands are mapped into ATSD commands.

graphite.conf contains patterns used to parse incoming metrics:

pattern = matches incoming metrics.

atsd-pattern = converts a graphite metric name into an ATSD metric name, entity, and tags.

If a metric name matches the regular expression pattern, the metric is parsed according to atsd-pattern.

Duplicate all \ characters in pattern.

graphite.conf is parsed from top to bottom, meaning that metric names are matched to patterns in the same order they are placed in the file. Matching stops as soon as a pattern in satisfied.

If a metric name has more tokens than atsd-pattern, additional tokens are ignored.

If a metric name has less tokens than atsd-pattern, but still satisfies pattern, then the metric is parsed. If there is no atsd-pattern for an incoming metric name, then everything before the first period is recorded as the entity and the rest is recorded as the metric. If there are no periods in the metric name, then the default entity is set to graphite, and the metric name is recorded as the metric.

metric: Metric token; multiple occurrences are combined.

entity: Entity token to replace the default entity (graphite); multiple occurrences are combined.

tag:tag_name: Token for the tag named tag_name.

metrics: Any number of metric tokens; can be used once per pattern.

String constants:

String constants in round brackets (tokens) are replaced.

String constants without round brackets (tokens) are added.

Empty tokens are omitted.
Empty atsd-pattern drops all incoming metrics that satisfy pattern.

Examples

INPUT: awgswgvml001.counters.axibase_com.wordpress.http.counts.api_bruteprotect_com.count
OUTPUT: e:awgswgvml001 m:wordpress.http.counts.count t:site=axibase_com t:url=api_bruteprotect_com
INPUT: servers.www-1.workers.busyWorkers
OUTPUT: e:www-1 m:workers.busyWorkers t:type=servers
INPUT: com.acmeCorp.instance01.jvm.memory.garbageCollections
OUTPUT: e:instance01 m:jvm.memory.garbageCollections t:type=com t:dep=acmeCorp
INPUT: collectd.nurdkr002.disk-sda1.disk_time.write
OUTPUT: e:nurdkr002 m:collectd.disk_time.write t:id=disk-sda1
INPUT: collectd.nurdkr002.interface-vethe538ad3.if_errors.tx
OUTPUT: e:nurdkr002 m:collectd.if_errors.tx t:id=interface-vethe538ad3
INPUT: collectd.nurdkr002.df-run-shm.percent_bytes-used
OUTPUT: e:nurdkr002 m:collectd.percent_bytes-used t:id=df-run-shm
INPUT: collectd.nurswgdkr002.df-run-shm.percent_bytes-used
OUTPUT: e:nurdkr002 m:collectd.percent_bytes-used t:id=run-shm

Note how the token containing df- is used to omit that precise part the of the input.

INPUT: collectd.nurswgdkr002.df-run-shm.percent_bytes-used
OUTPUT: e:nurdkr002 m:collectd.df.percent_bytes-used t:id=run-shm

Note how df is used to add it as a part of metric.

Collectl Example

Collectl is a universal system performance monitoring tool for Linux systems. Collectl can monitor a broad set of subsystems, which currently include buddyinfo, cpu, disk, inodes, infiniband, lustre, memory, network, nfs, processes, quadrics, slabs, sockets, and tcp.

You can instrument Collectl to send data to ATSD using the Graphite format.

Review the complete Collectl documentation here.

Review the complete list of collected metrics here.

Download Collectl:

http://sourceforge.net/projects/collectl/files/

Unpack Collectl:

tar xzf collectl-4.0.2.src.tar.gz

Install Collectl:

cd collectl-4.0.2
sudo ./INSTALL

Run Collectl:

collectl --export graphite,atsdserver:8081

Verify that Collectl is running correctly:

sudo collectl --export graphite,atsdserver:8081,d=1,s=dn --rawtoo -f /var/log/collectl

The output contains a log of data streamed to ATSD:

nurswgvml031.disktotals.reads 0 1434010410
nurswgvml031.disktotals.readkbs 0 1434010410
nurswgvml031.disktotals.writes 40 1434010410
nurswgvml031.disktotals.writekbs 228 1434010410
nurswgvml031.nettotals.kbin 0 1434010410
nurswgvml031.nettotals.pktin 8 1434010410
nurswgvml031.nettotals.kbout 1 1434010410
nurswgvml031.nettotals.pktout 9 1434010410
nurswgvml031.disktotals.reads 0 1434010411
nurswgvml031.disktotals.readkbs 0 1434010411
nurswgvml031.disktotals.writes 8 1434010411
nurswgvml031.disktotals.writekbs 36 1434010411

The entity and metrics collected by Collectl is visible under the Entity and Metrics tabs in ATSD.

Sensu Example

Sensu is a monitoring tool written in Ruby that uses RabbitMQ as a message broker and Redis for storing data. The tool is targeted at monitoring cloud environments.

You can instrument Sensu to send data to ATSD using the Graphite format.

To setup the server and client on separate servers, refer to the How to Configure Sensu guide.

Configure Sensu to send data to ATSD

To send data into ATSD, defined a TCP handler.

Official Sensu Handler guide.

{
  "handlers": {
    "tcp_socket": {
      "type": "tcp",
      "socket": {
        "host": "atsdserver",
        "port": 8081
      },
      "mutator": "only_check_output"
    }
  }
}

Including the "only_check_output" mutator is crucial. Without it the Sensu server is going to send the entire JSON output doc into ATSD, with the metadata and not just the necessary Graphite output.

You have to find and download a check plugin (or write one yourself). A large variety of Sensu community plugins, mainly written in Ruby, is available here: https://github.com/sensu-plugins

Next you have to create a check, for example:

{
  "checks": {
    "cpu_metrics": {
      "type": "metric",
      "handlers": [
        "debug",
        "tcp_socket"
      ],
      "command": "/etc/sensu/plugins/cpu-metrics.rb --scheme stats.:::name:::.cpu",
      "subscribers": ["webservers"],
      "interval": 10
    }
  }
}

The debug handler is included for logging purposes and can be omitted.

At least one element of subscribers has to match an element of the subscriptions in your client configuration file.

command is basically the address of the plugin you want to execute with the option --scheme enabled. It allows you to preface the metric name in the plugins output. Since ATSD accepts {prefix}.{hostname}.{check/task-name}.<...>, you can choose --scheme value accordingly. If you want your Sensu client name to be included in its place, you can also insert :::name::: into the prefix.