RabbitMQ Management HTTP Stats

Introduction

Most of the GET requests you can issue to the HTTP API return JSON objects with a large number of keys. While a few of these keys represent things you set yourself in a PUT request or AMQP command (e.g. queue durability or arguments), most of them represent statistics to do with the object in question. This page attempts to document them.

It should be read in conjunction with the manual page for rabbitmqctl (see your installation if on Unix / Linux, or the RabbitMQ website for the latest version). Any field which can be returned by a command of the form rabbitmqctl list_something will also be returned in the equivalent part of the HTTP API, so all those keys are not documented here. However, the HTTP API also adds a lot of extra fields which are not available in rabbitmqctl.

_details objects

Many fields represent a count of some kind: queue length, messages acknowledged, bytes received and so on. Such absolute counts returned by the HTTP API will often have a corresponding _details object which offers information on how this count has changed. So for example, from a queue:

    "messages": 123619,
    "messages_details": {
      "avg": 41206.333333333336,
      "avg_rate": 1030.1583333333333,
      "rate": 24723.8,
      "samples": [
        {
          "sample": 123619,
          "timestamp": 1400680560000
        },
        {
          "sample": 0,
          "timestamp": 1400680500000
        },
        {
          "sample": 0,
          "timestamp": 1400680440000
        }
      ]
    }

Here we have a messages count (the total messages in the queue), with some additional data:

avg The average value for the requested time period (see below).
avg_rate The average rate for the requested time period.
rate How much the count has changed per second in the most recent sampling interval.
samples Snapshots showing how the value has changed over the requested time period.

avg, avg_rate and samples will only appear if you request a specific time period by appending query parameters to the URL. To do this you need to set an age and an increment for the samples you want. The end of the range returned will always correspond to the present.

Different types of data take different query parameters to return samples, as in the following table. You can specify more than one set of parameters if the resource you are requesting can generate more than one type of sample (for example, queues can return message rates and queue lengths).

Messages sent and received msg_rates_age / msg_rates_incr
Bytes sent and received data_rates_age / data_rates_incr
Queue lengths lengths_age / lengths_incr
Node statistics (e.g. file descriptors, disk space free) node_stats_age / node_stats_incr

For example, appending ?lengths_age=3600&lengths_incr=60 will return the last hour's data on queue lengths, with a sample for every minute.

message_stats objects

Many objects (including queues, exchanges and channels) will return counts of messages passing through them. These are included in a message_stats object (which in turn will contain _details objects for each count, as described above).

These can contain:

publish Count of messages published.
publish_in Count of messages published "in" to an exchange, i.e. not taking account of routing.
publish_out Count of messages published "out" of an exchange, i.e. taking account of routing.
confirm Count of messages confirmed.
deliver Count of messages delivered in acknowledgement mode to consumers.
deliver_noack Count of messages delivered in no-acknowledgement mode to consumers.
get Count of messages delivered in acknowledgement mode in response to basic.get.
get_noack Count of messages delivered in no-acknowledgement mode in response to basic.get.
deliver_get Sum of all four of the above.
redeliver Count of subset of messages in deliver_get which had the redelivered flag set.
return Count of messages returned to publisher as unroutable.

Only fields for which some activity has taken place will appear.

Detailed message stats objects

In addition, queues, exchanges and channels can return a breakdown of message stats for each of their neighbours (i.e. adjacent objects in the chain: channel -> exchange -> queue -> channel). This will only happen if the rates_mode configuration item has been switched to detailed from its default of basic.

As this possibly constitutes a large quantity of data, it is also only returned when querying a single channel, queue or exchange rather than a list. Note also that the default sample retention policy means that these detailed message stats do not retain historical data for more than a few seconds.

The detailed message stats objects have different names depending on where they are (documented below). Each set of detailed stats consists of a list of objects with two fields, one identifying the partner object and one stats which is a message_stats object as described above.

For example, from a queue:

  "incoming": [
    {
      "stats": {
        "publish": 352593,
        "publish_details": {
          "rate": 100.2
        }
      },
      "exchange": {
        "name": "my-exchange",
        "vhost": "/"
      }
    }
    {
      "stats": {
        "publish": 543784,
        "publish_details": {
          "rate": 54.6
        }
      },
      "exchange": {
        "name": "amq.topic",
        "vhost": "/"
      }
    }
  ],

This queue is currently receiving messages from two exchanges: 100.2 msg/s from "my-exchange" and 54.6 msg/s from "amq.topic".

/api/overview

This has the following fields:

cluster_name The name of the entire cluster, as set with rabbitmqctl set_cluster_name.
contexts A list of web application contexts in the cluster.
erlang_full_version A string with extended detail about the Erlang VM and how it was compiled, for the node connected to.
erlang_version A string with the Erlang version of the node connected to. As clusters should all run the same version this can be taken as representing the cluster.
exchange_types A list of all exchange types available.
listeners All (non-HTTP) network listeners for all nodes in the cluster. (See contexts in /api/nodes for HTTP).
management_version Version of the management plugin in use.
message_stats A message_stats object for everything the user can see - for all vhosts regardless of permissions in the case of monitoring and administrator users, and for all vhosts the user has access to for other users.
node The name of the cluster node this management plugin instance is running on.
object_totals An object containing global counts of all connections, channels, exchanges, queues and consumers, subject to the same visibility rules as for message_stats.
queue_totals An object containing sums of the messages, messages_ready and messages_unacknowledged fields for all queues, again subject to the same visibility rules as for message_stats.
rabbitmq_version Version of RabbitMQ on the node which processed this request.
rates_mode 'none', 'basic' or 'detailed'.
statistics_db_event_queue Number of outstanding statistics events yet to be processed by the database.
statistics_db_node Name of the cluster node hosting the management statistics database.

/api/nodes

This has the following fields:

applications List of all Erlang applications running on the node.
auth_mechanisms List of all SASL authentication mechanisms installed on the node.
cluster_links A list of the other nodes in the cluster. For each node, there are details of the TCP connection used to connect to it and statistics on data that has been transferred.
config_files List of config files read by the node.
contexts List of all HTTP listeners on the node.
db_dir Location of the persistent storage used by the node.
disk_free Disk free space in bytes.
disk_free_alarm Whether the disk alarm has gone off.
disk_free_limit Point at which the disk alarm will go off.
enabled_plugins List of plugins which are both explicitly enabled and running.
exchange_types Exchange types available on the node.
fd_total File descriptors available.
fd_used Used file descriptors.
io_read_avg_time Average wall time (milliseconds) for each disk read operation in the last statistics interval.
io_read_bytes Total number of bytes read from disk by the persister.
io_read_count Total number of read operations by the persister.
io_reopen_count Total number of times the persister has needed to recycle file handles between queues. In an ideal world this number will be zero; if the number is large, performance might be improved by increasing the number of file handles available to RabbitMQ.
io_seek_avg_time Average wall time (milliseconds) for each seek operation in the last statistics interval.
io_seek_count Total number of seek operations by the persister.
io_sync_avg_time Average wall time (milliseconds) for each fsync() operation in the last statistics interval.
io_sync_count Total number of fsync() operations by the persister.
io_write_avg_time Average wall time (milliseconds) for each disk write operation in the last statistics interval.
io_write_bytes Total number of bytes written to disk by the persister.
io_write_count Total number of write operations by the persister.
log_file Location of main log file.
mem_used Memory used in bytes.
mem_alarm Whether the memory alarm has gone off.
mem_limit Point at which the memory alarm will go off.
mnesia_disk_tx_count Number of Mnesia transactions which have been performed that required writes to disk. (e.g. creating a durable queue). Only transactions which originated on this node are included.
mnesia_ram_tx_count Number of Mnesia transactions which have been performed that did not require writes to disk. (e.g. creating a transient queue). Only transactions which originated on this node are included.
msg_store_read_count Number of messages which have been read from the message store.
msg_store_write_count Number of messages which have been written to the message store.
name Node name.
net_ticktime Current kernel net_ticktime setting for the node.
os_pid Process identifier for the Operating System under which this node is running.
partitions List of network partitions this node is seeing.
proc_total Maximum number of Erlang processes.
proc_used Number of Erlang processes in use.
processors Number of cores detected and usable by Erlang.
queue_index_journal_write_count Number of records written to the queue index journal. Each record represents a message being published to a queue, being delivered from a queue, and being acknowledged in a queue.
queue_index_read_count Number of records read from the queue index.
queue_index_write_count Number of records written to the queue index.
rates_mode 'none', 'basic' or 'detailed'.
run_queue Average number of Erlang processes waiting to run.
running Boolean for whether this node is up. Obviously if this is false, most other stats will be missing.
sasl_log_file Location of sasl log file.
sockets_total File descriptors available for use as sockets.
sockets_used File descriptors used as sockets.
type 'disc' or 'ram'.
uptime Time since the Erlang VM started, in milliseconds.

/api/nodes/(name)

All of the above, plus:

memory Detailed memory use statistics. Only appears if ?memory=true is appended to the URL.
binary Detailed breakdown of the owners of binary memory. Only appears if ?binary=true is appended to the URL. Note that this can be an expensive query if there are many small binaries in the system.

/api/connections

/api/connections/(name)

See documentation for rabbitmqctl list_connections. No additional fields, although pid is replaced by node.

Note also that while non-AMQP connections will appear in this list (unlike rabbitmqctl list_connections), they will omit many of the connection-level statistics.

/api/connections/(name)/channels

/api/channels

See documentation for rabbitmqctl list_channels, with pid replaced by node, plus:

connection_details Some basic details about the owning connection.
message_stats See the section on message_stats above.

/api/channels/(name)

All the above, plus

publishes Detailed message stats (see section above) for publishes to exchanges.
deliveries Detailed message stats for deliveries from queues.
consumer_details List of consumers on this channel, with some details on each.

/api/exchanges

/api/exchanges/(vhost)

See documentation for rabbitmqctl list_exchanges, plus:

message_stats See the section on message_stats above.

/api/exchanges/(vhost)/(name)

All the above, plus:

incoming Detailed message stats (see section above) for publishes from channels into this exchange.
outgoing Detailed message stats for publishes from this exchange into queues.

/api/queues

/api/queues/(vhost)

See documentation for rabbitmqctl list_queues, with all references to pids replaced by nodes plus:

message_stats See the section on message_stats above.

/api/queues/(vhost)/(name)

All the above, plus:

incoming Detailed message stats (see section above) for publishes from exchanges into this queue.
deliveries Detailed message stats for deliveries from this queue into channels.
consumer_details List of consumers on this channel, with some details on each.

/api/vhosts/

/api/vhosts/(name)

All the fields from rabbitmqctl list_vhosts (i.e. name and tracing) plus:

message_stats Global message_stats for this vhost. Note that activity for other users in this vhost is shown, even for users without the monitoring tag.
messages messages_ready messages_acknowledged Sum of these fields for all queues in the vhost.
recv_oct send_oct Sum of these fields for all connections to the vhost.