Skip to main content

Logging

Overview

Log files is a very important aspect of system observability, much like monitoring.

Developers and operators should inspect logs when troubleshooting an issue or assessing the state of the system.

RabbitMQ supports a number of features when it comes to logging.

This guide covers topics such as:

and more.

Log Outputs

RabbitMQ nodes can log to multiple outputs. Logging to a file is one of the most common options for RabbitMQ installations.

Logging to standard output and error streams is another popular option. Syslog is a yet another one supported out of the box.

Different outputs can have different log levels. For example, the console output can log all messages including debug information while the file output can only log error and higher severity messages.

Default Log Output and Behavior

Nodes log to a file by default, if no outputs are explicitly configured. If some are configured, they will be used.

RabbitMQ versions prior to 3.9.0 would always log to a file unless explicitly configured not to do so. With later versions, the same behavior can be achieved by explicitly listing a file output next to other desired log outputs, such as the standard stream one.

Log File Location

Please see the File and Directory Location guide to find default log file location for various platforms.

There are two ways to configure log file location. One is the configuration file. This option is recommended. The other is the RABBITMQ_LOGS environment variable. It can be useful in development environments.

Use RabbitMQ management UI or rabbitmq-diagnostics -q log_location to find when a node stores its log file(s).

The RABBITMQ_LOGS variable value can be either a file path or a hyphen (-). RABBITMQ_LOGS=- will result in all log messages being sent to standard output. See Logging to Console (Standard Output).

The environment variable takes precedence over the configuration file. When in doubt, consider overriding log file location via the config file. As a consequence of the environment variable precedence, if the environment variable is set, the configuration key log.file will not have any effect.

Configuration

RabbitMQ starts logging early on node start. See the Configuration guide for a general overview of how to configure RabbitMQ.

Logging to a File

Logging to a file is one of the most common options for RabbitMQ installations. In modern releases, RabbitMQ nodes only log to a file if explicitly configured to do so using the configuration keys listed below:

  • log.file: log file path or false to deactivate the file output. Default value is taken from the RABBITMQ_LOGS environment variable or configuration file
  • log.file.level: log level for the file output. Default level is info
  • log.file.formatter: controls log entry format, text lines or JSON
  • log.file.rotation.date, log.file.rotation.size, log.file.rotation.count for log file rotation settings
  • log.file.formatter.time_format: controls timestamp formatting

The following example overrides log file name:

log.file = rabbit.log

The following example overrides log file location:

log.file = /opt/custom/var/log/rabbit.log

The following example instructs RabbitMQ to log to a file at the debug level:

log.file.level = debug

For a list of supported log levels, see Log Levels.

Logging to a file can be deactivated with

log.file = false

Logging in JSON format to a file:

log.file.formatter = json

By default, RabbitMQ will use RFC 3339 timestamp format. It is possible to switch to a UNIX epoch-based format:

log.file = true
log.file.level = info

# use microseconds since UNIX epoch for timestamp format
log.file.formatter.time_format = epoch_usecs

The rest of this guide describes more options, including more advanced ones.

Log Rotation

RabbitMQ nodes always append to the log files, so a complete log history is preserved. Log file rotation is not performed by default. Debian and RPM packages will set up log rotation via logrotate after package installation.

log.file.rotation.date, log.file.rotation.size, log.file.rotation.count settings control log file rotation for the file output.

Built-in Periodic Rotation

Use log.file.rotation.date to set up minimalistic periodic rotation:

# rotate every night at midnight
log.file.rotation.date = $D0

# keep up to 5 archived log files in addition to the current one
log.file.rotation.count = 5
# rotate every day at 23:00 (11:00 p.m.)
log.file.rotation.date = $D23
# rotate every night at midnight
log.file.rotation.date = $D0

Built-in File Size-based Rotation

log.file.rotation.size controls rotation based on the current log file size:

# rotate when the file reaches 10 MiB
log.file.rotation.size = 10485760

# keep up to 5 archived log files in addition to the current one
log.file.rotation.count = 5

Rotation Using Logrotate

On Linux, BSD and other UNIX-like systems, logrotate is an alternative way of log file rotation and compression.

RabbitMQ Debian and RPM packages will set up logrotate to run weekly on files located in default /var/log/rabbitmq directory. Rotation configuration can be found in /etc/logrotate.d/rabbitmq-server.

Logging to Console (Standard Output)

Logging to standard streams (console) is another popular option for RabbitMQ installations, in particular when RabbitMQ nodes are deployed in containers. RabbitMQ nodes only log to standard streams if explicitly configured to do so.

Here are the main settings that control console (standard output) logging:

  • log.console (boolean): set to true to activate console output. Default is `false
  • log.console.level: log level for the console output. Default level is info
  • log.console.formatter: controls log entry format, text lines or JSON
  • log.console.formatter.time_format: controls timestamp formatting

To activate console logging, use the following config snippet:

log.console = true

The following example deactivates console logging

log.console = false

The following example instructs RabbitMQ to use the debug logging level when logging to console:

log.console.level = debug

For a list of supported log levels, see Log Levels.

Logging to console in JSON format:

log.console.formatter = json

When console output is activated, the file output will also be activated by default. To deactivate the file output, set log.file to false:

log.console = true
log.console.level = info

log.file = false

By default, RabbitMQ will use RFC 3339 timestamp format. It is possible to switch to a UNIX epoch-based format:

log.console = true
log.console.level = info

log.file = false

# use microseconds since UNIX epoch for timestamp format
log.console.formatter.time_format = epoch_usecs

Please note that RABBITMQ_LOGS=- will deactivate the file output even if log.file is configured.

Logging to Syslog

RabbitMQ logs can be forwarded to a Syslog server via TCP or UDP. UDP is used by default and requires Syslog service configuration. TLS is also supported.

Syslog output has to be explicitly configured:

log.syslog = true

Syslog Endpoint Configuration

By default the Syslog logger will send log messages to UDP port 514 using the RFC 3164 protocol. RFC 5424 protocol also can be used.

In order to use UDP the Syslog service must have UDP input configured.

UDP and TCP transports can be used with both RFC 3164 and RFC 5424 protocols. TLS support requires the RFC 5424 protocol.

The following example uses TCP and the RFC 5424 protocol:

log.syslog = true
log.syslog.transport = tcp
log.syslog.protocol = rfc5424

To use TLS, a standard set of TLS options must be provided:

log.syslog = true
log.syslog.transport = tls
log.syslog.protocol = rfc5424

log.syslog.ssl_options.cacertfile = /path/to/ca_certificate.pem
log.syslog.ssl_options.certfile = /path/to/client_certificate.pem
log.syslog.ssl_options.keyfile = /path/to/client_key.pem

Syslog service IP address and port can be customised:

log.syslog = true
log.syslog.ip = 10.10.10.10
log.syslog.port = 1514

If a hostname is to be used rather than an IP address:

log.syslog = true
log.syslog.host = my.syslog-server.local
log.syslog.port = 1514

Syslog metadata identity and facility values also can be configured. By default identity will be set to the name part of the node name (for example, rabbitmq in rabbitmq@hostname) and facility will be set to daemon.

To set identity and facility of log messages:

log.syslog = true
log.syslog.identity = my_rabbitmq
log.syslog.facility = user

Logging to Syslog in JSON format:

log.syslog = true

log.syslog.formatter = json

Less commonly used Syslog client options can be configured using the advanced config file.

JSON Logging

RabbitMQ nodes can format log messages as JSON, which can be convenient for parsing by other pieces of software.

Logging to a file in JSON format:

log.file.level = info
log.file.formatter = json

Logging to the console in JSON format:

log.console = true
log.console.level = info
log.console.formatter = json

log.file = false

Logging to Syslog in JSON format:

log.syslog = true

log.syslog.formatter = json

Log Message Categories

RabbitMQ has several categories of messages, which can be logged with different levels or to different files.

The categories replace the rabbit.log_levels configuration setting in versions earlier than 3.7.0.

The categories are:

  • connection: connection lifecycle events for AMQP 0-9-1, AMQP 1.0, MQTT and STOMP.
  • channel: channel logs. Mostly errors and warnings on AMQP 0-9-1 channels.
  • queue: queue logs. Mostly debug messages.
  • mirroring: queue mirroring logs. Queue mirrors status changes: starting/stopping/synchronizing.
  • federation: federation plugin logs.
  • upgrade: verbose upgrade logs. These can be excessive.
  • default: all other log entries. You cannot override file location for this category.

It is possible to configure a different log level or file location for each message category using log.<category>.level and log.<category>.file configuration variables.

By default each category will not filter by level. If an is output configured to log debug messages, the debug messages will be printed for all categories. Configure a log level for a category to override.

For example, given debug level in the file output, the following will deactivate debug logging for connection events:

log.file.level = debug
log.connection.level = info

To redirect all federation logs to the rabbit_federation.log file, use:

log.federation.file = rabbit_federation.log

To deactivate a log type, you can use the none log level. For example, to deactivate upgrade logs:

log.upgrade.level = none

Log Levels

Log levels is another way to filter and tune logging. Log levels have a strict ordering. Each log message has a severity from debug being the lowest severity to critical being the highest.

Logging verbosity can be controlled on multiple layers by setting log levels for categories and outputs. More verbose log levels will include more log messages from debug being the most verbose to none being the least.

The following log levels are used by RabbitMQ:

Log levelVerbositySeverity
debugmost verboselowest severity
info
warning
error
criticalhighest severity
noneleast verbosenot applicable

The default log level is info.

If a log message has lower severity than the category level, the message will be dropped and not sent to any output.

If a category level is not configured, its messages will always be sent to all outputs.

To make the default category log only errors or higher severity messages, use

log.default.level = error

The none level means no logging.

Each output can use its own log level. If a message has lower severity than the output level, the message will not be logged.

For example, if no outputs are configured to log debug messages, even if the category level is set to debug, the debug messages will not be logged.

On the other hand, if an output is configured to log debug messages, it will get them from all categories, unless a category is configured with a less verbose level.

Changing Log Level

There are two ways of changing effective log levels:

  • Via configuration file(s): this is more flexible but requires a node restart between changes
  • Using CLI tools, rabbitmqctl set_log_level <level>: the changes are transient (will not survive node restart) but can be used to activate and deactivate, for example, debug logging at runtime for a period of time.

To set log level to debug on a running node:

rabbitmqctl -n rabbit@target-host set_log_level debug

To set the level to info:

rabbitmqctl -n rabbit@target-host set_log_level info

Tailing Logs Using CLI Tools

Modern releases support tailing logs of a node using CLI tools. This is convenient when log file location is not known or is not easily accessible but CLI tool connectivity is allowed.

To tail three hundred last lines on a node rabbitmq@target-host, use rabbitmq-diagnostics log_tail:

# This is semantically equivalent to using `tail -n 300 /path/to/rabbit@hostname.log`.
# Use -n to specify target node, -N is to specify the number of lines.
rabbitmq-diagnostics -n rabbit@target-host log_tail -N 300

This will load and print last lines from the log file. If only console logging is activated, this command will fail with a "file not found" (enoent) error.

To continuously inspect as a stream of log messages as they are appended to a file, similarly to tail -f or console logging, use rabbitmq-diagnostics log_tail_stream:

# This is semantically equivalent to using `tail -f /path/to/rabbit@hostname.log`.
# Use Control-C to stop the stream.
rabbitmq-diagnostics -n rabbit@target-host log_tail_stream

This will continuously tail and stream lines added to the log file. If only console logging is activated, this command will fail with a "file not found" (enoent) error.

The rabbitmq-diagnostics log_tail_stream command can only be used against a running RabbitMQ node and will fail if the node is not running or the RabbitMQ application on it was stopped using rabbitmqctl stop_app.

Activating Debug Logging

When debug logging is enabled, the node will log a lot of information that can be useful for troubleshooting. This log severity is meant to be used when troubleshooting, say, the peer discovery activity.

For example to log debug messages to a file:

log.file.level = debug

To print log messages to standard I/O streams:

log.console = true
log.console.level = debug

To switch to debug logging at runtime:

rabbitmqctl -n rabbit@target-host set_log_level debug

To set the level back to info:

rabbitmqctl -n rabbit@target-host set_log_level info

It is possible to deactivate debug logging for some categories:

log.file.level = debug

log.connection.level = info
log.channel.level = info

Service Logs

On systemd-based Linux distributions, system service logs can be inspected using journalctl --system

journalctl --system

which requires superuser privileges. Its output can be filtered to narrow it down to RabbitMQ-specific entries:

sudo journalctl --system | grep rabbitmq

Service logs will include standard output and standard error streams of the node. The output of journalctl --system will look similar to this:

Dec 26 11:03:04 localhost rabbitmq-server[968]: ##  ##
Dec 26 11:03:04 localhost rabbitmq-server[968]: ## ## RabbitMQ 3.11.6. Copyright (c) 2005-2024 Broadcom. All Rights Reserved. The term "Broadcom" refers to Broadcom Inc. and/or its subsidiaries.
Dec 26 11:03:04 localhost rabbitmq-server[968]: ########## Licensed under the MPL. See https://www.rabbitmq.com/
Dec 26 11:03:04 localhost rabbitmq-server[968]: ###### ##
Dec 26 11:03:04 localhost rabbitmq-server[968]: ########## Logs: /var/log/rabbitmq/rabbit@localhost.log
Dec 26 11:03:04 localhost rabbitmq-server[968]: /var/log/rabbitmq/rabbit@localhost_upgrade.log
Dec 26 11:03:04 localhost rabbitmq-server[968]: Starting broker...
Dec 26 11:03:05 localhost rabbitmq-server[968]: systemd unit for activation check: "rabbitmq-server.service"
Dec 26 11:03:06 localhost rabbitmq-server[968]: completed with 6 plugins.

Logged Events

Connection Lifecycle Events

Successful TCP connections that send at least 1 byte of data will be logged. Connections that do not send any data, such as health checks of certain load balancer products, will not be logged.

Here's an example:

2018-11-22 10:44:33.654 [info] <0.620.0> accepting AMQP connection <0.620.0> (127.0.0.1:52771 -> 127.0.0.1:5672)

The entry includes client IP address and port (127.0.0.1:52771) as well as the target IP address and port of the server (127.0.0.1:5672). This information can be useful when troubleshooting client connections.

Once a connection successfully authenticates and is granted access to a virtual host, that is also logged:

2018-11-22 10:44:33.663 [info] <0.620.0> connection <0.620.0> (127.0.0.1:52771 -> 127.0.0.1:5672): user 'guest' authenticated and granted access to vhost '/'

The examples above include two values that can be used as connection identifiers in various scenarios: connection name (127.0.0.1:57919 -> 127.0.0.1:5672) and an Erlang process ID of the connection (<0.620.0>). The latter is used by rabbitmqctl and the former is used by the HTTP API.

A client connection can be closed cleanly or abnormally. In the former case the client closes AMQP 0-9-1 (or 1.0, or STOMP, or MQTT) connection gracefully using a dedicated library function (method). In the latter case the client closes TCP connection or TCP connection fails. RabbitMQ will log both cases.

Below is an example entry for a successfully closed connection:

2018-06-17 06:23:29.855 [info] <0.634.0> closing AMQP connection <0.634.0> (127.0.0.1:58588 -> 127.0.0.1:5672, vhost: '/', user: 'guest')

Abruptly closed connections will be logged as warnings:

2018-06-17 06:28:40.868 [warning] <0.646.0> closing AMQP connection <0.646.0> (127.0.0.1:58667 -> 127.0.0.1:5672, vhost: '/', user: 'guest'):
client unexpectedly closed TCP connection

Abruptly closed connections can be harmless. For example, a short lived program can naturally stop and don't have a chance to close its connection. They can also hint at a genuine issue such as a failed application process or a proxy that closes TCP connections it considers to be idle.

Watching Internal Events

RabbitMQ nodes have an internal mechanism. Some of its events can be of interest for monitoring, audit and troubleshooting purposes. They can be consumed as JSON objects using a rabbitmq-diagnostics command:

# will emit JSON objects
rabbitmq-diagnostics consume_event_stream

When used interactively, results can be piped to a command line JSON processor such as jq:

rabbitmq-diagnostics consume_event_stream | jq

The events can also be exposed to applications for consumption with a plugin.

Events are published as messages with blank bodies. All event metadata is stored in message metadata (properties, headers).

Below is a list of published events.

Core Broker

Queue, Exchange and Binding events:

  • queue.deleted
  • queue.created
  • exchange.created
  • exchange.deleted
  • binding.created
  • binding.deleted

Connection and Channel events:

  • connection.created
  • connection.closed
  • channel.created
  • channel.closed

Consumer events:

  • consumer.created
  • consumer.deleted

Policy and Parameter events:

  • policy.set
  • policy.cleared
  • parameter.set
  • parameter.cleared

Virtual host events:

  • vhost.created
  • vhost.deleted
  • vhost.limits.set
  • vhost.limits.cleared

User management events:

  • user.authentication.success
  • user.authentication.failure
  • user.created
  • user.deleted
  • user.password.changed
  • user.password.cleared
  • user.tags.set

Permission events:

  • permission.created
  • permission.deleted
  • topic.permission.created
  • topic.permission.deleted

Alarm events:

  • alarm.set
  • alarm.cleared

Shovel Plugin

Worker events:

  • shovel.worker.status
  • shovel.worker.removed

Federation Plugin

Link events:

  • federation.link.status
  • federation.link.removed

Consuming Log Entries Using a System Log Exchange

RabbitMQ can forward log entries to a system exchange, amq.rabbitmq.log, which will be declared in the default virtual host.

This feature is deactivated by default. To activate this logging, set the log.exchange configuration key to true:

# activate log forwarding to amq.rabbitmq.log, a topic exchange
log.exchange = true

log.exchange.level can be used to control the log level that will be used by this logging target:

log.exchange = true
log.exchange.level = warning

amq.rabbitmq.log is a regular topic exchange and can be used as such. Log entries are published as messages. Message body contains the logged message and routing key is set to the log level.

Application that would like to consume log entries need to declare a queue and bind it to the exchange, using a routing key to filter a specific log level, or # to consume all log entries allowed by the configured log level.