Network can fail in many ways, sometimes pretty subtle (e.g. high ratio packet loss). Disrupted TCP connections take a moderately long time (about 11 minutes with default configuration on Linux, for example) to be detected by the operating system. AMQP 0-9-1 offers a heartbeat feature to ensure that the application layer promptly finds out about disrupted connections (and also completely unresponsive peers). Heartbeats also defend against certain network equipment which may terminate "idle" TCP connections when there is no activity on them for a certain period of time.
TCP keepalives is a TCP stack feature that serves a similar purpose and can be very useful (possibly in combination with heartbeats) but requires kernel tuning in order to be practical with most operating systems and distributions.
heartbeat timeout value defines after what period of time the peer TCP connection should be considered unreachable (down) by RabbitMQ and client libraries. This value is negotiated between the client and RabbitMQ server at the time of connection. The client must be configured to request heartbeats.
The negotiation process works like this: the server will suggest its configurable value and the client will reconcile it with its configured value, and send the result value back. The value is in seconds, and default value suggested by RabbitMQ is
Java, .NET and Erlang clients maintained by the RabbitMQ core team use the following negotiation algorithm:
0 (see below), the greater value of the two is used
A zero value indicates that a peer suggests disabling heartbeats entirely. To disable heartbeats, both peers have to opt in and use the value of
0. This is highly recommended against unless the environment is known to use TCP keepalives on every host.
Very low values are also highly recommended against.
Heartbeat frames are sent about every
heartbeat timeout / 2 seconds. This value is sometimes referred to as the
heartbeat interval. After two missed heartbeats, the peer is considered to be unreachable. Different clients manifest this differently but the TCP connection will be closed. When a client detects that RabbitMQ node is unreachable due to a heartbeat, it needs to re-connect.
It is important to not confuse the timeout value with the interval one. RabbitMQ configuration exposes the timeout value, so do the officially supported client libraries. However some clients might expose the interval, potentially causing confusion.
Any traffic (e.g. protocol operations, published messages, acknowledgements) counts for a valid heartbeat. Clients may choose to send heartbeat frames regardless of whether there was any other traffic on the connection but some only do it when necessary.
Heartbeats can be deactivated by setting the timeout interval to
0 on the client side at connection time, providing the server heartbeat has also been set to zero.
Deactivating heartbeats is not recommended unless the environment is known to use TCP keepalives on every host (both RabbitMQ nodes and applications).
Alternatively a very high (say, 1800 seconds) value can be used on both ends to effectively deactivate heartbeats as frame delivery will be too infrequent to make a practical difference.
Unless TCP keepalives are used instead with an adequately low inactivity detection period, deactivating heartbeats is highly discouraged. If heartbeats are deactivated, it will make timely peer unavailability detection much less likely, which would pose a significant risk to data safety, in particular for publishers.
To configure the heartbeat timeout in the Java client, set it with
ConnectionFactory#setRequestedHeartbeat before creating a connection:
ConnectionFactory cf = new ConnectionFactory(); // set the heartbeat timeout to 60 seconds cf.setRequestedHeartbeat(60);
Note that in case RabbitMQ server has a non-zero heartbeat timeout configured (which is the default), the client can only lower the value but not increase it.
To configure the heartbeat timeout in the .NET client, set it with
ConnectionFactory.RequestedHeartbeat before creating a connection:
var cf = new ConnectionFactory(); // set the heartbeat timeout to 60 seconds cf.RequestedHeartbeat = TimeSpan.FromSeconds(60);
Setting heartbeat timeout value too low can lead to false positives (peer being considered unavailable while it is not really the case) due to transient network congestion, short-lived server flow control, and so on.
This should be taken into consideration when picking a timeout value.
Several years worth of feedback from the users and client library maintainers suggest that values lower than 5 seconds are fairly likely to cause false positives, and values of 1 second or lower are very likely to do so. Values within the 5 to 20 seconds range are optimal for most environments.
STOMP 1.2 includes heartbeats. In STOMP, heartbeat timeouts can be asymmetrical: that is to say, client and server can use different values. RabbitMQ STOMP plugin fully supports this feature.
Heartbeats in STOMP are opt-in. To enable them, use the
heart-beat header when connecting. See STOMP specification for an example.
MQTT 3.1.1 includes heartbeats under a different name ("keepalives"). RabbitMQ MQTT plugin fully supports this feature.
Keepalives in MQTT are opt-in. To enable them, set the
keepalive interval when connecting. Please consult your MQTT client's documentation for examples.
Please refer to the AMQP 0-9-1 URI query parameters reference for details.
TCP contains a mechanism similar in purpose to the heartbeat (a.k.a. keepalive) one in messaging protocols and net tick timeout covered above: TCP keepalives. Due to inadequate defaults, TCP keepalives cannot be assumed to be suitable for messaging protocols. However, with proper tuning they can be useful as an additional defense mechanism in environments where applications cannot be expected to enable heartbeats or use reasonable values.
In certain rare cases when heartbeats alone are not sufficient (e.g. when connections involved use a protocol that does not have a heartbeat mechanism of some kind), TCP keepalives must be configured to use a reasonably low timeout value.
TCP keepalives cover all TCP connections on a host, both inbound and outgoing. This makes them useful in scenarios with a high outgoing connection churn, for example, Shovel or Federation plugin links that are often deactivated and re-activated (re-enabled) or interrupted.
TCP keepalives can also be used instead of heartbeats by configuring them to lower system-specific values. In that case heartbeats can be deactivated. The main benefit of this approach is that all TCP connections on a machine will use identical values regardless of the protocol and client library used.
See the Networking guide for details.
Certain networking tools (HAproxy, AWS ELB) and equipment (hardware load balancers) may terminate "idle" TCP connections when there is no activity on them for a certain period of time. Most of the time it is not desirable.
When heartbeats are activated on a connection, it results in periodic light network traffic. Therefore heartbeats have a side effect of guarding client connections that can go idle for periods of time against premature closure by proxies and load balancers.
With a heartbeat timeout of 30 seconds the connection will produce periodic network traffic roughly every 15 seconds. Activity in the 5 to 15 second range is enough to satisfy the defaults of most popular proxies and load balancers. Also see the section on low timeouts and false positives above.
RabbitMQ nodes will log connections closed due to missed heartbeats. So will all officially supported client libraries. Inspecting server and client logs will provide valuable information and should be the first troubleshooting step.
It may be necessary to inspect the connections open to or from a node, their state, origin, username and effective heartbeat timeout value. Network Troubleshooting guide provides an overview of the tools available to help with that.