open-consul/website/source/docs/faq.html.markdown

5.2 KiB

layout page_title sidebar_current
docs Frequently Asked Questions docs-faq

Frequently Asked Questions

Q: What is Checkpoint? / Does Consul call home?

Consul makes use of a HashiCorp service called Checkpoint which is used to check for updates and critical security bulletins. Only anonymous information, which cannot be used to identify the user or host, is sent to Checkpoint . An anonymous ID is sent which helps de-duplicate warning messages. This anonymous ID can be disabled. In fact, using the Checkpoint service is optional and can be disabled.

See disable_anonymous_signature and disable_update_check.

Q: How does Atlas integration work?

Consul makes use of a HashiCorp service called SCADA (Supervisory Control And Data Acquisition). The SCADA system allows clients to maintain long-running connections to Atlas. Atlas can in turn provide auto-join facilities for Consul agents (supervisory control) and an integrated dashboard showing the health of all connected agents (data acquisition).

Standard ACLs can be applied to the SCADA connection, ensuring that Atlas is given only those privileges that make sense for your deployment.

Using the SCADA service is optional. SCADA is only enabled by opt-in.

See the Atlas integration guide for more details.

Q: Does Consul rely on UDP Broadcast or Multicast?

Consul uses the Serf gossip protocol which relies on TCP and UDP unicast. Broadcast and Multicast are rarely available in a multi-tenant or cloud network environment. For that reason, Consul and Serf were both designed to avoid any dependence on those capabilities.

Q: Is Consul eventually or strongly consistent?

Consul has two important subsystems, the service catalog and the gossip protocol. The service catalog stores all the nodes, service instances, health check data, ACLs, and KV information. It is strongly consistent, and replicated using the consensus protocol.

The gossip protocol is used to track which nodes are part of the cluster and to detect a node or agent failure. This information is eventually consistent by nature. When the servers detects a change in membership, or receive a health update, they update the service catalog appropriately.

Because of this split, the answer to the question is subtle. Almost all client APIs interact with the service catalog and are strongly consistent. Updates to the catalog may come via the gossip protocol which is eventually consistent, meaning the current state of the catalog can lag behind until the state is reconciled.

Q: Are failed or left nodes ever removed?

To prevent an accumulation of dead nodes (nodes in either failed or left states), Consul will automatically remove dead nodes out of the catalog. This process is called reaping. This is currently done on a configurable interval of 72 hours. Reaping is similar to leaving, causing all associated services to be deregistered. Changing the reap interval for aesthetic reasons to trim the number of failed or left nodes is not advised (nodes in the failed or left state do not cause any additional burden on Consul).

Q: Does Consul support delta updates for watchers or blocking queries?

Consul does not currently support sending a delta or a change only response to a watcher or a blocking query. The API simply allows for an edge-trigger return with the full result. A client should keep the results of their last read and compute the delta client side.

By design, Consul offloads this to clients instead of attempting to support the delta calculation. This avoids expensive state maintenance on the servers as well as race conditions between data updates and watch registrations.

Q: What network ports does Consul use?

The Ports Used section of the Configuration documentation lists all ports that Consul uses.

Q: Does Consul require certain user process resource limits?

There should be only a small number of open file descriptors required for a Consul client agent. The gossip layers perform transient connections with other nodes, each connection to the client agent (such as for a blocking query) will open a connection, and there will typically be connections to one of the Consul servers. A small number of file descriptors are also required for watch handlers, health checks, log files, and so on.

For a Consul server agent, you should plan on the above requirements and an additional incoming connection from each of the nodes in the cluster. This should not be the common case, but in the worst case if there is a problem with the other servers you would expect the other client agents to all connect to a single server and so preparation for this possibility is helpful.

The default ulimits are usually sufficient for Consul, but you should closely scrutinize your own environment's specific needs and identify the root cause of any excessive resource utilization before arbitrarily increasing the limits.