consul/website/source/docs/guides/performance.html.markdown

layout	page_title	sidebar_current	description
docs	Server Performance	docs-guides-performance	Consul requires different amounts of compute resources, depending on cluster size and expected workload. This guide provides guidance on choosing compute resources.

Server Performance

Since Consul servers run a consensus protocol to process all write operations and are contacted on nearly all read operations, server performance is critical for overall throughput and health of a Consul cluster. Servers are generally I/O bound for writes because the underlying Raft log store performs a sync to disk every time an entry is appended. Servers are generally CPU bound for reads since reads work from a fully in-memory data store that is optimized for concurrent access.

Minimum Server Requirements

In Consul 0.7, the default server performance parameters were tuned to allow Consul to run reliably (but relatively slowly) on a server cluster of three AWS t2.micro instances. These thresholds were determined empirically using a leader instance that was under sufficient read, write, and network load to cause it to permanently be at zero CPU credits, forcing it to the baseline performance mode for that instance type. Real-world workloads typically have more bursts of activity, so this is a conservative and pessimistic tuning strategy.

This default was chosen based on feedback from users, many of whom wanted a low cost way to run small production or development clusters with low cost compute resources, at the expense of some performance in leader failure detection and leader election times.

The default performance configuration is equivalent to this:

{
  "performance": {
    "raft_multiplier": 5
  }
}

Production Server Requirements

When running Consul 0.7 and later in production, it is recommended to configure the server performance parameters back to Consul's original high-performance settings. This will let Consul servers detect a failed leader and complete leader elections much more quickly than the default configuration which extends key Raft timeouts by a factor of 5, so it quite slow during these events.

The high performance configuration is simple and looks like this:

{
  "performance": {
    "raft_multiplier": 1
  }
}

It's best to benchmark with a realistic workload when choosing a production server for Consul. Here are some general recommendations:

• Consul will make use of multiple cores, and at least 2 cores are recommended.

• For write-heavy workloads, disk speed on the servers is key for performance. Use SSDs or another fast disk technology for the best write throughput.

• Spurious leader elections can be caused by networking issues between the servers or insufficient CPU resources. Users in cloud environments often bump their servers up to the next instance class with improved networking and CPU until leader elections stabilize, and in Consul 0.7 or later the performance parameters configuration now gives you tools to trade off performance instead of upsizing servers. You can use the consul.raft.leader.lastContact telemetry to observe how the Raft timing is performing and guide the decision to de-tune Raft performance or add more powerful servers.

• For DNS-heavy workloads, configuring all Consul agents in a cluster with the allow_stale configuration option will allow reads to scale across all Consul servers, not just the leader. See Stale Reads in the DNS Caching guide for more details. It's also good to set reasonable, non-zero DNS TTL values if your clients will respect them.

• In other applications that perform high volumes of reads against Consul, consider using the stale consistency mode available to allow reads to scale across all the servers and not just be forwarded to the leader.