News HashiCorp to join IBM to accelerate multi-cloud automation. Read the blog →
Consul
Consul Home

DNS caching

This page describes the process to return cached results in response to DNS lookups. Consul agents can use DNS caching to reduce response time, but might provide stale information in the process.

Introduction

By default, Consul serves all DNS results with a 0 TTL value, which prevents any caching. This configuration returns the most recent information because each DNS lookup runs every time. However, this configuration adds latency to each lookup and can potentially exhaust the query throughput of a datacenter.

There are several ways you can modify to fine-tune Consul DNS lookup behavior to best suit your network's requirements.

TTL values

You can configure TTL values in the agent configuration file to allow DNS results to be cached downstream of Consul. Higher TTL values reduce the number of lookups on the Consul servers and speed lookups for clients, at the cost of increasingly stale results. By default, all TTLs are zero, preventing any caching.

dns_config {
  service_ttl {
    "*" = "0s"
  }
  node_ttl = "0s"
}

{
  "dns_config": {
    "service_ttl": {
      "*": "0s"
    },
    "node_ttl": "0s"
  }
}

Enable caching

To enable caching of node lookups, set the dns_config.node_ttl value. This can be set to 10s for example, and all node lookups will serve results with a 10 second TTL.

Service TTLs can be specified in a more granular fashion. You can set TTLs per-service, with a wildcard TTL as the default. This is specified using the dns_config.service_ttl map. The * is supported at the end of any prefix and has a lower precedence than strict match, so my-service-x has precedence over my-service-*. When performing wildcard match, the longest path is taken into account, thus my-service-* TTL will be used instead of my-* or *. With the same rule, * is the default value when nothing else matches. If no match is found the TTL defaults to 0.

For example, a dns_config that provides a wildcard TTL and a specific TTL for a service might look like this:

dns_config {
  service_ttl {
    "*" = "5s"
    "web" = "30s"
    "db*" = "10s"
    "db-master" = "3s"
  }
}

{
  "dns_config": {
    "service_ttl": {
      "*": "5s",
      "web": "30s",
      "db*": "10s",
      "db-master": "3s"
    }
  }
}

This sets all lookups to "web.service.consul" to use a 30 second TTL while lookups to "api.service.consul" will use the 5 second TTL from the wildcard. All lookups matching "db*" would get a 10 seconds TTL except "db-master" that would have a 3 seconds TTL.

Prepared queries

Prepared Queries provide an additional level of control over TTL. They allow for the TTL to be defined along with the query, and they can be changed on the fly by updating the query definition. If a TTL is not configured for a prepared query, then it will fall back to the service-specific configuration defined in the Consul agent as described above, and ultimately to 0 if no TTL is configured for the service in the Consul agent.

Stale reads

Stale reads can be used to reduce latency and increase the throughput of DNS queries. The settings used to control stale reads of DNS queries are:

With these two settings you can allow or prevent stale reads. Below we will discuss the advantages and disadvantages of both.

Allow stale reads

Since Consul 0.7.1, allow_stale is enabled by default and uses a max_stale value that defaults to a near-indefinite threshold (10 years). This allows DNS queries to continue to be served in the event of a long outage with no leader. A new telemetry counter has also been added at consul.dns.stale_queries to track when agents serve DNS queries that are stale by more than 5 seconds.

dns_config {
  allow_stale = true
  max_stale = "87600h"
}

{
  "dns_config": {
    "allow_stale": true,
    "max_stale": "87600h"
  }
}

Note

The above example is the default setting. You do not need to set it explicitly.

Doing a stale read allows any Consul server to service a query, but non-leader nodes may return data that is out-of-date. By allowing data to be slightly stale, you get horizontal read scalability. Now any Consul server can service the request, so you increase throughput by the number of servers in a datacenter.

Prevent stale reads

If you want to prevent stale reads or limit how stale they can be, you can set allow_stale to false or use a lower value for max_stale. Doing the first will ensure that all reads are serviced by a single leader node. The reads will then be strongly consistent but will be limited by the throughput of a single node.

dns_config {
  allow_stale = false
}

{
  "dns_config": {
    "allow_stale": false
  }
}

Negative response caching

Although DNS clients cache negative responses, Consul returns a "not found" style response when a service exists but there are no healthy endpoints. When using DNS for service discovery, cached negative responses may cause a service to appear down for longer than it is actually unavailable.

Configure SOA

In Consul v1.3.0 and newer, it is now possible to tune SOA responses and modify the negative TTL cache for some resolvers. It can be achieved using the soa.min_ttl configuration within the soa configuration.

dns_config {
  soa {
    min_ttl = 60
  }
}

{
  "dns_config": {
    "soa": {
      "min_ttl": 60
    }
  }
}

One common example is that Windows will default to caching negative responses for 15 minutes. DNS forwarders may also cache negative responses, with the same effect. To avoid this problem, check the negative response cache defaults for your client operating system and any DNS forwarder on the path between the client and Consul and set the cache values appropriately. In many cases "appropriately" means turning negative response caching off to get the best recovery time when a service becomes available again.

Edit this page on GitHub