» Managed Proxy Deprecation
Consul Connect was first released as a beta feature in Consul 1.2.0. The initial release included a feature called "Managed Proxies". Managed proxies were Connect proxies where the proxy process was started, configured, and stopped by Consul. They were enabled via basic configurations within the service definition.
Consul 1.3.0 deprecates Managed Proxies completely. It's strongly recommended you do not build anything using Managed proxies and consider using sidecar service registrations instead.
Even though this was a beta feature, managed proxies will continue to work at least until Consul 1.5 to prevent disruption to demonstration and proof-of-concept deployments of Consul Connect. Anyone using managed proxies though should aim to change their workflow as soon as possible to avoid issues with a later upgrade.
While the current functionality will remain present for a few major releases, new and known issues will not be fixed.
» Deprecation Rationale
Originally managed proxies traded higher implementation complexity for an easier "getting started" user experience. After seeing how Connect was investigated and adopted during beta it because obvious that they were not the best trade off.
Managed proxies only really helped in local testing or VM-per-service based models whereas a lot of users jumped straight to containers where they are not helpful. They also add only targeted fairly basic supervisor features which meant most people would want to use something else in production for consistency with other workloads. So the high implementation cost of building robust process supervision didn't actually benefit most real use-cases.
Instead of this Connect 1.3.0 introduces the concept of sidecar service registrations which have almost all of the benefits of simpler configuration but without any of the additional process management complexity. As a result they can be used to simplify configuration in both container-based and realistic production supervisor settings.
» Managed Proxy Documentation
As the managed proxy features continue to be supported for now, the rest of this page will document how they work in the interim.
Deprecation Note: It's strongly recommended you do not build anything using Managed proxies and consider using sidecar service registrations instead.
Managed proxies are given a unique proxy-specific ACL token that allows read-only access to Connect information for the specific service the proxy is representing. This ACL token is more restrictive than can be currently expressed manually in an ACL policy.
The default managed proxy is a basic proxy built-in to Consul and written in Go. Having a basic built-in proxy allows Consul to have a sane default with performance that is good enough for most workloads. In some basic benchmarks, the service-to-service communication over the built-in proxy could sustain 5 Gbps with sub-millisecond latency. Therefore, the performance impact of even the basic built-in proxy is minimal.
Consul will be integrating with advanced proxies in the near future to support more complex configurations and higher performance. The configuration below is all for the built-in proxy.
Security Note: 1.) Managed proxies can only be configured via agent configuration files. They cannot be registered via the HTTP API. And 2.) Managed proxies are not started at all if Consul is running as root. Both of these default configurations help prevent arbitrary process execution or privilege escalation. This behavior can be configured per-agent.
» Lifecycle
The Consul agent starts managed proxies on demand and supervises them, restarting them if they crash. The lifecycle of the proxy process is decoupled from the agent so if the agent crashes or is restarted for an upgrade, the managed proxy instances will not be stopped.
Note that this behaviour while desirable in production might leave proxy processes running indefinitely if you manually stop the agent and clear it's data dir during testing.
To terminate a managed proxy cleanly you need to deregister the service that requested it. If the agent is already stopped and will not be restarted again, you may choose to locate the proxy processes and kill them manually.
While in -dev
mode, unless a -data-dir
is explicitly set, managed proxies
switch to being killed when the agent exits since it can't store state in order
to re-adopt them on restart.
» Minimal Configuration
Managed proxies are configured within a service definition. The simplest possible managed proxy configuration is an empty configuration. This enables the default managed proxy and starts a listener for that service:
{
"service": {
"name": "redis",
"port": 6379,
"connect": { "proxy": {} }
}
}
The listener is started on random port within the configured Connect port range. It can be discovered using the DNS interface or Catalog API. In most cases, service-to-service communication is established by a proxy configured with upstreams (described below), which handle the discovery transparently.
» Upstream Configuration
To transparently discover and establish Connect-based connections to dependencies, they must be configured with a static port on the managed proxy configuration:
{
"service": {
"name": "web",
"port": 8080,
"connect": {
"proxy": {
"upstreams": [{
"destination_name": "redis",
"local_bind_port": 1234
}]
}
}
}
}
In the example above, "redis" is configured as an upstream with static port 1234 for service "web". When a TCP connection is established on port 1234, the proxy will find Connect-compatible "redis" services via Consul service discovery and establish a TLS connection identifying as "web".
Security: Any application that can communicate to the configured static port will be able to masquerade as the source service ("web" in the example above). You must either trust any loopback access on that port or use namespacing techniques provided by your operating system.
Deprecation Note: versions 1.2.0 to 1.3.0 required specifying upstreams
as part of the opaque config
that is passed to the proxy. However, since
1.3.0, the upstreams
configuration is now specified directily under the
proxy
key. Old service definitions using the nested config will continue to
work and have the values copied into the new location. This allows the upstreams
to be registered centrally rather than being part of the local-only config
passed to the proxy instance.
For full details of the additional configurable options available when using the built-in proxy see the built-in proxy configuration reference.
» Prepared Query Upstreams
The upstream destination may also be a prepared query. This allows complex service discovery behavior such as connecting to the nearest neighbor or filtering by tags.
For example, given a prepared query named "nearest-redis" that is configured to route to the nearest Redis instance, an upstream can be configured to route to this query. In the example below, any TCP connection to port 1234 will attempt a Connect-based connection to the nearest Redis service.
{
"service": {
"name": "web",
"port": 8080,
"connect": {
"proxy": {
"upstreams": [{
"destination_name": "redis",
"destination_type": "prepared_query",
"local_bind_port": 1234
}]
}
}
}
}
Note: Connect does not currently support cross-datacenter service communication. Therefore, prepared queries with Connect should only be used to discover services within a single datacenter. See Multi-Datacenter Connect for more information.
For full details of the additional configurable options available when using the built-in proxy see the built-in proxy configuration reference.
» Custom Managed Proxy
Custom proxies can also be configured to run as a managed proxy. To configure custom proxies, specify an alternate command to execute for the proxy:
{
"service": {
"name": "web",
"port": 8080,
"connect": {
"proxy": {
"exec_mode": "daemon",
"command": ["/usr/bin/my-proxy", "-flag-example"],
"config": {
"foo": "bar"
}
}
}
}
}
The exec_mode
value specifies how the proxy is executed. The only
supported value at this time is "daemon". The command is the binary and
any arguments to execute.
The "daemon" mode expects a proxy to run as a long-running, blocking
process. It should not double-fork into the background. The custom
proxy should retrieve its configuration (such as the port to run on)
via the custom proxy integration APIs.
The default proxy command can be changed at an agent-global level in the agent configuration. An example in HCL format is shown below.
connect {
proxy_defaults {
command = ["/usr/bin/my-proxy"]
}
}
With this configuration, all services registered without an explicit
proxy command will use my-proxy
instead of the default built-in proxy.
The config
key is an optional opaque JSON object which will be passed through
to the proxy via the proxy configuration endpoint to allow any configuration
options the proxy needs to be specified. See the built-in proxy
configuration reference
for details of config options that can be passed when using the built-in proxy.
» Managed Proxy Logs
Managed proxies have both stdout and stderr captured in log files in the agent's
data_dir
. They can be found in
<data_dir>/proxy/logs/<proxy_service_id>-std{err,out}.log
.
The built-in proxy will inherit it's log level from the agent so if the agent is
configured with log_level = DEBUG
, a proxy it starts will also output DEBUG
level logs showing service discovery, certificate and authorization information.