Merge pull request #14259 from hashicorp/docs/1-13-upgrade-considerations

docs: add 1.13 upgrade considerations
This commit is contained in:
Jared Kirschner 2022-08-18 18:18:33 -04:00 committed by GitHub
commit 92b718ad96
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 122 additions and 12 deletions

View file

@ -2084,7 +2084,7 @@ specially crafted certificate signed by the CA can be used to gain full access t
### Deprecated Options ((#tls_deprecated_options))
The following options were deprecated in Consul 1.12, please use the
[`tls`](#tls) stanza instead.
[`tls`](#tls-1) stanza instead.
- `ca_file` See: [`tls.defaults.ca_file`](#tls_defaults_ca_file).

View file

@ -31,6 +31,9 @@ we recommend reviewing the changelog for versions between the one you are on and
one you are upgrading to at each step to familiarize yourself with changes.
Select your _currently installed_ release series:
- 1.12.x: work upwards from [1.13 upgrade notes](/docs/upgrading/upgrade-specific#consul-1-13-x)
- 1.11.x: work upwards from [1.12 upgrade notes](/docs/upgrading/upgrade-specific#consul-1-12-0)
- 1.10.x: work upwards from [1.11 upgrade notes](/docs/upgrading/upgrade-specific#consul-1-11-0)
- [1.9.x](/docs/upgrading/instructions/upgrade-to-1-10-x)
- [1.8.x](/docs/upgrading/instructions/upgrade-to-1-10-x)
- [1.7.x](/docs/upgrading/instructions/upgrade-to-1-8-x)

View file

@ -18,8 +18,15 @@ upgrade flow.
### Service Mesh Compatibility
Existing Consul deployments using service mesh (i.e., containing any registered
Connect proxies) should upgrade to **at least Consul 1.13.1**.
Before upgrading existing Consul deployments using service mesh to Consul 1.13.x,
review the following guidances relevant to your deployment:
- [All service mesh deployments](#all-service-mesh-deployments)
- [Service mesh deployments using auto-encrypt or auto-config](#service-mesh-deployments-using-auto-encrypt-or-auto-config)
- [Service mesh deployments without the HTTPS port enabled on Consul agents](#service-mesh-deployments-without-the-https-port-enabled-on-consul-agents)
#### All service mesh deployments
Upgrade to **Consul version 1.13.1 or later**.
Consul 1.13.0 contains a bug that prevents Consul server agents from restoring
saved state on startup if the state
@ -29,17 +36,117 @@ saved state on startup if the state
This bug is fixed in Consul versions 1.13.1 and newer.
### gRPC TLS
#### Service mesh deployments using auto-encrypt or auto-config
In prior Consul versions if HTTPS was enabled for the client API and exposed
via `ports { https = NUMBER }` then the same TLS material was used to encrypt
the gRPC port used for xDS. Now this is decoupled and activating TLS on the
gRPC endpoint is controlled solely with the gRPC section of the new
[`tls` stanza](/docs/agent/config/config-files#tls-configuration-reference).
**Do not upgrade to Consul 1.13 yet** if using
[auto-encrypt](/docs/agent/config/config-files#auto_encrypt) or
[auto-config](/docs/agent/config/config-files#auto_config).
If you have not yet switched to the new `tls` stanza and were NOT using HTTPS
for the API then updating to Consul 1.13 will activate TLS for gRPC since the
deprecated TLS settings are used as defaults.
In Consul 1.13, auto-encrypt and auto-config both cause Consul
to require TLS for gRPC communication with Envoy proxies.
In environments where Envoy proxies are not already configured
to use TLS for gRPC, upgrading Consul 1.13 will cause
Envoy proxies to disconnect from the control plane (Consul agents).
The underlying cause is the same as discussed in
[deployments without the HTTPS port enabled on Consul agents](#service-mesh-deployments-without-the-https-port-enabled-on-consul-agents).
However, when using auto-encrypt or auto-config,
the problem **cannot** currently be avoided by
[modifying the agent's TLS configuration](#modify-the-consul-agent-s-tls-configuration)
because auto-encrypt and auto-config automatically set
interface-generic TLS configuration in a manner similar to
[`tls.defaults`](/docs/agent/config/config-files#tls_defaults).
We are working to address this problem in an upcoming 1.13 patch release.
#### Service mesh deployments without the HTTPS port enabled on Consul agents ((#grpc-tls))
If the HTTPS port is not enabled
([`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port))
on a pre-1.13 Consul agent,
**[modify the agent's TLS configuration before upgrading](#modify-the-consul-agent-s-tls-configuration)**
to avoid Envoy proxies disconnecting from the control plane (Consul agents).
Envoy proxies include service mesh sidecars and gateways.
##### Changes to gRPC and HTTP interface configuration
If a Consul agent's HTTP API is exposed externally,
enabling HTTPS (TLS encryption for HTTP) is important.
The gRPC interface is used for xDS communication between Consul and
Envoy proxies when using Consul service mesh.
A Consul agent's gRPC traffic is often loopback-only,
which TLS encryption is not important for.
Prior to Consul 1.13, if [`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port)
was configured, TLS was enabled for both HTTP *and* gRPC.
This was inconvenient for deployments that
needed TLS for HTTP, but not for gRPC.
Enabling HTTPS also required launching Envoy proxies
with the necessary TLS material for xDS communication
with its Consul agent via TLS over gRPC.
Consul 1.13 addresses this inconvenience by fully decoupling the TLS configuration for HTTP and gRPC interfaces.
TLS for gRPC is no longer enabled by setting
[`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port).
TLS configuration for gRPC is now determined exclusively by:
1. [`tls.grpc`](/docs/agent/config/config-files#tls_grpc), which overrides
1. [`tls.defaults`](/docs/agent/config/config-files#tls_defaults), which overrides
1. [Deprecated TLS options](/docs/agent/config/config-files#tls_deprecated_options) such as
[`ca_file`](/docs/agent/config/config-files#ca_file-4),
[`cert_file`](/docs/agent/config/config-files#cert_file-4), and
[`key_file`](/docs/agent/config/config-files#key_file-4).
This decoupling has a side effect that requires a
[TLS configuration change](#modify-the-consul-agent-s-tls-configuration)
for pre-1.13 agents without the HTTPS port enabled.
Without a TLS configuration change,
Consul 1.13 agents may now expect gRPC *with* TLS,
causing communication to fail with Envoy proxies
that continue to use gRPC *without* TLS.
##### Modify the Consul agent's TLS configuration
If [`tls.grpc`](/docs/agent/config/config-files#tls_grpc),
[`tls.defaults`](/docs/agent/config/config-files#tls_defaults),
or the [deprecated TLS options](/docs/agent/config/config-files#tls_deprecated_options)
define TLS material in their
`ca_file`, `ca_path`, `cert_file`, or `key_file` fields,
TLS for gRPC will be enabled in Consul 1.13, even if
[`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port)
is not set.
This will cause Envoy proxies to disconnect from the control plane
after upgrading to Consul 1.13 if associated pre-1.13 Consul agents
have **not** set
[`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port).
To avoid this problem, make the following agent configuration changes:
1. Remove TLS material from the Consul agents'
interface-generic TLS configuration options:
[`tls.defaults`](/docs/agent/config/config-files#tls_grpc) and
[deprecated TLS options](/docs/agent/config/config-files#tls_deprecated_options)
1. Reapply TLS material to the non-gRPC interfaces that need it with the
interface-specific TLS configuration stanzas
[introduced in Consul 1.12](/docs/upgrading/upgrade-specific#tls-configuration):
[`tls.https`](/docs/agent/config/config-files#tls_https) and
[`tls.internal_rpc`](/docs/agent/config/config-files#tls_internal_rpc).
If upgrading directly from pre-1.12 Consul,
the above configuration change cannot be made before upgrading.
Therefore, consider upgrading agents to Consul 1.12 before upgrading to 1.13.
If pre-1.13 Consul agents have set
[`ports { https = POSITIVE_INTEGER }`](/docs/agent/config/config-files#https_port),
this configuration change is not required to upgrade.
That setting means the pre-1.13 Consul agent requires TLS for gRPC *already*,
and will continue to do so after upgrading to 1.13.
If your pre-1.13 service mesh is working, you have already
configured your Envoy proxies to use TLS for gRPC when bootstrapping Envoy
via [`consul connect envoy`](/commands/connect/envoy),
such as with flags or environment variables like
[`-ca-file`](/commands/connect/envoy#ca-file) and
[`CONSUL_CACERT`](/commands#consul_cacert).
### 1.9 Telemetry Compatibility