Configuring your Mesh and multi-tenancy
This resource describes a very important concept in Kuma, and that is the ability of creating multiple isolated service meshes within the same Kuma cluster which in turn make Kuma a very simple and easy project to operate in environments where more than one mesh is required based on security, segmentation or governance requirements.
Typically, we would want to create a Mesh
per line of business, per team, per application or per environment or for any other reason. Typically multiple meshes are being created so that a service mesh can be adopted by an organization with a gradual roll-out that doesn’t require all the teams and their applications to coordinate with each other, or as an extra layer of security and segmentation for our services so that - for example - policies applied to one Mesh
do not affect another Mesh
.
Mesh
is the parent resource of every other resource in Kuma, including:
In order to use Kuma at least one Mesh
must exist, and there is no limit to the number of Meshes that can be created. When a data plane proxy connects to the control plane (kuma-cp
) it specifies to what Mesh
resource it belongs: a data plane proxy can only belong to one Mesh
at a time.
When starting a new Kuma cluster from scratch a default
Mesh is being created automatically.
Besides the ability of being able to create virtual service mesh, a Mesh
resource will also be used for:
-
Mutual TLS, to secure and encrypt our service traffic and assign an identity to the data plane proxies within the Mesh.
- Zone Egress, to setup if
ZoneEgress
should be used for cross zone and external service communication. - Non-mesh traffic, to setup if
passthrough
mode should be used for the non-mesh traffic.
To support cross-mesh communication an intermediate API Gateway must be used. Kuma checkout Kuma’s builtin gateway to set this up.
Previously, observability and locality awareness were configured within the Mesh
object.
However, for enhanced flexibility and granular control, these configurations have been extracted into separate policies:
MeshAccessLog
, MeshTrace
and MeshMetric
for observability, and MeshLoadBalancingStrategy
for locality awareness.
This separation allows for more fine-grained adjustments of each aspect, ensuring that observability and locality awareness are tailored to specific requirements.
Usage
The easiest way to create a Mesh
is to specify its name
. The name of a Mesh must be unique.
apiVersion: kuma.io/v1alpha1
kind: Mesh
metadata:
name: default
We will apply the configuration with kubectl apply -f [..]
.
Creating resources in a Mesh
It is possible to determine to what Mesh
other resources belong to in the following ways.
Data plane proxies
Every time we start a data plane proxy, we need to specify to what Mesh
it belongs, this can be done in the following way:
By using the kuma.io/mesh
annotation in a Deployment
, like:
apiVersion: apps/v1
kind: Deployment
metadata:
name: example-app
namespace: kuma-example
spec:
...
template:
metadata:
...
annotations:
# indicate to Kuma what is the Mesh that the data plane proxy belongs to
kuma.io/mesh: default
spec:
containers:
...
A Mesh
may span multiple Kubernetes namespaces. Any Kuma resource in the cluster which
specifies a particular Mesh
will be part of that Mesh
.
You can control which data plane proxies are allowed to join the mesh using mesh constraints.
Policies
When creating new Policies we also must specify to what Mesh
they belong. This can be done in the following way:
By using the kuma.io/mesh
label, like:
apiVersion: kuma.io/v1alpha1
kind: MeshHTTPRoute
metadata:
name: route-1
namespace: kuma-system
labels:
kuma.io/mesh: default # indicate to Kuma what is the Mesh that the resource belongs to
spec:
...
Kuma consumes all Policies on the cluster and joins each to an individual Mesh
, identified by this property.
Skipping default resource creation
By default, to help users get started we create the following default policies:
apiVersion: kuma.io/v1alpha1
kind: MeshTimeout
metadata:
name: mesh-gateways-timeout-all-default
namespace: kuma-system
labels:
kuma.io/mesh: default
spec:
targetRef:
kind: MeshSubset
proxyTypes:
- Gateway
to:
- targetRef:
kind: Mesh
default:
idleTimeout: 1h
http:
streamIdleTimeout: 5s
from:
- targetRef:
kind: Mesh
default:
idleTimeout: 5m
http:
streamIdleTimeout: 5s
requestHeadersTimeout: 500ms
---
apiVersion: kuma.io/v1alpha1
kind: MeshTimeout
metadata:
name: mesh-timeout-all-default
namespace: kuma-system
labels:
kuma.io/mesh: default
spec:
targetRef:
kind: Mesh
proxyTypes:
- Sidecar
to:
- targetRef:
kind: Mesh
default:
connectionTimeout: 5s
idleTimeout: 1h
http:
requestTimeout: 15s
streamIdleTimeout: 30m
from:
- targetRef:
kind: Mesh
default:
connectionTimeout: 10s
idleTimeout: 2h
http:
requestTimeout: 0s
streamIdleTimeout: 1h
maxStreamDuration: 0s
---
apiVersion: kuma.io/v1alpha1
kind: MeshRetry
metadata:
name: mesh-retry-all-default
namespace: kuma-system
labels:
kuma.io/mesh: default
spec:
targetRef:
kind: Mesh
to:
- targetRef:
kind: Mesh
default:
tcp:
maxConnectAttempt: 5
http:
numRetries: 5
perTryTimeout: 16s
backOff:
baseInterval: 25ms
maxInterval: 250ms
grpc:
numRetries: 5
perTryTimeout: 16s
backOff:
baseInterval: 25ms
maxInterval: 250ms
---
apiVersion: kuma.io/v1alpha1
kind: MeshCircuitBreaker
metadata:
name: mesh-circuit-breaker-all-default
namespace: kuma-system
labels:
kuma.io/mesh: default
spec:
targetRef:
kind: Mesh
to:
- targetRef:
kind: Mesh
default:
connectionLimits:
maxConnections: 1024
maxPendingRequests: 1024
maxRetries: 3
maxRequests: 1024
If you want to not have these policies be added on creation of the mesh set the configuration: skipCreatingInitialPolicies
:
apiVersion: kuma.io/v1alpha1
kind: Mesh
metadata:
name: default
spec:
skipCreatingInitialPolicies: ['*']
You can also skip creating the default mesh by setting the control-plane configuration: KUMA_DEFAULTS_SKIP_MESH_CREATION=true
.