Route Rules Alpha 1
Configuration affecting traffic routing. Here are a few terms useful to define in the context of routing rules.
Service is a unit of an application with a unique name that other services use to refer to the functionality being called. Service instances are pods/VMs/containers that implement the service.
Service versions - In a continuous deployment scenario, for a given service, there can be multiple sets of instances running potentially different variants of the application binary. These variants are not necessarily different API versions. They could be iterative changes to the same service, deployed in different environments (prod, staging, dev, etc.). Common scenarios where this occurs include A/B testing, canary rollouts, etc. The choice of a particular version can be decided based on various criterion (headers, url, etc.) and/or by weights assigned to each version. Each service has a default version consisting of all its instances.
Source - downstream client (browser or another service) calling the Envoy proxy/sidecar (typically to reach another service).
Destination - The remote upstream service to which the Envoy proxy/sidecar is talking to, on behalf of the source service. There can be one or more service versions for a given service (see the discussion on versions above). Envoy would choose the version based on various routing rules.
Access model - Applications address only the destination service without knowledge of individual service versions. The actual choice of the version is determined by Envoy, enabling the application code to decouple itself from the evolution of dependent services.
Circuit breaker configuration for Envoy. The circuit breaker implementation is fine-grained in that it tracks the success/failure rates of individual hosts in the load balancing pool. Hosts that continually return errors for API calls are ejected from the pool for a pre-defined period of time. See Envoy’s circuit breaker and outlier detection for more details.
A simple circuit breaker can be set based on a number of criteria such as connection and request limits. For example, the following destination policy sets a limit of 100 connections to “reviews” service version “v1” backends.
metadata: name: reviews-cb-policy namespace: default spec: destination: name: reviews labels: version: v1 circuitBreaker: simpleCb: maxConnections: 100
The following destination policy sets a limit of 100 connections and 1000 concurrent requests, with no more than 10 req/connection to “reviews” service version “v1” backends. In addition, it configures hosts to be scanned every 5 mins, such that any host that fails 7 consecutive times with 5XX error code will be ejected for 15 minutes.
metadata: name: reviews-cb-policy namespace: default spec: destination: name: reviews labels: version: v1 circuitBreaker: simpleCb: maxConnections: 100 httpMaxRequests: 1000 httpMaxRequestsPerConnection: 10 httpConsecutiveErrors: 7 sleepWindow: 15m httpDetectionInterval: 5m
Describes the Cross-Origin Resource Sharing (CORS) policy, for a given service. Refer to https://developer.mozilla.org/en-US/docs/Web/HTTP/AccesscontrolCORS for further details about cross origin resource sharing. For example, the following rule restricts cross origin requests to those originating from example.com domain using HTTP POST/GET, and sets the Access-Control-Allow-Credentials header to false. In addition, it only exposes X-Foo-bar header and sets an expiry period of 1 day.
metadata: name: my-rule namespace: default spec: destination: name: ratings route: - labels: version: v1 corsPolicy: allowOrigin: - example.com allowMethods: - POST - GET allowCredentials: false allowHeaders: - X-Foo-Bar maxAge: "1d"
DestinationPolicy defines client/caller-side policies that determine how to handle traffic bound to a particular destination service. The policy specifies configuration for load balancing and circuit breakers. For example, a simple load balancing policy for the ratings service would look as follows:
metadata: name: ratings-lb-policy namespace: default # optional (default is "default") spec: destination: name: ratings loadBalancing: name: ROUND_ROBIN
The FQDN of the destination service is composed from the destination name and meta namespace fields, along with a platform-specific domain suffix (e.g. on Kubernetes, “reviews” + “default” + “svc.cluster.local” -> “reviews.default.svc.cluster.local”).
A destination policy can be restricted to a particular version of a service or applied to all versions. It can also be restricted to calls from a particular source. For example, the following load balancing policy applies to version v1 of the ratings service running in the prod environment but only when called from version v2 of the reviews service:
metadata: name: ratings-lb-policy namespace: default spec: source: name: reviews labels: version: v2 destination: name: ratings labels: env: prod version: v1 loadBalancing: name: ROUND_ROBIN
Note: Destination policies will be applied only if the corresponding tagged instances are explicitly routed to. In other words, for every destination policy defined, at least one route rule must refer to the service version indicated in the destination policy.
Each routing rule is associated with one or more service versions (see glossary in beginning of document). Weights associated with the version determine the proportion of traffic it receives. For example, the following rule will route 25% of traffic for the “reviews” service to instances with the “v2” tag and the remaining traffic (i.e., 75%) to “v1”.
metadata: name: my-rule namespace: default spec: destination: name: reviews route: - labels: version: v2 weight: 25 - labels: version: v1 weight: 75
Egress rules describe the properties of a service outside Istio. When transparent proxying is used, egress rules signify a white listed set of external services that microserves in the mesh are allowed to access. A subset of routing rules and all destination policies can be applied on the service targeted by an egress rule. TCP services and HTTP-based services can be expressed by an egress rule. The destination of an egress rule for HTTP-based services must be an IP or a domain name, optionally with a wildcard prefix (e.g., *.foo.com). For TCP based services, the destination of an egress rule must be an IP or a block of IPs in CIDR notation.
If TLS origination from the sidecar is desired, the protocol associated with the service port must be marked as HTTPS, and the service is expected to be accessed over HTTP (e.g., http://gmail.com:443). The sidecar will automatically upgrade the connection to TLS when initiating a connection with the external service.
For example, the following egress rule describes the set of services hosted under the *.foo.com domain
kind: EgressRule metadata: name: foo-egress-rule spec: destination: service: *.foo.com ports: - port: 80 protocol: http - port: 443 protocol: https
The following egress rule describes the set of services accessed by a block of IPs
kind: EgressRule metadata: name: bar-egress-rule spec: destination: service: 184.108.40.206/27 ports: - port: 111 protocol: tcp
Port describes the properties of a specific TCP port of an external service.
HTTPFaultInjection can be used to specify one or more faults to inject while forwarding http requests to the destination specified in the route rule. Fault specification is part of a route rule. Faults include aborting the Http request from downstream service, and/or delaying proxying of requests. A fault rule MUST HAVE delay or abort or both.
Note: Delay and abort faults are independent of one another, even if both are specified simultaneously.
Abort specification is used to prematurely abort a request with a pre-specified error code. The following example will return an HTTP 400 error code for 10% of the requests to the “ratings” service “v1”.
metadata: name: my-rule spec: destination: name: reviews route: - labels: version: v1 httpFault: abort: percent: 10 httpStatus: 400
The httpStatus field is used to indicate the HTTP status code to return to the caller. The optional percent field, a value between 0 and 100, is used to only abort a certain percentage of requests. If not specified, all requests are aborted.
Delay specification is used to inject latency into the request forwarding path. The following example will introduce a 5 second delay in 10% of the requests to the “v1” version of the “reviews” service.
metadata: name: my-rule spec: destination: name: reviews route: - labels: version: v1 httpFault: delay: percent: 10 fixedDelay: 5s
The fixedDelay field is used to indicate the amount of delay in seconds. An optional percent field, a value between 0 and 100, can be used to only delay a certain percentage of requests. If left unspecified, all request will be delayed.
HTTPRedirect can be used to send a 302 redirect response to the caller, where the Authority/Host and the URI in the response can be swapped with the specified values. For example, the following route rule redirects requests for /v1/getProductRatings API on the ratings service to /v1/bookRatings provided by the bookratings service.
metadata: name: my-rule namespace: default spec: destination: name: ratings match: request: headers: uri: /v1/getProductRatings redirect: uri: /v1/bookRatings authority: bookratings.default.svc.cluster.local
Describes the retry policy to use when a HTTP request fails. For example, the following rule sets the maximum number of retries to 3 when calling ratings:v1 service, with a 2s timeout per retry attempt.
metadata: name: my-rule namespace: default spec: destination: name: ratings route: - labels: version: v1 httpReqRetries: simpleRetry: attempts: 3 perTryTimeout: 2s
HTTPRewrite can be used to rewrite specific parts of a HTTP request before forwarding the request to the destination. Rewrite primitive can be used only with the DestinationWeights. The following example demonstrates how to rewrite the URL prefix for api call (/ratings) to ratings service before making the actual API call.
metadata: name: my-rule namespace: default spec: destination: name: ratings match: request: headers: uri: prefix: /ratings rewrite: uri: /v1/bookRatings route: - labels: version: v1
Describes HTTP request timeout. For example, the following rule sets a 10 second timeout for calls to the ratings:v1 service
metadata: name: my-rule namespace: default spec: destination: name: ratings route: - labels: version: v1 httpReqTimeout: simpleTimeout: timeout: 10s
Ingress rules are routing rules applied to the ingress proxy pool. The ingress proxes serve as the receiving edge proxy for the entire mesh, but can also be addressed from inside the mesh. Each ingress rule defines a destination service and port. Rules that do not resolve to a service or a port in the mesh should be ignored.
The routing rules for the destination service are applied at the ingress proxy. That means the routing rule match conditions are composed and its actions are enforced. The traffic splitting for the destination service is also effective.
WARNING: This API is experimental and under active development
IstioService identifies a service and optionally service version. The FQDN of the service is composed from the name, namespace, and implementation-specific domain suffix (e.g. on Kubernetes, “reviews” + “default” + “svc.cluster.local” -> “reviews.default.svc.cluster.local”).
(– Faults can be injected into the connections from downstream by the Envoy, for testing the failure recovery capabilities of downstream services. Faults include aborting the connection from downstream service, delaying proxying of connection to the destination service, and throttling the bandwidth of the connection (either end). Bandwidth throttling for failure testing should not be confused with the rate limiting policy enforcement provided by the Mixer component. L4 fault injection is not supported at the moment. –)
Abruptly reset (terminate) the Tcp connection after it has been established, emulating remote server crash or link failure.
Bandwidth throttling for Tcp and Udp connections
(– L4 connection match attributes. Note that L4 connection matching support is incomplete. –)
Load balancing policy to use when forwarding traffic. These policies directly correlate to load balancer types supported by Envoy. Example,
metadata: name: reviews-lb-policy namespace: default spec: destination: name: reviews loadBalancing: name: RANDOM
Load balancing algorithms supported by Envoy.
Simple round robin policy.
The least request load balancer uses an O(1) algorithm which selects two random healthy hosts and picks the host which has fewer active requests.
The random load balancer selects a random healthy host. The random load balancer generally performs better than round robin if no health checking policy is configured.
Match condition specifies a set of criterion to be met in order for the route rule to be applied to the connection or HTTP request. The condition provides distinct set of conditions for each protocol with the intention that conditions apply only to the service ports that match the protocol. For example, the following route rule restricts the rule to match only requests originating from “reviews:v2”, accessing ratings service where the URL path starts with /ratings/v2/ and the request contains a “cookie” with value “user=jason”,
metadata: name: my-rule namespace: default spec: destination: name: ratings match: source: name: reviews labels: version: v2 request: headers: cookie: regex: "^(.*?;)?(user=jason)(;.*)?" uri: prefix: "/ratings/v2/"
MatchCondition CANNOT be empty. At least one source or request header must be specified.
MatchRequest specifies the attributes of an HTTP request to be used for matching a request.
Route rule provides a custom routing policy based on the source and destination service versions and connection/request metadata. The rule must provide a set of conditions for each protocol (TCP, UDP, HTTP) that the destination service exposes on its ports.
The rule applies only to the ports on the destination service for which it provides protocol-specific match condition, e.g. if the rule does not specify TCP condition, the rule does not apply to TCP traffic towards the destination service.
For example, a simple rule to send 100% of incoming traffic for a “reviews” service to version “v1” can be specified as follows:
metadata: name: my-rule namespace: default # optional (default is "default") spec: destination: name: reviews namespace: my-namespace # optional (default is metadata namespace field) route: - labels: version: v1 weight: 100
Describes how to match a given string in HTTP headers. Match is case-sensitive.