Matching API
Attention
The matching API is alpha and is currently under active development. Capabilities will be expanded over time and the configuration structures are likely to change.
Envoy makes use of a matching API to allow the various subsystems to express actions that should be performed based on incoming data.
The matching API is designed as a tree structure to allow for sublinear matching algorithms for better performance than the linear list matching as seen in Envoy’s HTTP routing. It makes heavy use of extension points to make it easy to extend to different inputs based on protocol or environment data as well as custom sublinear matchers and direct matchers.
Inputs and Matching Algorithms
Matching inputs define a way to extract the input value used for matching. The input functions are context-sensitive. For example, HTTP header inputs are applicable only in HTTP contexts, e.g. for matching HTTP requests.
HTTP Input Functions
These input functions are available for matching HTTP requests:
Network Input Functions
These input functions are available for matching TCP connections, UDP datagrams, and HTTP requests:
These input functions are available for matching TCP connections and HTTP requests:
These input functions are available for matching TCP connections:
SSL Input Functions
These input functions are available for matching TCP connections and HTTP requests:
Common Input Functions
These input functions are available in any context:
Custom Matching Algorithms
In addition to the built-in exact and prefix matchers, these custom matchers are available in some contexts:
Trie-based IP matcher applies to network inputs.
Filter Integration
Within supported environments (currently only HTTP filters), a wrapper proto can be used to instantiate a matching filter associated with the wrapped structure:
static_resources:
listeners:
- address:
socket_address:
address: 0.0.0.0
port_value: 443
listener_filters:
filter_chains:
- filters:
- name: envoy.filters.network.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
stat_prefix: ingress_http
http_filters:
- name: with-matcher
typed_config:
"@type": type.googleapis.com/envoy.extensions.common.matching.v3.ExtensionWithMatcher
extension_config:
name: envoy.filters.http.fault
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.fault.v3.HTTPFault
abort:
http_status: 503
percentage:
numerator: 0
denominator: HUNDRED
delay:
fixed_delay: 3s
percentage:
numerator: 0
denominator: HUNDRED
xds_matcher:
matcher_tree:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpRequestHeaderMatchInput
header_name: some-header
exact_match_map:
# Note this additional indirection; this is a workaround for Protobuf oneof limitations.
map:
some_value_to_match_on: # This is the header value we're trying to match against.
action:
name: skip
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.common.matcher.action.v3.SkipFilter
- name: envoy.filters.http.router
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
route_config:
virtual_hosts:
- name: default
domains: ["*"]
routes:
- match: {prefix: "/"}
route:
cluster: service_foo
clusters:
- name: service_foo
load_assignment:
cluster_name: some_service
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 127.0.0.1
port_value: 8080
The above example wraps a HTTP filter (the HTTPFault filter) in an ExtensionWithMatcher, allowing us to define a match tree to be evaluated in conjunction with evaluation of the wrapped filter. Prior to data being made available to the filter, it will be provided to the match tree, which will then attempt to evaluate the matching rules with the provided data, triggering an action if match evaluation results in an action.
In the above example, we are specifying that we want to match on the incoming request header
some-header by setting the input to
HttpRequestHeaderMatchInput
and configuring the header key to use. Using the value contained by this header, the provided
exact_match_map specifies which values we care about: we’ve configured a single value
(some_value_to_match_on) to match against. As a result, this config means that if we
receive a request which contains some-header: some_value_to_match_on as a header, the
SkipFilter
action will be resolved (causing the associated HTTP filter to be skipped). If no such header is
present, no action will be resolved and the filter will be applied as usual.
static_resources:
listeners:
- address:
socket_address:
address: 0.0.0.0
port_value: 443
listener_filters:
filter_chains:
- filters:
- name: envoy.filters.network.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
stat_prefix: ingress_http
http_filters:
- name: with-matcher
typed_config:
"@type": type.googleapis.com/envoy.extensions.common.matching.v3.ExtensionWithMatcher
extension_config:
name: envoy.filters.http.fault
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.fault.v3.HTTPFault
abort:
http_status: 503
percentage:
numerator: 0
denominator: HUNDRED
delay:
fixed_delay: 3s
percentage:
numerator: 0
denominator: HUNDRED
xds_matcher:
# The top level matcher is a matcher tree which conceptually selects one of several subtrees.
matcher_tree:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpRequestHeaderMatchInput
header_name: some-header
exact_match_map:
# Note this additional indirection; this is a workaround for Protobuf oneof limitations.
map:
some_value_to_match_on: # This is the header value we're trying to match against.
# The OnMatch resulting on matching with this branch of the exact matcher is another matcher, allowing for recursive matching.
matcher:
# The inner matcher is a matcher list, which attempts to match a list of predicates.
matcher_list:
matchers:
- predicate:
or_matcher:
predicate:
- single_predicate:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpRequestHeaderMatchInput
header_name: second-header
value_match:
exact: foo
- single_predicate:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpRequestHeaderMatchInput
header_name: second-header
value_match:
exact: bar
on_match:
action:
name: skip
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.common.matcher.action.v3.SkipFilter
- name: envoy.filters.http.router
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
route_config:
virtual_hosts:
- name: default
domains: ["*"]
routes:
- match: {prefix: "/"}
route:
cluster: service_foo
clusters:
- name: service_foo
load_assignment:
cluster_name: some_service
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 127.0.0.1
port_value: 8080
Above is a slightly more complicated example which combines a top level tree matcher with a
linear matcher. While the tree matchers provide very efficient matching, they are not very
expressive. The list matcher can be used to provide a much richer matching API, and can be combined
with the tree matcher in an arbitrary order. The example describes the following match logic: skip
the filter if some-header: skip_filter is present and second-header is set to either foo or
bar.
HTTP Filter Iteration Impact
The above example only demonstrates matching on request headers, which ends up being the simplest case due to it happening before the associated filter receives any data. Matching on other HTTP input sources is supported (e.g. response headers), but some discussion is warranted on how this works at a filter level.
Currently the match evaluation for HTTP filters does not impact control flow at all: if insufficient data is available to perform the match, callbacks will be sent to the associated filter as normal. Once sufficient data is available to match an action, this is provided to the filter. A consequence of this is that if the filter wishes to gate some behavior on a match result, it has to manage stopping the iteration on its own.
When it comes to actions such as SkipFilter, this means that if the skip condition is based on anything but the request headers, the filter might get partially applied, which might result in surprising behavior. An example of this would be to have a matching tree that attempts to skip the gRPC-Web filter based on response headers: clients assume that if they send a gRPC-Web request to Envoy, the filter will transform that into a gRPC request before proxying it upstream, then back into a gRPC-Web response on the encoding path. By skipping the filter based on response headers, the forward transformation will happen (the upstream receives a gRPC request), but the response is never converted back to gRPC-Web. As a result, the client will receive an invalid response back from Envoy. If the skip action was instead resolved on trailers, the same gRPC-Web filter would consume all the data but never write it back out (as this happens when it sees the trailers), resulting in a gRPC-Web response with an empty body.
HTTP Routing Integration
The matching API can be used with HTTP routing, by specifying a match tree as part of the virtual host and specifying a Route as the resulting action. See examples in the above sections for how the match tree can be configured.
Match Tree Validation
As the match tree structure is very flexible, some filters might need to impose additional restrictions on what kind of match trees can be used. This system is somewhat inflexible at the moment, only supporting limiting the input sources to a specific set. For example, a filter might specify that it only works with request headers: in this case a match tree that attempts to match on request trailers or response headers will fail during configuration load, reporting back which data input was invalid.
This is done for example to limit the issues talked about in the above section or to help users understand in what context a match tree can be used for a specific filter. Due to the limitations of the validation framework at the current time, it is not used for all filters.
For HTTP filters, the restrictions are specified by the filter implementation, so consult the individual filter documentation to understand whether there are restrictions in place.
For example, in the example below, the match tree could not be used with a filter that restricts the the match tree to only use HttpRequestHeaderMatchInput.
static_resources:
listeners:
- address:
socket_address:
address: 0.0.0.0
port_value: 443
listener_filters:
filter_chains:
- filters:
- name: envoy.filters.network.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
stat_prefix: ingress_http
http_filters:
- name: with-matcher
typed_config:
"@type": type.googleapis.com/envoy.extensions.common.matching.v3.ExtensionWithMatcher
extension_config:
name: envoy.filters.http.fault
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.fault.v3.HTTPFault
abort:
http_status: 503
percentage:
numerator: 0
denominator: HUNDRED
delay:
fixed_delay: 3s
percentage:
numerator: 0
denominator: HUNDRED
xds_matcher:
matcher_list:
matchers:
- predicate:
or_matcher:
predicate:
- single_predicate:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpRequestHeaderMatchInput
header_name: request-header
value_match:
exact: foo
- single_predicate:
input:
name: request-headers
typed_config:
"@type": type.googleapis.com/envoy.type.matcher.v3.HttpResponseHeaderMatchInput
header_name: response-header
value_match:
exact: bar
on_match:
action:
name: skip
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.common.matcher.action.v3.SkipFilter
- name: envoy.filters.http.router
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
route_config:
virtual_hosts:
- name: default
domains: ["*"]
routes:
- match: {prefix: "/"}
route:
cluster: service_foo
clusters:
- name: service_foo
load_assignment:
cluster_name: some_service
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 127.0.0.1
port_value: 8080