Local rate limit

The HTTP local rate limit filter applies a token bucket rate limit when the request’s route or virtual host has a per filter local rate limit configuration.

If the local rate limit token bucket is checked, and there are no tokens available, a 429 response is returned (the response is configurable). The local rate limit filter then sets the x-envoy-ratelimited response header. Additional response headers can be configured to be returned.

Request headers can be configured to be added to forwarded requests to the upstream when the local rate limit filter is enabled but not enforced.

Depending on the value of the config local_rate_limit_per_downstream_connection, the token bucket is either shared across all workers or on a per connection basis. This results in the local rate limits being applied either per Envoy process or per downstream connection. By default the rate limits are applied per Envoy process.

Example configuration

Example filter configuration for a globally set rate limiter (e.g.: all vhosts/routes share the same token bucket):

13          http_filters:
14          - name: envoy.filters.http.local_ratelimit
15            typed_config:
16              "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit
17              stat_prefix: http_local_rate_limiter
18              token_bucket:
19                max_tokens: 10000
20                tokens_per_fill: 1000
21                fill_interval: 1s
22              filter_enabled:
23                runtime_key: local_rate_limit_enabled
24                default_value:
25                  numerator: 100
26                  denominator: HUNDRED
27              filter_enforced:
28                runtime_key: local_rate_limit_enforced
29                default_value:
30                  numerator: 100
31                  denominator: HUNDRED
32              response_headers_to_add:
33                - append: false
34                  header:
35                    key: x-local-rate-limit
36                    value: 'true'
37              local_rate_limit_per_downstream_connection: false

Example filter configuration for a globally disabled rate limiter but enabled for a specific route:

13          http_filters:
14          - name: envoy.filters.http.local_ratelimit
15            typed_config:
16              "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit
17              stat_prefix: http_local_rate_limiter

The route specific configuration:

21          route_config:
22            name: local_route
23            virtual_hosts:
24            - name: local_service
25              domains: ["*"]
26              routes:
27              - match: { prefix: "/path/with/rate/limit" }
28                route: { cluster: service_protected_by_rate_limit }
29                typed_per_filter_config:
30                  envoy.filters.http.local_ratelimit:
31                    "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit
32                    stat_prefix: http_local_rate_limiter
33                    token_bucket:
34                      max_tokens: 10000
35                      tokens_per_fill: 1000
36                      fill_interval: 1s
37                    filter_enabled:
38                      runtime_key: local_rate_limit_enabled
39                      default_value:
40                        numerator: 100
41                        denominator: HUNDRED
42                    filter_enforced:
43                      runtime_key: local_rate_limit_enforced
44                      default_value:
45                        numerator: 100
46                        denominator: HUNDRED
47                    response_headers_to_add:
48                      - append: false
49                        header:
50                          key: x-local-rate-limit
51                          value: 'true'
52              - match: { prefix: "/" }
53                route: { cluster: default_service }

Note that if this filter is configured as globally disabled and there are no virtual host or route level token buckets, no rate limiting will be applied.

Using rate limit descriptors for local rate limiting

Rate limit descriptors can be used to override local per-route rate limiting. A route’s rate limit action is used to match up a local descriptor in the filter config descriptor list. The local descriptor’s token bucket settings are then used to decide if the request should be rate limited or not depending on whether the local descriptor’s entries match the route’s rate limit actions descriptor entries. If there is no matching descriptor entries, the default token bucket is used. All the matched local descriptors will be sorterd by tokens per second and try to consume tokens in order, in most cases if one of them is limited, the remaining descriptors will not consume their tokens. However, in some cases, it may not work, for example, we have two descriptors A and B, A is limited 3 requests per second, and B is limited 20 requests per 10 seconds. Obviously B is stricter than A (token per second), as a result, if we send requests above 3 in a second, the limited requests from A will also consume tokens of B. Note that global tokens are not sorted, so we suggest they should be larger than other descriptors.

Example filter configuration using descriptors:

21          route_config:
22            name: local_route
23            virtual_hosts:
24            - name: local_service
25              domains: ["*"]
26              routes:
27              - match: { prefix: "/foo" }
28                route:
29                  cluster: service_protected_by_rate_limit
30                  rate_limits:
31                  - actions: # any actions in here
32                    - request_headers:
33                        header_name: x-envoy-downstream-service-cluster
34                        descriptor_key: client_cluster
35                    - request_headers:
36                        header_name: ":path"
37                        descriptor_key: path
38                typed_per_filter_config:
39                  envoy.filters.http.local_ratelimit:
40                    "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit
41                    stat_prefix: test
42                    token_bucket:
43                      max_tokens: 1000
44                      tokens_per_fill: 1000
45                      fill_interval: 60s
46                    filter_enabled:
47                      runtime_key: test_enabled
48                      default_value:
49                        numerator: 100
50                        denominator: HUNDRED
51                    filter_enforced:
52                      runtime_key: test_enforced
53                      default_value:
54                        numerator: 100
55                        denominator: HUNDRED
56                    response_headers_to_add:
57                    - append: false
58                      header:
59                        key: x-test-rate-limit
60                        value: 'true'
61                    descriptors:
62                    - entries:
63                      - key: client_cluster
64                        value: foo
65                      - key: path
66                        value: /foo/bar
67                      token_bucket:
68                        max_tokens: 10
69                        tokens_per_fill: 10
70                        fill_interval: 60s
71                    - entries:
72                      - key: client_cluster
73                        value: foo
74                      - key: path
75                        value: /foo/bar2
76                      token_bucket:
77                        max_tokens: 100
78                        tokens_per_fill: 100
79                        fill_interval: 60s
80              - match: { prefix: "/" }
81                route: { cluster: default_service }

In this example, requests are rate-limited for routes prefixed with “/foo” as follow. If requests come from a downstream service cluster “foo” for “/foo/bar” path, then 10 req/min are allowed. But if they come from a downstream service cluster “foo” for “/foo/bar2” path, then 100 req/min are allowed. Otherwise, 1000 req/min are allowed.

Statistics

The local rate limit filter outputs statistics in the <stat_prefix>.http_local_rate_limit. namespace. 429 responses – or the configured status code – are emitted to the normal cluster dynamic HTTP statistics.

Name

Type

Description

enabled

Counter

Total number of requests for which the rate limiter was consulted

ok

Counter

Total under limit responses from the token bucket

rate_limited

Counter

Total responses without an available token (but not necessarily enforced)

enforced

Counter

Total number of requests for which rate limiting was applied (e.g.: 429 returned)

Runtime

The HTTP rate limit filter supports the following runtime fractional settings:

http_filter_enabled

% of requests that will check the local rate limit decision, but not enforce, for a given route_key specified in the local rate limit configuration. Defaults to 0.

http_filter_enforcing

% of requests that will enforce the local rate limit decision for a given route_key specified in the local rate limit configuration. Defaults to 0. This can be used to test what would happen before fully enforcing the outcome.