Rate Limiting in the Ambassador Edge Stack

Rate limiting in the Ambassador Edge Stack is composed of two parts:

  • Labels that get attached to requests; a label is basic metadata that is used by the RateLimitService to decide which limits to apply to the request.
  • RateLimits configure the Ambassador Edge Stack's built-in RateLimitService, and set limits based on the labels on the request.

Attaching labels to requests

There are two ways of setting labels on a request:

  1. Per Mapping. Labels set here will only apply to requests that use that Mapping

    ---
    apiVersion: getambassador.io/v2
    kind: Mapping
    metadata:
    name: foo-app
    spec:
    prefix: /foo/
    service: foo
    labels:
    "my_first_label_domain":
    - "my_first_label_group":
    - "my_label_specifier_1"
    - "my_label_specifier_2"
    - "my_second_label_group":
    - "my_label_specifier_3"
    - "my_label_specifier_4"
    "my_second_label_domain":
    - ...
  2. Globally, in the ambassadorModule. Labels set here are applied to every single request that goes through Ambassador. This includes requests go through a Mapping that sets more labels; for those requests, the global labels are prepended to each of the Mapping's label groups for the matching domain; otherwise the global labels are put in to a new label group named "default" for that domain.

    ---
    apiVersion: getambassador.io/v2
    kind: Module
    metadata:
    name: ambassador
    spec:
    config:
    default_labels:
    "my_first_label_domain":
    defaults:
    - "my_label_specifier_a"
    - "my_label_specifier_b"
    "my_second_label_domain":
    defaults:
    - "my_label_specifier_c"
    - "my_label_specifier_d"

Labels on a request are lists of key/value pairs, organized in to label groups. Because a label group is a list of key/value pairs (rather than a map),

  • it is possible to have multiple labels with the same key
  • the order of labels matters

Your Module and Mappings contain label specifiers that tell Ambassador what labels to set on the request.

Note: The terminology used by the Envoy documentation differs from the terminology used by Ambassador:

AmbassadorEnvoy
label groupdescriptor
labeldescriptor entry
label specifierrate limit action

The Mappings' listing of the groups of specifiers have names for the groups; the group names are useful for humans dealing with the YAML, but are ignored by Ambassador, all Ambassador cares about are the contents of the groupings of label specifiers.

There are 5 types of label specifiers in Ambassador:

#Label SpecifierAction, in human termsAction, in Envoy gRPC terms
1"source_cluster"Sets the label "source_cluster=«Envoy source cluster name»"{ "source_cluster": {} }
2"destination_cluster"Sets the label "destination_cluster=«Envoy destination cluster name»"{ "destination_cluster": {} }
3{ "my_key": { "header": "my_hdr" } }If the my_hdr header is set, then set the label "«my_key»=«Value of the my_hdr header»"; otherwise skip applying this label group{ "request_headers": { "header_name": "my_hdr", descriptor_key: "my_key" } }
4"remote_address"Sets the label "remote_address=«IP address of the client»"{ "remote_address": {} }
5{ "generic_key": "my_val" }Sets the label "generic_keymy_val»"{ "generic_key": { "descriptor_value": "my_val" } }
5 (shorthand)"my_val"Shorthand for { "generic_key": "my_val" }
  1. The Envoy source cluster name is the name of the Envoy listener cluster that the request name in on.
  2. The Envoy destination cluster is the name of the Envoy cluster that the Mapping routes the request to. Typically, there is a 1:1 correspondence between upstream services (pointed to by Mappings) and clusters. You can get the name for a cluster from the diagnostics service or Edge Policy Console.
  3. When setting a label from an HTTP request header, be aware that if that header is not set in the request, then the entire label group is skipped.
  4. The IP address of the HTTP client could be the actual IP of the client talking directly to Ambassador, or it could be the IP address from X-Forwarded-For if Ambassador is configured to trust the X-Fowarded-For header.
  5. generic_key allows you to apply a simple string label to requests flowing through that Mapping.

Rate limiting requests based on their labels

A RateLimit resource defines a list of limits that apply to different requests.

---
apiVersion: getambassador.io/v2
kind: RateLimit
metadata:
name: example-limits
spec:
domain: "my_domain"
limits:
- pattern:
- "my_key1": "my_value1"
"my_key2": "my_value2"
- "my_key3": "my_value3"
rate: 5
unit: "minute"
injectRequestHeaders: # optional
- name "header-name-string-1" # required
value: "go-template-string" # required
- name "header-name-string-2" # required
value: "go-template-string" # required
injectResponseHeaders: # optional
- name "header-name-string-1" # required
value: "go-template-string" # required
- pattern:
- "my_key4": "" # check the key but not the value
- "my_key5": "*" # check the key but not the value
rate: 5
unit: "second"
...

It makes no difference whether a limits are defined together in one RateLimit resource or are defined separately in many RateLimit resources.

  • pattern: Each limit has a pattern that matches against a label group on a request to decide if that limit should apply to that request. For a pattern to match, the request's label group must start with exactly the labels specified in the pattern, in order. If a label in a pattern has an empty string or "*" as the value, then it only checks the key of that label on the request; not the value. If a list item in the pattern has multiple key/value pairs, if any of them match the label then it is considered a match.

    For example, the pattern

    pattern:
    - "key1": "foo"
    "key1": "bar"
    - "key2": ""

    matches the label group

    - key1: foo
    - key2: baz
    - otherkey: knob

    and

    - key1: bar
    - key2: baz
    - otherkey: knob

    but not the label group

    - key0: frob
    - key1: foo
    - key2: baz

    If a label group is matched by multiple patterns, the pattern with the longest list of items wins.

    If a request has multiple label groups, then multiple limits may apply to that request; if any of the limits are being hit, then Ambassador will reject the request as an HTTP 429.

  • rate, unit: The limit itself is specified as an integer number of requests per a unit of time. Valid units of time are second, minute, hour, or day (all case-insensitive).

    So for example

    rate: 5
    unit: minute

    would allow 5 requests per minute, and any requests in excess of that would result in HTTP 429 errors.

  • injectRequestHeaders, injectResponseHeaders: If this limit's pattern matches the request, then injectRequestHeaders injects HTTP header fields in to the request before sending it to the upstream service (assuming the limit even allows the request to go to the upstream service), and injectResponseHeaders injects headers in to the response sent back to the client (whether the response came from the upstream service or is an HTTP 429 response because it got rate limited). This is very similar to injectRequestHeaders in a JWT Filter. The header value is specified as a Go text/template string, with the following data made available to it:

    • .RateLimitResponse.OverallCodeint : 1 for OK, 2 for OVER_LIMIT.
    • .RateLimitResponse.Statuses → [[]*RateLimitResponse_DescriptorStatus]]v2.RateLimitResponse_DescriptorStatus The itemized status codes for each limit that was selected for this request.
    • .RetryAftertime.Duration the amount of time until all of the limits would allow access again (0 if they all currently allow access).

    Also available to the template are the standard functions available to Go text/templates, as well as:

    • a hasKey function that takes the a string-indexed map as arg1, and returns whether it contains the key arg2. (This is the same as the Sprig function of the same name.)

    • a doNotSet function that causes the result of the template to be discarded, and the header field to not be adjusted. This is useful for only conditionally setting a header field; rather than setting it to an empty string or "<no value>". Note that this does not unset an existing header field of the same name.

Examples

An example service-level rate limit

The following Mapping resource will add a my_default_generic_key_label generic_key label to every request to the foo-app service:

---
apiVersion: getambassador.io/v2
kind: Mapping
metadata:
name: foo-app
spec:
prefix: /foo/
service: foo
labels:
ambassador:
- label_group:
- my_default_generic_key_label

You can then create a default RateLimit for every request that matches this label:

---
apiVersion: getambassador.io/v2
kind: RateLimit
metadata:
name: default-rate-limit
spec:
domain: ambassador
limits:
- pattern:
- generic_key: "my_default_generic_key_label"
rate: 10
unit: minute

Tip: For testing purposes, it is helpful to configure per-minute rate limits before switching the rate limits to per second or per hour.

An example with multiple labels

Mappings can have multiple labels which annotate a given request.

---
apiVersion: getambassador.io/v2
kind: Mapping
metadata:
name: catalog
spec:
prefix: /catalog/
service: catalog
labels:
ambassador: # the label domain
- string_request_label: # the label group name -- useful for humans, ignored by Ambassador
- catalog # annotate the request with `generic_key=catalog`
- header_request_label: # another label group name
- headerkey: # The name of the label
header: ":method" # annotate the request with the specific HTTP method used
- multi_request_label_group:
- authorityheader:
header: ":authority"
omit_if_not_present: true
- xuserheader:
header: "x-user"
omit_if_not_present: true

Let's digest the above example:

  • Request labels must be part of the "ambassador" label domain. Or rather, it must match the domain in your RateLimitService.spec.domain which defaults to Module.spec.default_label_domain which defaults to ambassador; but normally you should accept the default and just accept that the domain on the Mappings must be set to "ambassador".
  • Each label must have a name, e.g., one_request_label
  • The string_request_label simply adds the string catalog to every incoming request to the given mapping. The string is referenced with the key generic_key.
  • The header_request_label adds a specific HTTP header value to the request, in this case, the method. Note that HTTP/2 request headers must be used here (e.g., the host header needs to be specified as the :authority header).
  • Multiple labels can be part of a single named label, e.g., multi_request_label specifies two different headers to be added
  • When an HTTP header is not present, the entire named label is omitted. The omit_if_not_present: true is an explicit notation to remind end-users of this limitation. false is not a supported value.

An example with multiple limits

Labels can be grouped. This allows for a single request to count against multiple different RateLimit resources. For example, imagine the following scenario:

  1. Users should be limited on the total number of requests that can be sent to a set of endpoints
  2. On a specific service, stricter limits are desirable

The following Mapping resources could be configured:

---
apiVersion: getambassador.io/v2
kind: Mapping
metadata:
name: foo-app
spec:
prefix: /foo/
service: foo
labels:
ambassador:
- foo-app_label_group:
- foo-app
- total_requests_group:
- remote_address
---
apiVersion: getambassador.io/v2
kind: Mapping
metadata:
name: bar-app
spec:
prefix: /bar/
service: bar
labels:
ambassador:
- bar-app_label_group:
- bar-app
- total_requests_group:
- remote_address

Now requests to the foo-app and the bar-app would be labeled with

- "generic_key": "foo-app"
- "remote_address": "10.10.11.12"

and

- "generic_key": "bar-app"
- "remote_address": "10.10.11.12"

respectively. RateLimits on these two services could be created as such:

---
apiVersion: getambassador.io/v2
kind: RateLimit
metadata:
name: foo-rate-limit
spec:
domain: ambassador
limits:
- pattern: [{generic_key: "foo-app"}]
rate: 10
unit: second
---
apiVersion: getambassador.io/v2
kind: RateLimit
metadata:
name: bar-rate-limit
spec:
domain: ambassador
limits:
- pattern: [{generic_key: "bar-app"}]
rate: 20
unit: second
---
apiVersion: getambassador.io/v2
kind: RateLimit
metadata:
name: user-rate-limit
spec:
domain: ambassador
limits:
- pattern: [{remote_address: "*"}]
rate: 100
unit: minute

An example with global labels and groups

Global labels are prepended to every single label group. In the above example, if the following global label was added in the ambassador Module:

---
apiVersion: getambassador.io/v2
kind: Module
metadata:
name: ambassador
spec:
config:
default_label_domain: ambassador
default_labels:
ambassador:
defaults:
- "my_default_label"

The labels metadata would change

  • from
    - "generic_key": "foo-app"
    - "remote_address": "10.10.11.12"
    to
    - "generic_key": "my_default_label"
    - "generic_key": "foo-app"
    - "remote_address": "10.10.11.12"

and

  • from
    - "generic_key": "bar-app"
    - "remote_address": "10.10.11.12"
    to
    - "generic_key": "my_default_label"
    - "generic_key": "bar-app"
    - "remote_address": "10.10.11.12"

respectively.

And thus our RateLimits would need to change to appropriately handle the new labels.

Questions?

We’re here to help. If you have questions, join our Slack, contact us, or request a demo.