package envoy.extensions.filters.http.bandwidth_share.v3
Get desktop application:
View/edit binary Protocol Buffers messages
[#next-free-field: 9]
optional request_limit = 1
Configuration for request limiting. If unset, requests are not limited. If ``kbps`` is zero, requests are not limited.
optional response_limit = 2
Configuration for response limiting. If unset, responses are not limited. If ``kbps`` is zero, responses are not limited.
optional fill_interval = 3
Optional fill interval in milliseconds for the token refills. Defaults to 50ms. It must be at least 20ms to avoid too aggressive refills. Lower values will use more CPU and provide smoother bandwidth distribution when the limit is being enforced.
bool enable_response_trailers = 4
Enable response trailers. .. note:: If set true, the following 4 trailers will be added, prefixed by ``response_trailer_prefix``: * bandwidth-request-delay-ms: delay time in milliseconds added by the limiter. * bandwidth-response-delay-ms: delay time in milliseconds added by the limiter. * bandwidth-request-duration-ms: total duration of receiving and processing the request. * bandwidth-response-duration-ms: total duration of receiving and processing the response. If ``response_limit`` is unset or has ``kbps`` of 0, the trailers will not be set. If both the request and response delay time is 0, the trailers will not be set. Note that the delay time may overlap with other factors such that it may not actually increase the duration by the stated amount - for example, the source data may still be arriving into a buffer during the artificial delay, such that it's possible to fully "catch up" the lost time by reading from that buffer at a later, less-constrained time.
string response_trailer_prefix = 5
The prefix for the response trailers. Empty string is a valid prefix. If set, ``enable_response_trailers`` must be true. For example, if this value is ``x-banana-`` then the trailer keys will be ``x-banana-bandwidth-request-delay-ms``, ``x-banana-bandwidth-response-duration-ms``, etc.
optional tenant_name_selector = 6
An optional matcher which returns a string to use as a tenant name. If unset, the tenant name will always be the empty string, and all requests will have equal weight when limiting is being enforced. If tenant names are set when limiting is being enforced, traffic may be distributed unevenly - for example, assuming all requests want all possible bandwidth, if there are five streams with tenant ``foo`` and two streams with tenant ``bar``, and no custom weights, ``foo`` and ``bar`` will get an equal share of the bandwidth (half), then the requests within each tenant will get an equal share of that tenant's share, i.e. the ``foo`` requests will each get a fifth of the ``foo`` half, a tenth of the total, and the ``bar`` requests will each get half of the ``bar`` half, a quarter of the total. An example way this might be used is to have ``tenant_name_selector`` make each source IP address a tenant. Then if one source IP is trying to make 100 requests in parallel, and 99 other IPs are making one request each, the total bandwidth distribution (assuming the limit is being exceeded) would be equal per IP, versus with no tenants it would be equal per request and the greedy IP address would be getting 100x more bandwidth than the others. Note that if the limit is not consistently exceeded, all available bandwidth is still distributed - so in this case if the 99 other IPs were only using half the available bandwidth total between them, the greedy IP would still be able to get 50x more bandwidth than the rest. In this example the value is clearer - the less greedy tenants get all the bandwidth they want, and only the greedy tenant gets throttled, vs. without tenants all the requests would be throttled equally.
map<string, > tenant_configs = 7
A map from tenant names to configurations. If a tenant name is not present in the map, ``default_tenant_config`` is used.
optional default_tenant_config = 8
Optional configuration override for the default tenant. If unset, the default values are used.
Used in:
string bucket_id = 1
The unique id of the limiter. Should be set if you want to apply distinct limits on different listeners, or distinct limits for requests and responses. With matching ``bucket_id`` values, multiple listeners can share a single limit (e.g. for https and http listener that share a network bandwidth constraint). Also used for the ``bucket_id`` stats tag. The empty string is a valid ``bucket_id``.
optional kbps = 2
The limit supplied in KiB/s. .. note:: The limit is associated with the ``bucket_id``. If the same ``bucket_id`` is used in multiple listeners/routes, it is a config error for the configurations to not be identical. If the configuration is updated dynamically via xds, changing the runtime key, the ``bucket_id`` must also be changed or it is a config error. The limit can be updated dynamically via runtime. Setting the limit to zero, or unset, disables the filter.
Used in:
uint32 weight = 1
Weight of the tenant. For example, if two tenants are competing for limited bandwidth, one has weight=3 and the other has weight=1, the contested bandwidth will be distributed 3/4 to the weight=3 tenant and 1/4 to the weight=1 tenant. If unset or 0, the default weight of 1 will be used.
bool include_stats_tag = 2
True to record stats with ``tenant`` tag set to the tenant name. If false or unset, the stats tag will be empty. It is recommended not to set this to default true if tenant names are dynamically generated from untrusted sources (or otherwise with unlimited scope), to avoid creating unlimited stats cardinality.