Description of ConcurrencyLimitingPolicy fields - Alibaba Cloud Service Mesh

The ConcurrencyLimitingPolicy CustomResourceDefinition (CRD) is part of the Service Mesh (ASM) traffic scheduling suite. It declaratively configures the maximum number of concurrent requests for global traffic of a service in an ASM instance.

How it works

The ASM traffic scheduling suite tracks in-flight requests against the max_concurrency threshold. The selectors field determines which requests the policy applies to. When limit_by_label_key is set, requests are grouped by label, and each group maintains its own independent concurrency count.

To handle edge cases such as pod restarts where a request termination event may not be recorded, max_inflight_duration sets a timeout. Requests that exceed this timeout are automatically treated as completed, preventing stale records from permanently consuming concurrency slots.

Sample configuration

The following YAML shows a minimal ConcurrencyLimitingPolicy spec:

spec:
  concurrency_limiter:
    max_concurrency: 100
    parameters:
      max_inflight_duration: 60s
    selectors:
      - service: my-service.default.svc.cluster.local
        port:
          number: 8080

max_concurrency: 100 -- Allows up to 100 concurrent in-flight requests.
max_inflight_duration: 60s -- Treats any request not completed within 60 seconds as finished, freeing its concurrency slot.
selectors -- Applies the limit to traffic that matches the specified service and port.

ConcurrencyLimitingPolicySpec

The spec section is the top-level configuration for ConcurrencyLimitingPolicy.

Field	Type	Required	Description
concurrency_limiter	ConcurrencyLimiter	Yes	Concurrency limiter configuration. See ConcurrencyLimiter.

ConcurrencyLimiter

Defines the concurrency limit, matching rules, and request handling behavior.

Field	Type	Required	Description
max_concurrency	int64	Yes	Maximum number of concurrent in-flight requests allowed.
parameters	ConcurrencyLimiterParameters	Yes	Parameters that control how the concurrency limiter groups and tracks requests. See ConcurrencyLimiterParameters.
request_parameters	RateLimiterRequestParameters	No	Actions the concurrency limiter performs on requests. Uses the same schema as the RateLimitingPolicy CRD.
selectors	[]Selector	Yes	Requests that the concurrency limit applies to.

ConcurrencyLimiterParameters

Controls how the concurrency limiter tracks in-flight requests.

Field	Type	Required	Description
limit_by_label_key	string	No	Groups requests by label for independent concurrency tracking. When set, each unique label value maintains its own concurrency limit. For details on labeling requests, see Description of request labels.
max_idle_time	Duration	No	Maximum idle time before releasing the in-flight request record for a specific label. If no requests with a given label arrive within this period, the record is released. Takes effect only when `limit_by_label_key` is set. Default: `7200s`.
max_inflight_duration	Duration	Yes	Maximum time a request is considered in-flight. After this timeout, the request is treated as completed regardless of its actual state. This prevents stale records caused by events such as pod restarts from permanently consuming concurrency slots. Set this value to the expected maximum response time of your service.