Workflow Concurrency Limits

Dapr provides concurrency limits for workflows and activities at two levels:

• Global limits control the total across all replicas, enforced by the scheduler.
• Per-sidecar limits control how many workflows or activities a single Dapr instance can execute concurrently.

Both levels can be configured independently and work together. Global limits enforce namespace-wide capacity constraints, for example, to respect rate limits on downstream services. Per-sidecar limits protect individual instances from resource exhaustion.

Prefer global limits when you want to express a true concurrency limit. Because a per-sidecar limit applies to each instance independently, the effective namespace-wide limit is the per-sidecar value multiplied by the number of replicas, so it drifts every time you scale up or down. A global limit holds the same total regardless of how many replicas are running, which is usually what you mean when you say “at most N at a time”.

Global limits

Global limits enforce a maximum across all replicas of your application. The Dapr scheduler divides the limit among its instances and holds back triggers when the limit is reached, dispatching them as capacity becomes available.

Because the total is fixed regardless of how many replicas are running, a global limit is the most accurate way to express a true concurrency limit. This is the recommended starting point for most workloads.

All workflows or all activities

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: appconfig
spec:
  workflow:
    globalMaxConcurrentWorkflowInvocations: 50
    globalMaxConcurrentActivityInvocations: 200

Per-name limits

You can set concurrency limits for specific workflow or activity names. This is useful when certain workflows or activities call rate-limited external services.

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: appconfig
spec:
  workflow:
    globalMaxConcurrentActivityInvocations: 200
    activityConcurrencyLimits:
      - name: SendEmail
        maxConcurrent: 5
      - name: CallPaymentAPI
        maxConcurrent: 10
    workflowConcurrencyLimits:
      - name: OrderProcess
        maxConcurrent: 20

A trigger must satisfy all applicable limits. For example, if globalMaxConcurrentActivityInvocations is 200 and SendEmail has a per-name limit of 5, then at most 5 SendEmail activities can run, and all activities combined cannot exceed 200.

Per-sidecar limits

Per-sidecar limits restrict concurrency within a single Dapr sidecar. Because they apply to each instance independently, the effective namespace-wide capacity scales with the number of replicas: if you have 10 replicas with a per-sidecar limit of 100, the effective namespace-wide capacity is up to 1000. For this reason, reach for per-sidecar limits to protect individual instances from resource exhaustion rather than to enforce a true namespace-wide concurrency limit.

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: appconfig
spec:
  workflow:
    maxConcurrentWorkflowInvocations: 100
    maxConcurrentActivityInvocations: 1000

These limits do not distinguish between different workflow or activity names. They apply to all workflows and activities running in the sidecar.

How the levels interact

When both per-sidecar and global limits are configured, both apply. The global limit prevents the namespace-wide total from exceeding the configured value, while the per-sidecar limit prevents any single instance from consuming too much local resources.

How global limits work with multiple scheduler replicas

The scheduler divides global limits evenly among its instances using floor division. With a global limit of 100 and 3 scheduler replicas, each scheduler enforces a local limit of 33, for an effective namespace max of 99. This ensures the configured limit is never exceeded.

Comparison with other rate limiting options

Dapr provides several ways to control concurrency and rate limiting:

Related links

• Dapr Configuration reference
• Control concurrency and rate limit applications
• Rate limit middleware
• Workflow overview
• Workflow API reference