How do I stop orchestrated jobs from hitting cloud API rate limits (HTTP 429)?

HTTP 429 means you started more requests in a period than the provider allows. Retrying with backoff recovers individual calls, but the reliable fix is to limit how fast jobs start in the first place, using a rate limit shared by everything that calls the same API: a token bucket if short bursts are fine, a leaky bucket for an even flow, or a sliding window for a hard "at most N per period" rule.

Why it happens in scheduled work

Schedules cluster on round times. Ten tasks scheduled at 02:00, five dependency children released when one parent finishes, and a batch of retries can all hit the same API within a second. Each task is fine alone; together they exceed the quota. Azure OpenAI deployments, Power BI refreshes, the Databricks Jobs API and most cloud management APIs all enforce limits like this.

Three ways to limit starts

Algorithm Behaviour Use when
Token bucket Allows a burst up to the bucket size, then releases at the refill rate The provider tolerates short bursts but enforces an average
Leaky bucket Releases at a steady rate regardless of how many are waiting The provider is sensitive to spikes
Sliding window At most N releases in any rolling window of T seconds The quota is stated as "N requests per minute"

Applying it

  • In your own code: put a limiter in front of the API client, shared across every process that calls it.
  • In Polysync: create a Rate Limit contention profile with one of the three algorithms, link it to the platform, and choose what happens when the limit is reached: wait in the queue (default), reject immediately, or reject only low-priority runs. Every Polysync task on that platform shares the same budget, whatever started it; calls made outside Polysync are not counted.

Combine it with a concurrency limit if the service also has a cap on simultaneous jobs.

Retries still matter, carefully

A 429 is a transient error: retrying after a delay normally succeeds. Use exponential backoff with a maximum delay, and make sure a retry cannot repeat a side effect. See when a retry is safe.

Frequently asked questions

Is retrying enough to handle 429 errors?

For occasional spikes, yes. If 429s are routine, retries add load at the worst moment and stretch the schedule. Limit the start rate instead.

What is the difference between a rate limit and a concurrency limit?

A rate limit controls how many runs start per period. A concurrency limit controls how many are active at once. Long jobs are usually constrained by concurrency; short API calls by rate.

Does the rate limit apply to manual runs and retries?

In Polysync, yes. Every run is queued and checked against the platform's linked profiles before it is released.

Microsoft, Azure, Azure Data Factory, Microsoft Fabric, Power BI, Logic Apps and Microsoft Marketplace are trademarks of the Microsoft group of companies. Apache Airflow and Airflow are trademarks of the Apache Software Foundation. Databricks and Lakeflow are trademarks of Databricks, Inc. All other product names are the property of their respective owners and are used for identification only. No endorsement by their owners is implied.

Data pipeline orchestration for Azure-first teams

Keep your ADF, Databricks, Fabric and Power BI jobs. Polysync connects them with shared scheduling, cross-platform dependencies and one run history, including jobs on AWS and Google Cloud. No orchestration code to write.

Learn more about Polysync