An integration can work during a demonstration and fail during a busy morning because a provider accepts only a limited request rate. Repeatedly sending faster makes recovery harder. Ask how the connection slows down, preserves unfinished work and tells staff what is waiting.
Related service: Custom software development- 01Measure the allowed request budget.
- 02Queue work with an identity and priority.
- 03Pause according to provider feedback.
- 04Resume with bounded retry behavior.
- 05Escalate an aging or growing backlog.
SECTION 01
Explain the request budget in business terms
Backpressure means reducing incoming or outgoing work when the next system cannot keep up. The allowed rate may apply to an account, credential, endpoint or shared application. Ask the agency which budget actually governs your connection and whether other integrations consume the same allowance.
GitHub's REST guidance illustrates provider-specific waiting instructions and rate-limit headers. Azure's retry guidance recommends bounded retries and distinguishes recoverable problems from permanent ones. The implementation should follow the connected provider's rules, not assume that every error wants an immediate retry or that every service uses the same reset period. Authentication and invalid-data failures need different handling from temporary throttling.
SECTION 02
Illustrative example: a catalogue refresh meets live orders
Imagine an integration updating 8,000 product descriptions while new orders also need to reach an operations system. Both activities share a request allowance. If catalogue refreshes consume the whole budget, staff may wait for current orders even though the connection appears busy and healthy.
A hypothetical first release reserves capacity for order updates and processes catalogue changes in controlled batches. It shows the oldest queued order and remaining catalogue work separately. Superseded product edits may be safely combined if only the latest state matters; order actions usually cannot be discarded on the same assumption. The business owner approves those distinctions before the agency implements them.
SECTION 03
Test a slow recovery, not only a quick retry
An established connector may already provide request throttling. Check whether it exposes queue age, priority and permanent failures before building a replacement. A simple bounded queue can be enough; a complex scheduling platform is unnecessary when a few predictable jobs fit comfortably within the available request budget.
Ask the agency to simulate repeated throttling, a long outage and a workload larger than normal. Confirm that the system follows retry instructions, avoids synchronized retry storms and stops after its defined limits. Test worker restarts without dropping accepted items. Agree what happens when the backlog exceeds capacity: pause new imports, reject them clearly or request an operational decision. Quietly accumulating unlimited work is not a recovery plan.
SECTION 04
Frequently asked questions
Can we solve throttling by adding more workers?
Usually not when all workers share the same provider allowance. More concurrency can consume the budget faster and increase repeated failures.
Should every queued update be preserved separately?
Not always. Some replaceable state updates can be combined, while commands with independent business meaning need separate handling. Agree this per workflow.
What should staff see during a backlog?
Show affected work, oldest waiting time, relevant failure reasons and who owns recovery. Distinguish a temporary delay from a connection that requires intervention.
PRIMARY REFERENCES
Sources and further reading
These references cover the standards, platforms, or published prices discussed in the guide. Worked examples and checklists are our editorial guidance.
- REST API rate-limit handlingGitHub Docs
- Retry pattern and failure handlingMicrosoft Learn
EDITORIAL METHOD
About this guide
We use AI to assist with drafting and editing. Catapult AI Work is responsible for the published content. Examples illustrate possible approaches; they are not client case studies unless identified as such.
Budget examples are not Catapult package prices. Check linked provider pages for current fees and plan limits before making a purchase.
Read the editorial policy