Build a reliable API integration
Plan timeouts, retries, resource mappings and verification around the effects of each API operation.
A reliable integration can explain what it requested, what IFBI accepted and what it verified. Design that record before adding automatic retries or broad provisioning loops.
Keep a small operation ledger
Record the business-event key, installation, operation, resource IDs, submission time and verified outcome. Store the minimum diagnostic detail needed to reconcile failures. Keep credentials, tokens, exported connection details and full exception data out of ordinary application logs.
For multi-step setup, retain each completed resource ID before starting the next step. A dataset created successfully is still present if canvas creation later fails.
Retry according to the operation’s effect
| Situation | Application behaviour |
|---|---|
| An ordinary read fails temporarily | Use bounded retries and a delay appropriate to the workload. |
| A token expires | Obtain a valid token and retry only when the operation’s outcome is understood. |
| A create or manual-run request times out | Reconcile existing resources or execution evidence before repeating it. |
| A save returns HTTP 200 | Validate the documented body and application error state. |
| An update changes a complete model | Preserve intended fields and relationships, then read back the result. |
Method alone is not a safe classifier. Some documented GET operations have side effects. Reconcile uncertain writes before retrying, coordinate concurrent edits and keep request volume appropriate to the installation.
Use controlled request behaviour
Set explicit connection and response timeouts. Bound parallel work and polling so your integration does not overwhelm the installation. Honour a server-provided retry instruction when present and investigate repeated failures rather than retrying indefinitely.
Cache reusable tokens according to expires_in. Coordinate renewal among concurrent workers. For configuration changes, agree an editing process and deployment window where competing updates could overwrite each other.
Verify what matters to the user
A saved GUID confirms a different milestone from a working dashboard view or a delivered report. Read configuration back after a write, test reader access after audience changes, and verify execution and destination results for reporting jobs.
Use separate mappings and credentials for development and production installations. Resource IDs and connection settings can differ between them; an exported definition is not a complete environment backup.
For a repeatable setup sequence, see customer analytics setup. For a failed request or run, use troubleshooting.
Updated 3 days ago