Performance Best Practices

At Skyflow, we want your integration to perform well at any scale. Beyond the security and implementation best practices, here are the performance recommendations we suggest as you build with Skyflow’s APIs and SDKs.

  • Batch your requests
    • Combine multiple records into a single request per record to reduce round trips.
    • Requests default to a cap of 25 records per call across operations, but this limit is configurable.
  • Reuse SDK clients and connections
    • Initialize the client once per application or service and reuse it, rather than constructing a new one per call, so the underlying HTTP layer can pool connections and keep them alive between requests.
  • Cache and reuse bearer tokens
    • Tokens are valid for 60 minutes. Server-side SDKs (Python, Java, Node, Go) cache and reuse valid tokens automatically.
    • If your SDK doesn’t provide this out of the box, implement bearer token caching on your end.
  • Handle retries and rate limits deliberately
    • Implement exponential backoff for 429 and 5xx responses, increasing the wait time on each subsequent attempt, and cap the total number of retries.
  • Use concurrency within your rate limit
    • Parallelize independent calls for bulk workloads.
    • Size your concurrency to stay under your rate limit.
  • Tune client timeouts for your environment
    • Set connect, read, and write timeouts that match your latency requirements instead of relying on long defaults.
  • Optimize the network path
    • Use AWS PrivateLink or GCP Private Service Connect for direct, private, low-latency connectivity.
    • Choose a vault region close to your application, consistent with your data residency requirements.
  • Monitor and instrument your integration
    • Track latency, error rate, and rate-limit consumption. Stream audit logs to your SIEM or observability stack to correlate API activity with performance.
  • Keep your SDKs up to date
    • Upgrade to newer SDK versions as they’re released to get functional, security, and performance fixes.