Errors, Limits & ReliabilityRetry Behavior

Retry Behavior

How Protecto handles internal retries, and recommended client retry strategies for each error type.

Server-side retry behavior

Protecto retries transient internal failures automatically before returning an error.

ConditionRetry policy
Internal processing errorsUp to 3 internal retries
Client errors (4xx)No retry — client must fix the request
Rate limit (429)No retry — client must implement backoff
Async failuresExposed as FAILED status — client decides
Error typeClient action
400 Bad RequestFix the request payload — do not retry as-is
401 UnauthorizedRefresh or replace the auth token, then retry
429 Too Many RequestsRetry with exponential backoff
500 Internal Server ErrorInspect the error message. Do not retry Permission Denied or Toxicity output is prohibited; correct permissions or policy configuration first. Retry other transient server failures once after a short delay.
Async FAILEDInspect error_message before deciding to retry

The HTTP status alone does not determine retryability. Inspect the error message before retrying any 500 Internal Server Error, especially for permission or policy failures.

Exponential backoff example

import time

def mask_with_retry(payload, max_retries=3):
    for attempt in range(max_retries):
        response = protecto.mask(payload)
        if response.status_code == 429:
            wait = 2 ** attempt  # 1s, 2s, 4s
            time.sleep(wait)
            continue
        return response
    raise Exception("Rate limit exceeded after retries")

Reliability guarantees

  • Async APIs expose clear lifecycle states: PENDING, IN-PROGRESS, SUCCESS, FAILED, PURGED
  • All failures include actionable error messages
  • No partial success responses — requests either succeed fully or fail entirely
  • Token determinism ensures the same input produces the same token if retried (safe to retry masking)

Because masking is deterministic, it is always safe to retry a failed mask request with the same payload. You will get the same token even if the first request partly succeeded before failing.

Was this page helpful?