import os _DEFAULT_TTL = 60 * 60 # 1 hour default _ENV_TTL = os.getenv("HERMES_CREDENTIAL_TTL_SECONDS", "").strip() if _ENV_TTL: try: _DEFAULT_TTL = max(60, int(_ENV_TTL)) # minimum 60 seconds except ValueError: pass EXHAUSTED_TTL_429_SECONDS = _DEFAULT_TTL EXHAUSTED_TTL_DEFAULT_SECONDS = _DEFAULT_TTL

TL;DR

Configure the credential pool exhaustion TTL using an environment variable or config.yaml to prevent hardcoded 1-hour cooldowns and reduce fallback cascade risks.

Guidance

• Introduce an environment variable, such as HERMES_CREDENTIAL_TTL_SECONDS, to override the default 1-hour TTL, allowing for more flexible cooldown durations.
• Alternatively, add a credential_pool section to config.yaml with an exhausted_ttl_seconds key to achieve the same configurability.
• When setting the TTL, ensure a minimum value of 60 seconds to prevent rapid retry loops and potential quota burns.
• Test the new configuration with burst traffic scenarios to verify the effectiveness of the reduced TTL in preventing fallback cascades.

Example

import os
_DEFAULT_TTL = 60 * 60  # 1 hour default
_ENV_TTL = os.getenv("HERMES_CREDENTIAL_TTL_SECONDS", "").strip()
if _ENV_TTL:
    try:
        _DEFAULT_TTL = max(60, int(_ENV_TTL))  # minimum 60 seconds
    except ValueError:
        pass
EXHAUSTED_TTL_429_SECONDS = _DEFAULT_TTL
EXHAUSTED_TTL_DEFAULT_SECONDS = _DEFAULT_TTL

Notes

The proposed solution focuses on introducing configurability for the credential pool exhaustion TTL. However, the optimal TTL value may vary depending on specific use cases and traffic patterns. Monitoring and adjusting the TTL as needed will be crucial to achieving the desired balance between preventing fallback cascades and avoiding quota burns.

Recommendation

Apply the workaround by introducing an environment variable or config.yaml configuration for the credential pool exhaustion TTL, as this provides the necessary flexibility to adapt to different scenarios and traffic conditions.