diff --git a/check_vllm.py b/check_vllm.py index 55654e2..f1aaba6 100644 --- a/check_vllm.py +++ b/check_vllm.py @@ -48,37 +48,33 @@ def main(critical, warning, timeout, target_url, does_not_contain, verify): text_result = f"failed with status code {response.status_code}\n{response.text}" exit_code = nagios.STATE_CRIT else: - if not json_data.get('text'): + if not json_data.get('text') or not len(json_data['text']): text_result = f"did not receive valid JSON\n{response.text}" exit_code = nagios.STATE_CRIT else: - if not len(json_data['text']): - text_result = f"did not receive valid JSON\n{response.text}" + tokenizer = tiktoken.get_encoding("cl100k_base") + num_tokens = len(tokenizer.encode(json_data['text'][0])) + perfdata.update({ + 'tokens': { + 'value': num_tokens, + 'min': 0, + } + }) + if not len(json_data['text'][0]): + text_result = f"response was empty.\n{response.text}" + exit_code = nagios.STATE_CRIT + elif elapsed_time >= warning: + text_result = f"response was {elapsed_time} seconds" + exit_code = nagios.STATE_WARN + elif elapsed_time >= critical: + text_result = f"response was {elapsed_time} seconds" + exit_code = nagios.STATE_CRIT + elif does_not_contain and does_not_contain in json_data['text'][0]: + text_result = f"\"{does_not_contain}\" was in the response:\n{json_data['text'][0]}" exit_code = nagios.STATE_CRIT else: - tokenizer = tiktoken.get_encoding("cl100k_base") - num_tokens = len(tokenizer.encode(json_data['text'][0])) - perfdata.update({ - 'tokens': { - 'value': num_tokens, - 'min': 0, - } - }) - if not len(json_data['text'][0]): - text_result = f"response was empty.\n{response.text}" - exit_code = nagios.STATE_CRIT - elif elapsed_time >= warning: - text_result = f"response was {elapsed_time} seconds" - exit_code = nagios.STATE_WARN - elif elapsed_time >= critical: - text_result = f"response was {elapsed_time} seconds" - exit_code = nagios.STATE_CRIT - elif does_not_contain and does_not_contain in json_data['text'][0]: - text_result = f"\"{does_not_contain}\" was in the response:\n{json_data['text'][0]}" - exit_code = nagios.STATE_CRIT - else: - text_result = f"generation time was {elapsed_time} seconds and generated {num_tokens} tokens" - exit_code = nagios.STATE_OK + text_result = f"generation time was {elapsed_time} seconds and generated {num_tokens} tokens" + exit_code = nagios.STATE_OK else: perfdata = {} text_result = f"CRITICAL - request failed - {error.__class__.__name__}: {error}"