Fix engine init num_gpu_blocks logging

Avoid exception but still needs more work to be functional with multiple api server procs.

Signed-off-by: Nick Hill <nhill@redhat.com>

Author: njhill

vllm/v1/engine/async_llm.py

@@ -124,8 +124,9 @@ def __init__(
             client_addresses=client_addresses,
             client_index=client_index,
         )
-        for stat_logger in self.stat_loggers[0]:
-            stat_logger.log_engine_initialized()
+        if self.stat_loggers:
+            for stat_logger in self.stat_loggers[0]:
+                stat_logger.log_engine_initialized()
         self.output_handler: Optional[asyncio.Task] = None
         try:
             # Start output handler eagerly if we are in the asyncio eventloop.

vllm/v1/metrics/loggers.py

@@ -135,10 +135,11 @@ def log(self):
             self.spec_decoding_logging.log(log_fn=log_fn)
 
     def log_engine_initialized(self):
-        logger.info(
-            "vllm cache_config_info with initialization " \
-            "after num_gpu_blocks is: %d",
-            self.vllm_config.cache_config.num_gpu_blocks)
+        if self.vllm_config.cache_config.num_gpu_blocks:
+            logger.info(
+                "Engine %03d: vllm cache_config_info with initialization "
+                "after num_gpu_blocks is: %d", self.engine_index,
+                self.vllm_config.cache_config.num_gpu_blocks)
 
 
 class PrometheusStatLogger(StatLoggerBase):