[Deprecation] Deprecate profiling envs (#33722)

Signed-off-by: yewentao256 <zhyanwentao@126.com>

docs/usage/security.md

@@ -178,7 +178,7 @@ These endpoints are **only available when the environment variable `VLLM_SERVER_
 - `/is_sleeping` - Check if engine is sleeping
 - `/collective_rpc` - Execute arbitrary RPC methods on the engine (extremely dangerous)
 
-**Profiler endpoints (only when `VLLM_TORCH_PROFILER_DIR` or `VLLM_TORCH_CUDA_PROFILE` are set):**
+**Profiler endpoints (only when profiling is enabled via `--profiler-config`):**
 
 These endpoints are only available when profiling is enabled and should only be used for local development:
 
@@ -207,7 +207,7 @@ An attacker who can reach the vLLM HTTP server can:
 - Cache manipulation that can disrupt service
 - Detailed server configuration disclosure
 
-Similarly, never enable profiler endpoints (`VLLM_TORCH_PROFILER_DIR` or `VLLM_TORCH_CUDA_PROFILE`) in production.
+Similarly, never enable profiler endpoints in production.
 
 **Be cautious with `--enable-tokenizer-info-endpoint`:** Only enable the `/tokenizer_info` endpoint if you need to expose tokenizer configuration information. This endpoint reveals chat templates and tokenizer settings that may contain sensitive implementation details or prompt engineering strategies.
 

vllm/envs.py

@@ -855,53 +855,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
     "VLLM_LORA_RESOLVER_HF_REPO_LIST": lambda: os.getenv(
         "VLLM_LORA_RESOLVER_HF_REPO_LIST", None
     ),
-    # Enables torch CUDA profiling if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_CUDA_PROFILE": lambda: os.getenv("VLLM_TORCH_CUDA_PROFILE"),
-    # Enables torch profiler if set.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_DIR": lambda: os.getenv("VLLM_TORCH_PROFILER_DIR"),
-    # Enable torch profiler to record shapes if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_RECORD_SHAPES": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_RECORD_SHAPES")
-    ),
-    # Enable torch profiler to profile memory if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY")
-    ),
-    # Enable torch profiler to profile stack if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_WITH_STACK": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_WITH_STACK")
-    ),
-    # Enable torch profiler to profile flops if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_WITH_FLOPS": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_WITH_FLOPS")
-    ),
-    # Disable torch profiling of the AsyncLLMEngine process if set to 1.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_DISABLE_ASYNC_LLM": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_DISABLE_ASYNC_LLM")
-    ),
-    # Delay number of iterations before starting profiling when using
-    # the torch/torch CUDA profiler. If set to 0, will start profiling immediately.
-    # Deprecated, see profiler_config.
-    "VLLM_PROFILER_DELAY_ITERS": lambda: (os.getenv("VLLM_PROFILER_DELAY_ITERS")),
-    # Maximum number of iterations to profile when using the torch/torch CUDA profiler.
-    # If set to 0, will not limit the number of iterations.
-    "VLLM_PROFILER_MAX_ITERS": lambda: os.getenv("VLLM_PROFILER_MAX_ITERS"),
-    # Control whether torch profiler gzip-compresses profiling files.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_USE_GZIP": lambda: os.getenv("VLLM_TORCH_PROFILER_USE_GZIP"),
-    # Control whether torch profiler dumps the self_cuda_time_total table.
-    # Set to 0 to disable dumping the table.
-    # Deprecated, see profiler_config.
-    "VLLM_TORCH_PROFILER_DUMP_CUDA_TIME_TOTAL": lambda: (
-        os.getenv("VLLM_TORCH_PROFILER_DUMP_CUDA_TIME_TOTAL")
-    ),
     # If set, vLLM will use Triton implementations of AWQ.
     "VLLM_USE_TRITON_AWQ": lambda: bool(int(os.getenv("VLLM_USE_TRITON_AWQ", "0"))),
     # If set, allow loading or unloading lora adapters in runtime,