@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / bd98e97557
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 2, 03:30 UTC

[Misc] Remove dead VLLM_RPC_TIMEOUT env var and fix profiling doc that references it (#44128)

Signed-off-by: Daoyuan Li <94409450+DaoyuanLi2816@users.noreply.github.com>

Daoyuan Li committed Jun 2, 2026 at 17:22 UTC bd98e97557fd8f5451357f79e83d37dd3cc46289
11 files changed +1 -15
.buildkite/performance-benchmarks/tests/latency-tests-arm64-cpu.json
-1
@@ -2,7 +2,6 @@
2 {
3 "test_name": "latency_llama8B_tp1",
4 "environment_variables": {
5 - "VLLM_RPC_TIMEOUT": 100000,
5 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
6 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
7 "VLLM_CPU_KVCACHE_SPACE": 40
.buildkite/performance-benchmarks/tests/latency-tests-cpu.json
-1
@@ -2,7 +2,6 @@
2 {
3 "test_name": "latency_llama8B_tp2",
4 "environment_variables": {
5 - "VLLM_RPC_TIMEOUT": 100000,
5 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
6 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
7 "VLLM_CPU_SGL_KERNEL": 1,
.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json
-1
@@ -13,7 +13,6 @@
13 200
14 ],
15 "server_environment_variables": {
16 - "VLLM_RPC_TIMEOUT": 100000,
16 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
17 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
18 "VLLM_CPU_SGL_KERNEL": 1,
.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json
-1
@@ -5,7 +5,6 @@
5 ],
6 "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
7 "server_environment_variables": {
8 - "VLLM_RPC_TIMEOUT": 100000,
8 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120
9 },
10 "server_parameters": {
.buildkite/performance-benchmarks/tests/serving-tests-cpu-embed.json
-1
@@ -9,7 +9,6 @@
9 128
10 ],
11 "server_environment_variables": {
12 - "VLLM_RPC_TIMEOUT": 100000,
12 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
13 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
14 "VLLM_CPU_SGL_KERNEL": 1,
.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json
-1
@@ -5,7 +5,6 @@
5 ],
6 "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
7 "server_environment_variables": {
8 - "VLLM_RPC_TIMEOUT": 100000,
8 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
9 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
10 "VLLM_CPU_SGL_KERNEL": 1,
.buildkite/performance-benchmarks/tests/serving-tests-cpu.json
-1
@@ -5,7 +5,6 @@
5 ],
6 "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
7 "server_environment_variables": {
8 - "VLLM_RPC_TIMEOUT": 100000,
8 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
9 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
10 "VLLM_CPU_SGL_KERNEL": 1,
.buildkite/performance-benchmarks/tests/throughput-tests-arm64-cpu.json
-1
@@ -2,7 +2,6 @@
2 {
3 "test_name": "throughput_llama8B_tp1",
4 "environment_variables": {
5 - "VLLM_RPC_TIMEOUT": 100000,
5 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
6 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
7 "VLLM_CPU_KVCACHE_SPACE": 40
.buildkite/performance-benchmarks/tests/throughput-tests-cpu.json
-1
@@ -2,7 +2,6 @@
2 {
3 "test_name": "throughput_llama8B_tp2",
4 "environment_variables": {
5 - "VLLM_RPC_TIMEOUT": 100000,
5 "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
6 "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
7 "VLLM_CPU_SGL_KERNEL": 1,
docs/contributing/profiling.md
+1 -2
@@ -35,8 +35,7 @@ Traces can be visualized using <https://ui.perfetto.dev/>.
35
36 !!! tip
37 To stop the profiler - it flushes out all the profile trace files to the directory. This takes time, for example for about 100 requests worth of data for a llama 70b, it takes about 10 minutes to flush out on a H100.
38 - Set the env variable VLLM_RPC_TIMEOUT to a big number before you start the server. Say something like 30 minutes.
39 - `export VLLM_RPC_TIMEOUT=1800000`
38 + The engine client waits for this flush to complete without timing out, so simply allow the stop call to run to completion.
39
40 ### Example commands and usage
41
vllm/envs.py
-4
@@ -95,7 +95,6 @@ if TYPE_CHECKING:
95 CMAKE_BUILD_TYPE: Literal["Debug", "Release", "RelWithDebInfo"] | None = None
96 VERBOSE: bool = False
97 VLLM_ALLOW_LONG_MAX_MODEL_LEN: bool = False
98 - VLLM_RPC_TIMEOUT: int = 10000 # ms
98 VLLM_HTTP_TIMEOUT_KEEP_ALIVE: int = 5 # seconds
99 VLLM_MAX_N_SEQUENCES: int = 16384
100 VLLM_PLUGINS: list[str] | None = None
@@ -1015,9 +1014,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
1014 "VLLM_TEST_FORCE_LOAD_FORMAT": lambda: os.getenv(
1015 "VLLM_TEST_FORCE_LOAD_FORMAT", "dummy"
1016 ),
1018 - # Time in ms for the zmq client to wait for a response from the backend
1019 - # server for simple data operations
1020 - "VLLM_RPC_TIMEOUT": lambda: int(os.getenv("VLLM_RPC_TIMEOUT", "10000")),
1017 # Timeout in seconds for keeping HTTP connections alive in API server
1018 "VLLM_HTTP_TIMEOUT_KEEP_ALIVE": lambda: int(
1019 os.environ.get("VLLM_HTTP_TIMEOUT_KEEP_ALIVE", "5")