perf(agentx): refresh dsv4-fp4-gb300-dynamo-sglang-agentic harness

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/agg-gb300-tp4-mtp-kvoffload.yaml

@@ -87,6 +87,7 @@ backend:
   sglang_config:
     aggregated:
       served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      enable-metrics: true
       trust-remote-code: true
       stream-interval: 50
       watchdog-timeout: 1000000
@@ -139,16 +140,14 @@ benchmark:
     # in process_agentic_result.py instead. TP must match sglang_config tp-size.
     IS_MULTINODE: "false"
     TP: "4"
-    # Enable dynamo conv-aware routing (per-session affinity so multi-turn
-    # KV cache is reused across turns). Previously opted out because the
-    # frontend 400'd on aiperf's nvext.session_control actions; re-enabled
-    # to test with the current build.
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    # Bind each conversation through the supported Dynamo session header.
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     # Container-side path of the aiperf mmap dataset cache; host-side mount
     # wired via launch_gb300-*.sh srtslurm.yaml default_mounts. Without it,
     # aiperf re-tokenizes + re-writes the dataset mmap on every run.
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     # Persistent HF hub cache (also via default_mounts) so the trace dataset
     # isn't re-downloaded each run; overrides the workflow-level HF_HUB_CACHE.
     HF_HUB_CACHE: "/hf_hub_cache"
-

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/disagg-gb300-12p4d-dep8-dep16-c1024-mtp-kvoffload.yaml

@@ -46,6 +46,9 @@ frontend:
   num_additional_frontends: 4
   env:
     DYN_ROUTER_TEMPERATURE: "10000000"
+    # AgentX warmup can legitimately keep the single wide decode worker busy
+    # for longer than Dynamo's 10-second TCP request-plane default.
+    DYN_TCP_REQUEST_TIMEOUT: "60"
     PIP_BREAK_SYSTEM_PACKAGES: "1"
   args:
     router-mode: "kv"
@@ -129,6 +132,7 @@ backend:
     prefill:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -163,6 +167,7 @@ backend:
     decode:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -206,7 +211,8 @@ benchmark:
     RESULT_DIR: /logs/agentic
     PORT: "8000"
     IS_MULTINODE: "true"
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     HF_HUB_CACHE: "/hf_hub_cache"
-

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/disagg-gb300-2p1d-dep8-tp4-c8-mtp-kvoffload.yaml

@@ -129,6 +129,7 @@ backend:
     prefill:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -163,6 +164,7 @@ backend:
     decode:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -205,7 +207,8 @@ benchmark:
     RESULT_DIR: /logs/agentic
     PORT: "8000"
     IS_MULTINODE: "true"
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     HF_HUB_CACHE: "/hf_hub_cache"
-

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/disagg-gb300-2p4d-dep8-dep16-c256-mtp-kvoffload.yaml

@@ -129,6 +129,7 @@ backend:
     prefill:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -163,6 +164,7 @@ backend:
     decode:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -206,7 +208,8 @@ benchmark:
     RESULT_DIR: /logs/agentic
     PORT: "8000"
     IS_MULTINODE: "true"
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     HF_HUB_CACHE: "/hf_hub_cache"
-

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/disagg-gb300-4p4d-dep8-dep16-c256-mtp-kvoffload.yaml

@@ -129,6 +129,7 @@ backend:
     prefill:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -163,6 +164,7 @@ backend:
     decode:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -206,7 +208,8 @@ benchmark:
     RESULT_DIR: /logs/agentic
     PORT: "8000"
     IS_MULTINODE: "true"
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     HF_HUB_CACHE: "/hf_hub_cache"
-

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic/disagg-gb300-6p4d-dep8-dep16-c512-mtp-kvoffload.yaml

@@ -129,6 +129,7 @@ backend:
     prefill:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -163,6 +164,7 @@ backend:
     decode:
       host: 0.0.0.0
       served-model-name: deepseek-ai/DeepSeek-V4-Pro
+      enable-metrics: true
       model-path: /model/
       trust-remote-code: true
       watchdog-timeout: 86400
@@ -206,7 +208,8 @@ benchmark:
     RESULT_DIR: /logs/agentic
     PORT: "8000"
     IS_MULTINODE: "true"
-    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: "true"
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
+    AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "sglang:"
     AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
     HF_HUB_CACHE: "/hf_hub_cache"
-

perf-changelog.yaml

@@ -5697,3 +5697,15 @@
     - "Use the official lmsysorg/sglang:v0.5.16-rocm720-mi30x image, enable SGLang prompt-cache reporting, and pass the backend Prometheus metrics endpoint explicitly to AIPerf"
     - "Allow up to 30 minutes for healthy AgentX responses admitted near the end of the measurement window to drain before AIPerf finalizes profile metric coverage"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2528
+
+- config-keys:
+    - dsv4-fp4-gb300-dynamo-sglang-agentic-agg
+    - dsv4-fp4-gb300-dynamo-sglang-agentic-disagg
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Refresh submission with up to date AgentX harness."
+    - "Collect every SGLang worker's metrics through AgentX and fail when required sglang: series are absent."
+    - "Enable SGLang metrics on every aggregate, prefill, and decode engine."
+    - "Use supported header-based Dynamo session routing with the in-repo AIPerf build."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2520

runners/launch_gb300-nv.sh

@@ -206,13 +206,15 @@ if [[ "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX ==
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5" \
         recipes/sglang/qwen3.5
 elif [[ "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "dsv4" ]]; then
-    # DSv4 GB300 sglang agentic: NVIDIA/srt-slurm v1.0.10 has the nginx
-    # client_max_body_size fix (>1 MiB agentic warmup bodies), the
-    # session-affinity frontend, and the BenchmarkType.CUSTOM / extra_mount
-    # schema these recipes need.
+    # DSv4 GB300 SGLang agentic uses NVIDIA/srt-slurm v1.0.38. In addition to
+    # the nginx body-size fix, session-affinity frontend, and custom benchmark
+    # schema required by these recipes, this release injects every logical
+    # SGLang worker leader's /metrics URL into AIPERF_SERVER_METRICS_URLS.
+    # AgentX forwards that list to aiperf's --server-metrics argument so its
+    # trace artifacts include backend metrics for every engine.
     git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR"
-    git checkout v1.0.10
+    git checkout v1.0.38
     mkdir -p recipes/sglang/deepseek-v4/agentic
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/agentic" \
         recipes/sglang/deepseek-v4/agentic