feat(perf): migrate python perf image config from test-plans #808

.github/actions/detect-changed-impls/action.yml

@@ -84,15 +84,23 @@ runs:
           git show "origin/$BASE_REF:$IMPLS_FILE" > /tmp/impls-old.yaml 2>/dev/null || echo "" > /tmp/impls-old.yaml
           cp "$IMPLS_FILE" /tmp/impls-new.yaml
 
+          # Normalize objects before comparing so YAML comments/formatting
+          # don't make unchanged entries look modified.
+          normalize_object() {
+            local file="$1"
+            local query="$2"
+            yq eval -o=json -I=0 "$query" "$file" 2>/dev/null || echo ""
+          }
+
           # Find changed/new implementations
           OLD_IMPLS=$(yq eval '.implementations[].id' /tmp/impls-old.yaml 2>/dev/null | sort || echo "")
           NEW_IMPLS=$(yq eval '.implementations[].id' /tmp/impls-new.yaml 2>/dev/null | sort || echo "")
 
           for impl_id in $NEW_IMPLS; do
             if echo "$OLD_IMPLS" | grep -q "^${impl_id}$"; then
               # Exists in old, check if content changed
-              OLD_IMPL=$(yq eval ".implementations[] | select(.id == \"$impl_id\")" /tmp/impls-old.yaml 2>/dev/null || echo "")
-              NEW_IMPL=$(yq eval ".implementations[] | select(.id == \"$impl_id\")" /tmp/impls-new.yaml 2>/dev/null || echo "")
+              OLD_IMPL=$(normalize_object /tmp/impls-old.yaml ".implementations[] | select(.id == \"$impl_id\")")
+              NEW_IMPL=$(normalize_object /tmp/impls-new.yaml ".implementations[] | select(.id == \"$impl_id\")")
 
               if [ "$OLD_IMPL" != "$NEW_IMPL" ]; then
                 CHANGED_IMPLS="${CHANGED_IMPLS:+$CHANGED_IMPLS|}$impl_id"
@@ -113,8 +121,8 @@ runs:
 
             for relay_id in $NEW_RELAYS; do
               if echo "$OLD_RELAYS" | grep -q "^${relay_id}$"; then
-                OLD_RELAY=$(yq eval ".relays[] | select(.id == \"$relay_id\")" /tmp/impls-old.yaml 2>/dev/null || echo "")
-                NEW_RELAY=$(yq eval ".relays[] | select(.id == \"$relay_id\")" /tmp/impls-new.yaml 2>/dev/null || echo "")
+                OLD_RELAY=$(normalize_object /tmp/impls-old.yaml ".relays[] | select(.id == \"$relay_id\")")
+                NEW_RELAY=$(normalize_object /tmp/impls-new.yaml ".relays[] | select(.id == \"$relay_id\")")
                 if [ "$OLD_RELAY" != "$NEW_RELAY" ]; then
                   CHANGED_RELAYS="${CHANGED_RELAYS:+$CHANGED_RELAYS|}$relay_id"
                   echo "  → Changed relay: $relay_id"
@@ -131,8 +139,8 @@ runs:
 
             for router_id in $NEW_ROUTERS; do
               if echo "$OLD_ROUTERS" | grep -q "^${router_id}$"; then
-                OLD_ROUTER=$(yq eval ".routers[] | select(.id == \"$router_id\")" /tmp/impls-old.yaml 2>/dev/null || echo "")
-                NEW_ROUTER=$(yq eval ".routers[] | select(.id == \"$router_id\")" /tmp/impls-new.yaml 2>/dev/null || echo "")
+                OLD_ROUTER=$(normalize_object /tmp/impls-old.yaml ".routers[] | select(.id == \"$router_id\")")
+                NEW_ROUTER=$(normalize_object /tmp/impls-new.yaml ".routers[] | select(.id == \"$router_id\")")
                 if [ "$OLD_ROUTER" != "$NEW_ROUTER" ]; then
                   CHANGED_ROUTERS="${CHANGED_ROUTERS:+$CHANGED_ROUTERS|}$router_id"
                   echo "  → Changed router: $router_id"
@@ -153,8 +161,8 @@ runs:
             for baseline_id in $NEW_BASELINES; do
               if echo "$OLD_BASELINES" | grep -q "^${baseline_id}$"; then
                 # Exists in old, check if content changed
-                OLD_BASELINE=$(yq eval ".baselines[] | select(.id == \"$baseline_id\")" /tmp/impls-old.yaml 2>/dev/null || echo "")
-                NEW_BASELINE=$(yq eval ".baselines[] | select(.id == \"$baseline_id\")" /tmp/impls-new.yaml 2>/dev/null || echo "")
+                OLD_BASELINE=$(normalize_object /tmp/impls-old.yaml ".baselines[] | select(.id == \"$baseline_id\")")
+                NEW_BASELINE=$(normalize_object /tmp/impls-new.yaml ".baselines[] | select(.id == \"$baseline_id\")")
                 if [ "$OLD_BASELINE" != "$NEW_BASELINE" ]; then
                   CHANGED_BASELINES="${CHANGED_BASELINES:+$CHANGED_BASELINES|}$baseline_id"
                   echo "  → Changed baseline: $baseline_id"

.github/actions/run-bash-perf-test/action.yml

@@ -62,6 +62,10 @@ inputs:
     description: "Enable debug mode (sets DEBUG=true in test containers)"
     required: false
     default: "false"
+  timeout:
+    description: "Max seconds per test before harness abort"
+    required: false
+    default: ""
 
 runs:
   using: "composite"
@@ -129,6 +133,16 @@ runs:
           LATENCY_ITERATIONS_FLAG="--latency-iterations ${{ inputs.latency-iterations }}"
         fi
 
+        TIMEOUT_FLAG=""
+        if [ -n "${{ inputs.timeout }}" ]; then
+          TIMEOUT_FLAG="--timeout ${{ inputs.timeout }}"
+        fi
+
+        if echo '${{ inputs.test-select }}' | grep -q 'python'; then
+          export PY_YAMUX_DISABLE_HYSTERESIS=1
+          export PY_YAMUX_ASSUME_RTT_MS=1
+        fi
+
         ./run.sh \
           --impl-select '${{ inputs.test-select }}' \
           --test-ignore '${{ inputs.test-ignore }}' \
@@ -140,6 +154,7 @@ runs:
           $DOWNLOAD_BYTES_FLAG \
           $DURATION_FLAG \
           $LATENCY_ITERATIONS_FLAG \
+          $TIMEOUT_FLAG \
           $SNAPSHOT_FLAG \
           $EXPORT_DOCKER_FLAG \
           $FORCE_MATRIX_FLAG \

.github/workflows/perf-pr.yml

@@ -59,6 +59,11 @@ on:
         required: false
         default: false
         type: boolean
+      timeout:
+        description: 'Max seconds per test before harness abort'
+        required: false
+        default: 300
+        type: number
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -78,6 +83,7 @@ jobs:
       debug: ${{ steps.resolve.outputs.debug }}
       snapshot: ${{ steps.resolve.outputs.snapshot }}
       export-docker-images: ${{ steps.resolve.outputs.export-docker-images }}
+      timeout: ${{ steps.resolve.outputs.timeout }}
       should-run-tests: ${{ steps.resolve.outputs.should-run-tests }}
     steps:
       - name: Checkout code
@@ -108,6 +114,7 @@ jobs:
             echo "debug=${{ github.event.inputs.debug }}" >> $GITHUB_OUTPUT
             echo "snapshot=${{ github.event.inputs.snapshot }}" >> $GITHUB_OUTPUT
             echo "export-docker-images=${{ github.event.inputs.export-docker-images }}" >> $GITHUB_OUTPUT
+            echo "timeout=${{ github.event.inputs.timeout }}" >> $GITHUB_OUTPUT
             echo "should-run-tests=true" >> $GITHUB_OUTPUT
             echo "→ Manual trigger: using workflow inputs"
           else
@@ -120,14 +127,25 @@ jobs:
             echo "test-ignore=~failing" >> $GITHUB_OUTPUT
 
             CHANGED_BASELINES="${{ steps.detect.outputs.changed-baselines }}"
+            CHANGED_IMPLS="${{ steps.detect.outputs.changed-impls }}"
             if [ -z "$CHANGED_BASELINES" ]; then
               echo "baseline-select=~default_baselines" >> $GITHUB_OUTPUT
             else
               echo "baseline-select=$CHANGED_BASELINES" >> $GITHUB_OUTPUT
             fi
 
-            # Same rule for baselines: avoid negating selected values.
-            echo "baseline-ignore=~failing" >> $GITHUB_OUTPUT
+            # Skip baseline builds/tests on impl-only PRs to save runner disk/time.
+            if [ -z "$CHANGED_BASELINES" ] && [ -n "$CHANGED_IMPLS" ]; then
+              echo "baseline-ignore=~baselines" >> $GITHUB_OUTPUT
+            else
+              echo "baseline-ignore=~failing" >> $GITHUB_OUTPUT
+            fi
+
+            if echo "$CHANGED_IMPLS" | grep -q 'python'; then
+              echo "timeout=900" >> $GITHUB_OUTPUT
+            else
+              echo "timeout=300" >> $GITHUB_OUTPUT
+            fi
 
             echo "iterations=10" >> $GITHUB_OUTPUT
             echo "force-matrix-rebuild=${{ steps.detect.outputs.impls-yaml-changed }}" >> $GITHUB_OUTPUT
@@ -167,3 +185,4 @@ jobs:
           force-matrix-rebuild: ${{ needs.resolve-parameters.outputs.force-matrix-rebuild }}
           force-image-rebuild: ${{ needs.resolve-parameters.outputs.force-image-rebuild }}
           debug: ${{ needs.resolve-parameters.outputs.debug }}
+          timeout: '${{ needs.resolve-parameters.outputs.timeout }}'

lib/lib-inputs-yaml.sh

@@ -88,6 +88,7 @@ EOF
   DOWNLOAD_BYTES: "${DOWNLOAD_BYTES:-}"
   DURATION: "${DURATION:-}"
   LATENCY_ITERATIONS: "${LATENCY_ITERATIONS:-}"
+  PERF_TEST_TIMEOUT_SECS: "${PERF_TEST_TIMEOUT_SECS:-}"
 EOF
             ;;
         hole-punch)

perf/README.md

@@ -73,8 +73,8 @@ Control test behavior with these options:
 # Number of iterations for upload/download tests (default: 10)
 ./run.sh --iterations 10
 
-# Duration per iteration for throughput tests in seconds (default: 20)
-./run.sh --duration 20
+# Max seconds per test before the harness aborts (default: 300)
+./run.sh --timeout 900
 
 # Number of iterations for latency tests (default: 100)
 ./run.sh --latency-iterations 100
@@ -83,7 +83,11 @@ Control test behavior with these options:
 ./run.sh --cache-dir /srv/cache
 ```
 
-**Note**: The `--iterations` flag controls both upload and download iterations. The framework measures throughput by transferring data over a time period and measuring bytes/second.
+**Note**: The `--iterations` flag controls both upload and download iterations. Modern dialers measure throughput by transferring a fixed amount of data (`UPLOAD_BYTES` / `DOWNLOAD_BYTES`) per iteration and reporting Gbps.
+
+**Note**: `--timeout` sets `PERF_TEST_TIMEOUT_SECS` (harness limit for the whole docker-compose test). This is separate from `TEST_TIMEOUT_SECS` inside dialer/listener containers (peer/redis wait limits).
+
+**Legacy**: `--duration` (default 20s) is passed to dialers as `DURATION` but is unused by current implementations; prefer `--upload-bytes`, `--iterations`, and `--timeout`.
 
 ## Test Filtering
 
@@ -586,8 +590,10 @@ For now, all tests run locally using Docker networking. Multi-machine testing su
 
 **Tests failing with timeout**
 - Check container logs: `$TEST_PASS_DIR/logs/<test-name>.log`
-- Increase test duration: `--duration 30`
-- Enable debug mode: `--debug`
+- Each perf test is capped at **300 seconds** by default (`--timeout`). For slow stacks: `./run.sh --timeout 900 --test-select "..." --yes`
+- Do not use `--debug` for throughput runs (adds log I/O and often lowers Gbps)
+- `--duration` does not extend the harness limit and is unused by modern dialers (they transfer fixed `UPLOAD_BYTES`/`DOWNLOAD_BYTES` per iteration, not “run for N seconds”)
+- Python yamux tuning uses `PY_YAMUX_*` env vars (e.g. `PY_YAMUX_DISABLE_HYSTERESIS=1`); they are injected into **python-v0.x** containers only, not Go/Rust
 
 **Cache not working / Matrix regenerates every time**
 - Check TEST_RUN_KEY in output (should be consistent for same configuration)

perf/images.yaml

@@ -9,7 +9,9 @@ test-aliases:
   - alias: "none"
     value: "!~all"
   - alias: "images"
-    value: "dotnet-v1.0|go-v0.45|js-v3.x|rust-v0.56|lua-v0.1.0|eth-p2p-z-v0.0.2|nim-v1.15"
+    value: "dotnet-v1.0|go-v0.45|js-v3.x|python-v0.x|rust-v0.56|lua-v0.1.0|eth-p2p-z-v0.0.2|nim-v1.15"
+  - alias: "python"
+    value: "python-v0.x"
   - alias: "baselines"
     value: "https|quic-go|iperf|quinn|tquic|tokio-quiche|tokio-tquic|quiche"
   - alias: "default_baselines"
@@ -28,7 +30,7 @@ test-aliases:
     value: "eth-p2p-z-v0.0.2"
   - alias: "nim"
     value: "nim-v1.15"
-  - alias: "failing" # As of 26 Feb 2026
+  - alias: "failing" # As of 7 Jul 2026
     value: "go-v0.45 x js-v3.x (webtransport)|js-v3.x x go-v0.45 (webtransport)|js-v3.x x js-v3.x (webtransport)|js-v3.x x dotnet-v1.0 (tcp, tls, yamux)|rust-v0.56 x dotnet-v1.0 (tcp, tls, yamux)|dotnet-v1.0 x go-v0.45 (tcp, noise, yamux)|dotnet-v1.0 x go-v0.45 (tcp, tls, yamux)|dotnet-v1.0 x go-v0.45 (quic-v1)|dotnet-v1.0 x rust-v0.56 (tcp, noise, yamux)|dotnet-v1.0 x rust-v0.56 (tcp, tls, yamux)|dotnet-v1.0 x rust-v0.56 (quic-v1)|dotnet-v1.0 x js-v3.x (tcp, noise, yamux)|dotnet-v1.0 x js-v3.x (tcp, tls, yamux)|dotnet-v1.0 x dotnet-v1.0 (quic-v1)"
 
 # Baseline implementations - Separate from main tests
@@ -156,6 +158,26 @@ implementations:
     secureChannels: [noise, tls]
     muxers: [yamux]
 
+  - id: python-v0.x
+    source:
+      type: github
+      repo: libp2p/py-libp2p
+      commit: 34d7bcda5e616d05af88f4a787d9193564ecc018
+      dockerfile: interop/perf/Dockerfile
+    transports: [tcp, ws]
+    secureChannels: [noise, tls]
+    muxers: [yamux, mplex]
+
+  # Local python config example (same Dockerfile; context = repo root = images/python/0.x/py-libp2p)
+#  - id: python-v0.x
+#    source:
+#      type: local
+#      path: images/python/0.x/py-libp2p
+#      dockerfile: interop/perf/Dockerfile
+#    transports: [tcp, ws]
+#    secureChannels: [noise, tls]
+#    muxers: [yamux, mplex]
+
   - id: lua-v0.1.0
     source:
       type: github