test(ci): mitigate e2e flakes with targeted baseload shrink

Author: mikehardy

.github/workflows/scripts/flake-summary.sh

@@ -25,9 +25,7 @@ DETOX_LOG="${RNFB_DETOX_LOG:-}"
   }
 
   summarize "detox-log" "${DETOX_LOG}"
-  summarize "simulator-log" "simulator.log"
-  summarize "testing-log" "testing.log"
-  summarize "springboard-log" "springboard-invertase.log"
+  summarize "sim-app-log" "sim-app.log"
   summarize "resource-monitor" "resource-monitor.log"
   summarize "resource-monitor-android" "resource-monitor-android.log"
   summarize "metro-log" "metro.log"
@@ -43,7 +41,7 @@ DETOX_LOG="${RNFB_DETOX_LOG:-}"
   echo ""
 
   echo "=== log stream death markers ==="
-  for f in simulator.log testing.log springboard-invertase.log; do
+  for f in sim-app.log; do
     if [[ -f "$f" ]]; then
       rg -n 'Child process terminated|log stream restarted' "$f" 2>/dev/null | tail -20 || true
     fi

.github/workflows/scripts/simulator-logging.sh

@@ -1,22 +1,22 @@
 #!/bin/bash
-# Start or restart iOS simulator log streams and resource monitor (sourced by boot-simulator.sh).
+# Start or restart iOS simulator focused log stream and resource monitor (sourced by boot-simulator.sh).
 set -uo pipefail
 
 restart_simulator_logging() {
   local with_stdout="${RNFB_SIM_LOG_STDOUT:-0}"
+  local record_screens="${RNFB_RECORD_SCREENS:-0}"
   local log_dir="${RNFB_SIM_LOG_DIR:-.}"
-  local simulator_log="${log_dir}/simulator.log"
-  local testing_log="${log_dir}/testing.log"
-  local springboard_log="${log_dir}/springboard-invertase.log"
+  local sim_app_log="${log_dir}/sim-app.log"
   local resource_log="${log_dir}/resource-monitor.log"
   local repo_root="${RNFB_REPO_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/../../.." && pwd)}"
+  local log_predicate='process == "testing" OR (process == "SpringBoard" AND eventMessage CONTAINS "invertase")'
 
   if ! xcrun simctl list devices booted 2>/dev/null | grep -q Booted; then
     echo "[boot-status] phase=log_streams skipped=no_booted_simulator"
     return 1
   fi
 
-  echo "[boot-status] phase=log_streams_restart ts=$(date -u +%Y-%m-%dT%H:%M:%SZ) stdout=${with_stdout} dir=${log_dir}"
+  echo "[boot-status] phase=log_streams_restart ts=$(date -u +%Y-%m-%dT%H:%M:%SZ) stdout=${with_stdout} record_screens=${record_screens} dir=${log_dir}"
 
   pkill -f 'simctl spawn booted log stream' 2>/dev/null || true
   pkill -f 'simctl io booted recordVideo' 2>/dev/null || true
@@ -25,33 +25,26 @@ restart_simulator_logging() {
 
   {
     echo ""
-    echo "=== log stream restarted $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
-  } >>"$simulator_log" 2>/dev/null || : >"$simulator_log"
-  {
-    echo ""
-    echo "=== log stream restarted $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
-  } >>"$testing_log" 2>/dev/null || : >"$testing_log"
-  {
-    echo ""
-    echo "=== log stream restarted $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
-  } >>"$springboard_log" 2>/dev/null || : >"$springboard_log"
+    echo "=== log stream restarted $(date -u +%Y-%m-%dT%H:%M:%SZ) predicate=${log_predicate} ==="
+  } >>"$sim_app_log" 2>/dev/null || : >"$sim_app_log"
+
+  if [[ "$record_screens" == "1" ]]; then
+    if [[ "$with_stdout" == "1" ]]; then
+      xcrun simctl io booted recordVideo --codec=h264 -f "${log_dir}/simulator.mp4" 2>&1 &
+    else
+      nohup sh -c "xcrun simctl io booted recordVideo --codec=h264 -f '${log_dir}/simulator.mp4' 2>&1 &" >/dev/null 2>&1 &
+    fi
+  fi
 
   if [[ "$with_stdout" == "1" ]]; then
-    xcrun simctl io booted recordVideo --codec=h264 -f "${log_dir}/simulator.mp4" 2>&1 &
-    xcrun simctl spawn booted log stream --level debug --style compact 2>&1 | tee -a "$simulator_log" &
-    xcrun simctl spawn booted log stream --level debug --style compact \
-      --predicate 'process == "testing"' 2>&1 | tee -a "$testing_log" &
     xcrun simctl spawn booted log stream --level debug --style compact \
-      --predicate 'process == "SpringBoard" AND eventMessage CONTAINS "invertase"' 2>&1 | tee -a "$springboard_log" &
+      --predicate "$log_predicate" 2>&1 | tee -a "$sim_app_log" &
   else
-    nohup sh -c "xcrun simctl io booted recordVideo --codec=h264 -f '${log_dir}/simulator.mp4' 2>&1 &" >/dev/null 2>&1 &
-    nohup sh -c "xcrun simctl spawn booted log stream --level debug --style compact >>'${simulator_log}' 2>&1 &" >/dev/null 2>&1 &
-    nohup sh -c "xcrun simctl spawn booted log stream --level debug --style compact --predicate 'process == \"testing\"' >>'${testing_log}' 2>&1 &" >/dev/null 2>&1 &
-    nohup sh -c "xcrun simctl spawn booted log stream --level debug --style compact --predicate 'process == \"SpringBoard\" AND eventMessage CONTAINS \"invertase\"' >>'${springboard_log}' 2>&1 &" >/dev/null 2>&1 &
+    nohup sh -c "xcrun simctl spawn booted log stream --level debug --style compact --predicate '${log_predicate}' >>'${sim_app_log}' 2>&1 &" >/dev/null 2>&1 &
   fi
 
   chmod +x "${repo_root}/.github/workflows/scripts/resource-monitor.sh"
   RNFB_RESOURCE_MONITOR_LOG="$resource_log" nohup "${repo_root}/.github/workflows/scripts/resource-monitor.sh" >/dev/null 2>&1 &
 
-  echo "[boot-status] phase=log_streams_started video=${log_dir}/simulator.mp4 monitor=${resource_log}"
+  echo "[boot-status] phase=log_streams_started sim_app_log=${sim_app_log} video=$([[ "$record_screens" == "1" ]] && echo "${log_dir}/simulator.mp4" || echo disabled) monitor=${resource_log}"
 }

.github/workflows/scripts/wait-for-load-settle.sh

@@ -4,7 +4,7 @@ set -euo pipefail
 
 MAX_WAIT_SEC="${RNFB_LOAD_SETTLE_MAX_WAIT_SEC:-1200}"
 POLL_SEC="${RNFB_LOAD_SETTLE_POLL_SEC:-5}"
-MAX_LOAD="${RNFB_LOAD_SETTLE_MAX_LOAD:-10}"
+MAX_LOAD="${RNFB_LOAD_SETTLE_MAX_LOAD:-20}"
 
 parse_load1() {
   # uptime: ... load averages: 1.23 4.56 7.89

.github/workflows/tests_e2e_ios.yml

@@ -1,13 +1,23 @@
 name: Testing E2E iOS
 
 on:
+  workflow_call:
+    inputs:
+      record_screens:
+        description: 'Record host screencapture and simulator video artifacts'
+        type: boolean
+        default: false
   workflow_dispatch:
     inputs:
       iterations:
         description: 'Number of iterations to run. Default 1. Max 122.'
         required: true
         default: 1
         type: number
+      record_screens:
+        description: 'Record host screencapture and simulator video artifacts'
+        type: boolean
+        default: false
   pull_request:
     branches:
       - '**'
@@ -97,6 +107,7 @@ jobs:
     timeout-minutes: 107
     env:
       CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+      RNFB_RECORD_SCREENS: ${{ (github.event_name == 'workflow_dispatch' || github.event_name == 'workflow_call') && inputs.record_screens && '1' || '0' }}
       CCACHE_SLOPPINESS: clang_index_store,file_stat_matches,include_file_ctime,include_file_mtime,ivfsoverlay,pch_defines,modules,system_headers,time_macros
       CCACHE_FILECLONE: true
       CCACHE_DEPEND: true
@@ -112,14 +123,15 @@ jobs:
         # https://github.com/guidepup/setup-action/releases
         uses: guidepup/setup-action@5ab89fbb6641406f6f6c91057225f3fb397c2eea # v0.20.0
         continue-on-error: true
+        if: env.RNFB_RECORD_SCREENS == '1'
         with:
           record: true
 
       - name: Upload Screen Recording Environment Setup
         # https://github.com/actions/upload-artifact/releases
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         continue-on-error: true
-        if: always()
+        if: always() && env.RNFB_RECORD_SCREENS == '1'
         with:
           name: screenrecording-setup-${{ matrix.buildmode }}-${{ matrix.iteration }}.mov
           path: ./recordings/
@@ -327,11 +339,11 @@ jobs:
           curl --output /dev/null --silent --head --fail "http://localhost:8081/index.bundle?platform=ios&dev=true&minify=false&inlineSourceMap=true"
           echo "...javascript bundle ready"
 
-      - name: Start Screen and System Logging
+      - name: Start host screen recording
+        if: env.RNFB_RECORD_SCREENS == '1'
         continue-on-error: true
         run: |
           nohup sh -c "sleep 314159265 | screencapture -v -C -k -T0 -g screenrecording.mov > screenrecording.log 2>&1 &"
-          nohup sh -c "log stream --backtrace --color none --style syslog > syslog.log 2>&1 &"
 
       - name: Pre-Boot Simulator
         # Separate Simulator boot from Detox run. boot-simulator.sh polls bootstatus and logs
@@ -344,12 +356,12 @@ jobs:
           max_attempts: 3
           command: RNFB_START_SIM_LOGS=0 ./.github/workflows/scripts/boot-simulator.sh
 
-      - name: Start Simulator Recordings and Log
-        # Log streams are owned by boot-simulator.sh (logs-only mode after pre-boot).
+      - name: Start Simulator App Log
+        # One filtered sim log stream (testing + SpringBoard/invertase); optional sim video when record_screens.
         continue-on-error: true
         run: |
           chmod +x ./.github/workflows/scripts/boot-simulator.sh
-          RNFB_SIM_BOOT_MODE=logs ./.github/workflows/scripts/boot-simulator.sh
+          RNFB_RECORD_SCREENS="${RNFB_RECORD_SCREENS}" RNFB_SIM_BOOT_MODE=logs ./.github/workflows/scripts/boot-simulator.sh
 
       - name: Wait for host load to settle
         continue-on-error: true
@@ -388,43 +400,27 @@ jobs:
         continue-on-error: true
         run: |
           killall -int simctl
-          killall -int screencapture
-          killall -int log
+          if [[ "${RNFB_RECORD_SCREENS}" == "1" ]]; then
+            killall -int screencapture
+          fi
           pkill -f resource-monitor.sh || true
 
       - name: Upload App Video
         # https://github.com/actions/upload-artifact/releases
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         continue-on-error: true
-        if: always()
+        if: always() && env.RNFB_RECORD_SCREENS == '1'
         with:
           name: simulator-${{ matrix.buildmode }}-${{ matrix.iteration }}_video
           path: simulator.mp4
 
-      - name: Upload Simulator Log
-        # https://github.com/actions/upload-artifact/releases
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        continue-on-error: true
-        if: always()
-        with:
-          name: simulator-${{ matrix.buildmode }}-${{ matrix.iteration }}_log
-          path: simulator.log
-
-      - name: Upload testing process log
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        continue-on-error: true
-        if: always()
-        with:
-          name: testing-${{ matrix.buildmode }}-${{ matrix.iteration }}_log
-          path: testing.log
-
-      - name: Upload SpringBoard invertase log
+      - name: Upload Simulator App Log
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         continue-on-error: true
         if: always()
         with:
-          name: springboard-${{ matrix.buildmode }}-${{ matrix.iteration }}_log
-          path: springboard-invertase.log
+          name: sim-app-${{ matrix.buildmode }}-${{ matrix.iteration }}_log
+          path: sim-app.log
 
       - name: Upload resource monitor log
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
@@ -463,7 +459,7 @@ jobs:
         # https://github.com/actions/upload-artifact/releases
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         continue-on-error: true
-        if: always()
+        if: always() && env.RNFB_RECORD_SCREENS == '1'
         with:
           name: screenrecording-${{ matrix.buildmode }}-${{ matrix.iteration }}
           path: screenrecording.*

okf-bundle/ci-workflows/index.md

@@ -4,7 +4,7 @@ GitHub Actions job shape, platform reliability, and artifact triage.
 
 ## Platforms
 
-* [iOS](ios.md) — simulator boot, logging, troubleshooting
+* [iOS](ios.md) — simulator boot, logging, troubleshooting, [CI baseload policy](ios.md#ci-baseload-policy-instrumentation)
 * [Android](android.md) — idling, adb teardown, native coverage
 * [Other](other.md) — macOS Jet/Metro, Windows/shared
 
@@ -13,7 +13,7 @@ GitHub Actions job shape, platform reliability, and artifact triage.
 * [Jet host orchestration](../testing/running-e2e.md#jet-host-orchestration-ports-and-launch-gate) — ports 8090/8091, defer-run launch gate (canonical)
 * [Detox patches](detox-patches.md) — inventory, `ECOMPROMISED`, patch workflow
 * [Cloud API quota triage](../testing/firebase-testing-project.md#ci-triage-cloud-api-quota-pressure) — live FIS/RC pressure (all platforms)
-* [iOS diagnostics](ios.md#operational-notes) — `resource-monitor.sh`, `flake-summary.sh`
+* [iOS diagnostics](ios.md#operational-notes) — `resource-monitor.sh`, `flake-summary.sh`, [CI baseload / load settle](ios.md#ci-baseload-policy-instrumentation)
 
 ## Related