fix(ci): tester/builder start race conditions (#10893)

- lock around builder install
- reuse code in tester/builder

.github/ensure-builder/action.yml

@@ -31,7 +31,7 @@ runs:
         if [[ $TYPE == builder-x86 ]]; then
           # 128-core x86 instance types, aws chooses for us based on capacity
           echo "instance_type=m6a.32xlarge m6i.32xlarge m6in.32xlarge m7a.32xlarge r6a.32xlarge r6i.32xlarge r6in.32xlarge" >> $GITHUB_OUTPUT
-          echo "ami_id=ami-052a1e16394277fdf" >> $GITHUB_OUTPUT
+          echo "ami_id=ami-044f8e99cd65daf76" >> $GITHUB_OUTPUT
           echo "runner_concurrency=20" >> $GITHUB_OUTPUT
           echo "runner_label=$USERNAME-x86" >> $GITHUB_OUTPUT
           echo "ttl=40" >> $GITHUB_OUTPUT

.github/ensure-builder/chron

@@ -0,0 +1,11 @@
+set -eu
+# One-time config only on builder.
+if ! [ -f ~/maybe_exit_spot.sh ] ; then
+  cp scripts/ci/maybe_exit_spot.sh ~/maybe_exit_spot.sh
+  # Run maybe_exit_spot.sh every minute
+  chmod +x ~/maybe_exit_spot.sh
+  echo "* * * * * ~/maybe_exit_spot.sh" | crontab -
+  echo "Configured instance exit cron job."
+else
+  echo "Chron jobs already configured."
+fi

.github/ensure-builder/wrapper → .github/ensure-builder/install

@@ -32,12 +32,10 @@ if ! command -v jq >/dev/null; then
   sudo mv ./jq /usr/bin/jq
 fi
 function install_parallel() {
-  sudo apt update && sudo apt install parallel
+  sudo apt update && sudo systemctl stop unattended-upgrades && sudo apt install parallel
 }
 export -f install_parallel
 if ! command -v parallel >/dev/null; then
-  DENOISE=1 ci3/denoise install_parallel
+  DENOISE=1 ci3/denoise ci3/retry install_parallel
 fi
-[ -x /usr/local/bin/earthly ] || ci3/dump_fail ci3/earthly_install
-# Run the test.
-ci3/aws_handle_evict "set -eu; $2"
+[ -x /usr/local/bin/earthly ] || ci3/dump_fail ci3/earthly_install

.github/ensure-builder/run

@@ -1,26 +1,34 @@
 #!/usr/bin/env bash
+set -u
 exit_code=254
 ttl=$1
 scripts/run_on_builder "
   set -eu;
   sudo shutdown -P $ttl;
-  if ! [ -d ~/run-$RUN_ID ]; then
-    mkdir -p ~/run-$RUN_ID;
-    cd ~/run-$RUN_ID;
-    git init >/dev/null 2>&1;
-    git remote add origin https://github.com/aztecprotocol/aztec-packages >/dev/null 2>&1;
-    git fetch --depth 1 origin $GIT_COMMIT >/dev/null 2>&1;
-    git checkout FETCH_HEAD >/dev/null 2>&1;
-  fi;
-  cd ~/run-$RUN_ID;
-  .github/ensure-builder/wrapper $DOCKERHUB_PASSWORD '$INPUT';
+  function clone {
+    if ! [ -d ~/run-$RUN_ID ]; then
+      mkdir -p ~/run-$RUN_ID;
+      cd ~/run-$RUN_ID;
+      git init >/dev/null 2>&1;
+      git remote add origin https://github.com/aztecprotocol/aztec-packages >/dev/null 2>&1;
+      git fetch --depth 1 origin $GIT_COMMIT >/dev/null 2>&1;
+      git checkout FETCH_HEAD >/dev/null 2>&1;
+    fi;
+  }
+  export RUN_ID GIT_COMMIT
+  export -f clone
+  flock /var/lock/clone.lock bash -c clone
+  cd ~/run-$RUN_ID
+  flock /var/lock/install.lock .github/ensure-builder/chron
+  flock /var/lock/install.lock .github/ensure-builder/install $DOCKERHUB_PASSWORD
+  ci3/aws_handle_evict 'set -eu; $INPUT'
 "
 exit_code=$?
 if [ $exit_code = 255 ]; then
-  echo "Treating ssh termination as spot eviction.";
+  echo "Treating ssh termination as spot eviction."
   exit_code=155
-fi;
-echo "exit_code=$exit_code" >> $GITHUB_OUTPUT;
+fi
+echo "exit_code=$exit_code" >> $GITHUB_OUTPUT
 if [ $exit_code = 155 ]; then
   echo "Spot eviction detected - retrying with on-demand."
 fi

.github/ensure-tester/run

@@ -1,19 +1,27 @@
 #!/usr/bin/env bash
+set -u
 exit_code=254
 ttl=$1
 scripts/run_on_tester "
   set -eu;
   sudo shutdown -P $ttl;
-  if ! [ -d ~/run-$RUN_ID ]; then
-    mkdir -p ~/run-$RUN_ID;
-    cd ~/run-$RUN_ID;
-    git init >/dev/null 2>&1;
-    git remote add origin https://github.com/aztecprotocol/aztec-packages >/dev/null 2>&1;
-    git fetch --depth 1 origin $GIT_COMMIT >/dev/null 2>&1;
-    git checkout FETCH_HEAD >/dev/null 2>&1;
-  fi;
-  cd ~/run-$RUN_ID;
-  .github/ensure-tester/wrapper $DOCKERHUB_PASSWORD '$INPUT';
+  function clone {
+    if ! [ -d ~/run-$RUN_ID ]; then
+      mkdir -p ~/run-$RUN_ID;
+      cd ~/run-$RUN_ID;
+      git init >/dev/null 2>&1;
+      git remote add origin https://github.com/aztecprotocol/aztec-packages >/dev/null 2>&1;
+      git fetch --depth 1 origin $GIT_COMMIT >/dev/null 2>&1;
+      git checkout FETCH_HEAD >/dev/null 2>&1;
+    fi;
+  }
+  export RUN_ID GIT_COMMIT
+  export -f clone
+  flock /var/lock/clone.lock bash -c clone
+  cd ~/run-$RUN_ID
+  # reuse script from ensure-builder, but don't set up chron
+  flock /var/lock/install.lock .github/ensure-builder/install $DOCKERHUB_PASSWORD
+  ci3/aws_handle_evict 'set -eu; $INPUT'
 "
 exit_code=$?
 if [ $exit_code = 255 ]; then
@@ -23,4 +31,4 @@ fi
 echo "exit_code=$exit_code" >> $GITHUB_OUTPUT
 if [ $exit_code = 155 ]; then
   echo "Spot eviction detected - retrying with on-demand."
-fi
+fi

.github/workflows/ci.yml

@@ -153,7 +153,7 @@ jobs:
     steps:
       - uses: actions/checkout@v4
         with: { ref: "${{ env.GIT_COMMIT }}" }
-      - name: "CI (l1-contracts, avm-transpiler, noir-projects, yarn-project)"
+      - name: "Bootstrap (l1-contracts, avm-transpiler, noir-projects, yarn-project)"
         uses: ./.github/ensure-builder
         timeout-minutes: 40
         with:

ci3/retry

@@ -0,0 +1,19 @@
+#!/bin/bash
+set -u # not -e
+[ "${BUILD_SYSTEM_DEBUG:-}" = 1 ] && set -x
+
+if [ -n "${RETRY_DISABLED:-}" ]; then
+  set -e
+  eval "$"
+  exit
+fi
+
+ATTEMPTS=3
+# Retries up to 3 times with 5 second intervals
+for i in $(seq 1 $ATTEMPTS); do
+  eval "$*" && exit
+  [ "$i" != "$ATTEMPTS" ] && sleep 5
+done
+
+>&2 echo "$@ failed after $ATTEMPTS attempts"
+exit 1