multipass/run_parallel_iiab_test.sh

@@ -33,14 +33,16 @@
 
 set -euo pipefail
 
+DPKG_ARCH="$(dpkg --print-architecture)"
 # Debian 13 (Trixie) official cloud image (qcow2). Multipass can launch from URL/file:// on Linux.
 # Source: Debian cloud images live under cloud.debian.org/images/cloud/ ('genericcloud' includes cloud-init).
-DEBIAN13_IMAGE_URL="${DEBIAN13_IMAGE_URL:-https://cloud.debian.org/images/cloud/trixie/latest/debian-13-genericcloud-amd64.qcow2}"
+DEBIAN13_IMAGE_URL="${DEBIAN13_IMAGE_URL:-https://cloud.debian.org/images/cloud/trixie/latest/debian-13-genericcloud-${DPKG_ARCH}.qcow2}"
 
 IMAGE="${IMAGE:-24.04}"
 BASE="${BASE:-ubu2404}"
 COUNT="${COUNT:-1}"
-CPUS="${CPUS:-3}"
+[ "$DPKG_ARCH" = "arm64" ] && CPUS="${CPUS:-2}" # SBC don't have spare CPUs.
+[ "$DPKG_ARCH" = "amd64" ] && CPUS="${CPUS:-3}"
 MEM="${MEM:-4G}"
 DISK="${DISK:-20G}"
 
@@ -53,6 +55,8 @@ LOCAL_VARS_URL="${LOCAL_VARS_URL:-https://raw.githubusercontent.com/iiab/iiab/re
 WAIT_TRIES="${WAIT_TRIES:-60}"     # used ONLY for the first auto-resume
 WAIT_SLEEP="${WAIT_SLEEP:-5}"
 STAGGER="${STAGGER:-20}"
+RESUME_TRIES="${RESUME_TRIES:-3}"
+RESUME_RETRY_SLEEP="${RESUME_RETRY_SLEEP:-8}"
 
 ACTION="run"
 modules=()
@@ -92,7 +96,7 @@ PR options:
   --run-pr N    Add PR number and force --run
 
 Env:
-  IMAGE BASE COUNT CPUS MEM DISK IIAB_PR IIAB_FAST LOCAL_VARS_URL WAIT_TRIES WAIT_SLEEP STAGGER
+  IMAGE BASE COUNT CPUS MEM DISK IIAB_PR IIAB_FAST LOCAL_VARS_URL WAIT_TRIES WAIT_SLEEP STAGGER RESUME_TRIES RESUME_RETRY_SLEEP
 EOF
 }
 
@@ -145,7 +149,7 @@ if [[ ( "$FIRST_UBUNTU" == "1" || "$FIRST_DEBIAN" == "1" ) && "$BOTH_DISTROS" !=
   echo "[ERROR] --first-ubuntu/--first-debian requires --both-distros."
   exit 2
 fi
-# ---- 
+# ----
 
 # Default module
 if [[ "${#modules[@]}" -eq 0 ]]; then
@@ -439,17 +443,56 @@ resume_iiab() {
       fi
     fi
 
-    multipass exec "$vm" -- bash -lc '
+    # Retry on rc=255 (SSH connection dropped; often reboot/network restart during apt/upgrade)
-      set -euo pipefail
+    # Also neutralize apt-listchanges/pagers to avoid "Waiting for data..." stalls.
-      echo "--- resume: sudo iiab -f ---"
+    local attempt r
-      if command -v iiab >/dev/null 2>&1; then
+    r=1
-        sudo iiab -f
+    for attempt in $(seq 1 "$RESUME_TRIES"); do
-      else
+      echo "[INFO] Resume attempt ${attempt}/${RESUME_TRIES} on $vm"
-        echo "[ERROR] iiab command not found; install likely not finished."
+
-        exit 89
+      multipass exec "$vm" -- bash -lc '
+        set -euo pipefail
+        echo "--- resume: sudo /usr/sbin/iiab -f ---"
+
+        if ! sudo test -x /usr/sbin/iiab; then
+          echo "[ERROR] /usr/sbin/iiab not found/executable; install likely not finished."
+          exit 89
+        fi
+
+        # Avoid interactive/pager behaviour during apt actions that IIAB may trigger.
+        # (apt-listchanges commonly causes: "Waiting for data... (interrupt to abort)")
+        sudo env \
+          DEBIAN_FRONTEND=noninteractive \
+          APT_LISTCHANGES_FRONTEND=none \
+          NEEDRESTART_MODE=a \
+          TERM=dumb \
+          PAGER=cat \
+          /usr/sbin/iiab -f
+
+        echo "--- resume done ---"
+      '
+
+      r=$?
+      if [[ "$r" -eq 0 ]]; then
+        break
       fi
-      echo "--- resume done ---"
+
-    '
+      if [[ "$r" -eq 255 ]]; then
+        echo "[WARN] multipass exec rc=255 (connection dropped; likely reboot). Waiting for VM and retrying..."
+        if ! wait_for_vm "$vm"; then
+          echo "[ERROR] VM did not become ready after reconnect wait: $vm"
+          r=88
+          break
+        fi
+        sleep "$RESUME_RETRY_SLEEP"
+        continue
+      fi
+
+      # Any other non-zero: don't loop forever.
+      break
+    done
+
+    exit "$r"
   } >"$log" 2>&1
   echo "$?" >"$rc"
   set -e