Change version from 3.2.0 to 3.0.0-foxtrot

Install UEK8 in so-kernel-upgrade when no UEK kernel is present

VERSION

@@ -1 +1 @@
-3.2.0
+3.0.0-foxtrot

salt/common/tools/sbin/so-kernel-upgrade

@@ -5,53 +5,239 @@
 # https://securityonion.net/license; you may not use this file except in compliance with the
 # Elastic License 2.0.
 #
-# so-kernel-upgrade — switch the boot default to the installed UEK8 (6.x) kernel.
+# so-kernel-upgrade — install the UEK8 (6.x) kernel and make it the boot default.
 #
-# Security Onion is moving off the EL9 stock kernel / UEK7 (5.x) onto UEK8 (6.x).
+# Security Onion is moving off the EL9 stock kernel (RHCK, 5.14) and UEK7 (5.15) onto UEK8
-# Installing the kernel-uek-core package adds a UEK8 boot entry but does NOT make it the
+# (6.x). Three things have to happen, and the tool has to drive each one:
-# default: kernel-install/grubby only auto-promote a new kernel within the running
-# kernel's flavor lineage, and we're crossing from a 5.x kernel to the new 6.x UEK flavor.
-# So even with UPDATEDEFAULT=yes and DEFAULTKERNEL=kernel-uek-core the box keeps booting
-# the old kernel. This tool finds the newest installed 6.x UEK kernel and makes it the
-# GRUB default via grubby so the next boot comes up on UEK8.
 #
-# Idempotent: if the UEK8 kernel is already the default it does nothing. It only sets the
+#   1. Populate. The manager mirrors the UEK8 packages into /nsm/kernelrepo via so-repo-sync,
-# boot default; it does NOT reboot — the admin reboots the node on their own schedule.
+#      and serves them to the grid over https://<manager>/kernelrepo. Until that sync runs the
+#      repo is valid but EMPTY -- dnf resolves it happily and installs nothing, with no error.
+#   2. Install. A node on RHCK has no kernel-uek* package at all, so there is nothing for
+#      'dnf update' to upgrade. A node on UEK7 does have kernel-uek installed, so
+#      'dnf install kernel-uek' reports "Nothing to do" and exits 0 without installing 6.x.
+#      Both cases need an explicit install of the UEK8 NEVRA.
+#   3. Boot it. Whether a newly installed UEK8 kernel becomes the boot default depends on the
+#      RUNNING kernel's flavor. kernel-install/grubby (with UPDATEDEFAULT=yes) only auto-promote
+#      within the running kernel's flavor lineage:
+#        - From UEK7 (5.x, kernel-uek) the install stays in the kernel-uek lineage and IS
+#          auto-promoted, so no grubby change is needed -- just make sure the repo is populated
+#          and install UEK8.
+#        - From the stock EL9 kernel (RHCK, 5.14, no UEK) it is a flavor CROSS that is NOT
+#          auto-promoted, so the box keeps booting RHCK until grubby is told otherwise.
+#      This tool inspects the running kernel and only runs 'grubby --set-default' for RHCK.
+#
+# Every one of those failure modes is silent by default. This tool handles each case and fails
+# loudly when it cannot, rather than reporting success while changing nothing.
+#
+# Manager vs minion: only the manager owns /nsm/kernelrepo, so only the manager can populate
+# it. If the repo is empty here, a manager runs so-repo-sync itself; a minion has no way to
+# fix it and exits non-zero telling the admin to sync the manager first.
+#
+# Idempotent: an already-installed, already-default UEK8 kernel is left alone. It only sets
+# the boot default; it does NOT reboot -- the admin reboots the node on their own schedule.
+
+. /usr/sbin/so-common
+
+# Client-side repo id (what dnf enables on this node, from repo/client/oracle.sls) vs the
+# reposync-side section in repodownload.conf that the manager mirrors from (mirrors the
+# securityonion/securityonionsync split for the main repo).
+KERNEL_REPO="securityonionkernel"
+KERNEL_REPO_SYNC="securityonionkernelsync"
+KERNEL_PKG="kernel-uek"
+KERNEL_REPO_DIR="/nsm/kernelrepo"
+REPOSYNC_CONF="/opt/so/conf/reposync/repodownload.conf"
+GLOBAL_PILLAR="/opt/so/saltstack/local/pillar/global/soc_global.sls"
 
 log() { echo "[so-kernel-upgrade] $*"; }
+die() { echo "[so-kernel-upgrade] ERROR: $*" >&2; exit 1; }
 
-[ "$(id -u)" -eq 0 ] || { log "must run as root"; exit 1; }
+command -v grubby >/dev/null 2>&1 || die "grubby not found"
-command -v grubby >/dev/null 2>&1 || { log "grubby not found"; exit 1; }
+command -v dnf >/dev/null 2>&1 || die "dnf not found"
+
+ARCH="$(rpm -E '%{_arch}')"
+
+is_airgap() {
+    [ -f "$GLOBAL_PILLAR" ] && grep -q 'airgap: *[Tt]rue' "$GLOBAL_PILLAR"
+}
 
 # Newest installed UEK8 (6.x) kernel known to the bootloader. UEK8 vmlinuz paths look like
-# /boot/vmlinuz-6.12.0-203.76.7.5.el9uek.x86_64; the 5.x UEK7 and 5.14 RHCK won't match.
+# /boot/vmlinuz-6.12.0-204.92.4.2.el9uek.x86_64; UEK7 (5.15) and RHCK (5.14) won't match.
-target="$(grubby --info=ALL 2>/dev/null \
+find_uek8() {
-    | sed -n 's/^kernel="\(.*\)"$/\1/p' \
+    grubby --info=ALL 2>/dev/null \
-    | grep -E '/vmlinuz-6\.[0-9]+.*uek' \
+        | sed -n 's/^kernel="\(.*\)"$/\1/p' \
-    | sort -V | tail -1)"
+        | grep -E '/vmlinuz-6\.[0-9]+.*uek' \
+        | sort -V | tail -1
+}
 
-if [ -z "$target" ]; then
+# Classify the RUNNING kernel (uname -r) -- this, not what's installed, is what decides whether
-    log "no installed 6.x UEK (UEK8) kernel found — confirm the kernel repo is assigned and"
+# a UEK8 install auto-promotes to the boot default:
-    log "'dnf update' has installed kernel-uek-core. Nothing to do."
+#   uek8  6.x UEK    already on the target line; nothing to do
+#   uek7  5.x UEK    a UEK8 install stays in the kernel-uek lineage and auto-promotes (no grubby)
+#   rhck  5.14 EL9   crossing into the UEK flavor does NOT auto-promote (needs grubby --set-default)
+running_flavor() {
+    case "$(uname -r)" in
+        6.*uek*) echo uek8 ;;
+        *uek*)   echo uek7 ;;
+        *)       echo rhck ;;
+    esac
+}
+
+# Newest UEK8 kernel-uek NEVRA offered by the kernel repo, empty if the repo has none.
+# Restricted to the kernel repo so a UEK7 kernel-uek in the main repo can't be picked up,
+# and filtered to 6.x so we never "succeed" by reinstalling the 5.15 we already have.
+uek8_available() {
+    dnf -q repoquery --disablerepo='*' --enablerepo="$KERNEL_REPO" \
+        --arch="$ARCH" --latest-limit=1 \
+        --qf '%{name}-%{evr}.%{arch}\n' "$KERNEL_PKG" 2>/dev/null \
+        | grep -E "^${KERNEL_PKG}-6\." | tail -1
+}
+
+kernelrepo_rpm_count() {
+    find "$KERNEL_REPO_DIR" -maxdepth 1 -name '*.rpm' 2>/dev/null | wc -l
+}
+
+# The kernel repo starts life as valid-but-empty (kernelrepo_init_empty in
+# salt/manager/init.sls) and is filled by so-repo-sync. During a soup, so-repo-sync runs
+# BEFORE the highstate deploys the [securityonionkernelsync] section into repodownload.conf, so
+# the first kernel-aware soup leaves the repo empty until the next nightly sync.
+sync_kernel_repo() {
+    if is_airgap; then
+        log "airgap install: $KERNEL_REPO_DIR is populated from the airgap ISO, not by so-repo-sync."
+        return 1
+    fi
+    if ! grep -q "^\[${KERNEL_REPO_SYNC}\]" "$REPOSYNC_CONF" 2>/dev/null; then
+        log "$REPOSYNC_CONF has no [${KERNEL_REPO_SYNC}] section -- run a highstate to deploy it."
+        return 1
+    fi
+
+    log "populating $KERNEL_REPO_DIR with so-repo-sync (mirrors upstream; can take several minutes)"
+    su socore -c '/usr/sbin/so-repo-sync' || { log "so-repo-sync failed"; return 1; }
+
+    dnf -q clean expire-cache >/dev/null 2>&1
+    return 0
+}
+
+# Make the kernel repo actually able to serve a UEK8 package, or fail trying.
+ensure_kernel_repo() {
+    # The repo is assigned by the repo.client highstate, and only once NICs are pinned by MAC
+    # (/opt/so/state/nic_names_pinned) so the kernel swap can't renumber interfaces SO binds
+    # by name. skip_if_unavailable=1 means a broken repo is silently ignored, so check first.
+    if ! dnf -q repolist --enabled 2>/dev/null | awk '{print $1}' | grep -qx "$KERNEL_REPO"; then
+        log "repo '$KERNEL_REPO' is not enabled on this node."
+        log "Run a highstate first; the repo is skipped until /opt/so/state/nic_names_pinned"
+        log "exists (run so-nic-pin) and this node's salt matches the version this release ships."
+        die "kernel repo unavailable"
+    fi
+
+    [ -n "$(uek8_available)" ] && return 0
+
+    log "repo '$KERNEL_REPO' is enabled but offers no UEK8 $KERNEL_PKG package"
+
+    if ! is_manager_node; then
+        log "This is a minion; it consumes the kernel repo from the manager and cannot populate it."
+        log "On the manager, run:  su socore -c /usr/sbin/so-repo-sync"
+        log "then re-run this script here."
+        die "manager's kernel repo is empty"
+    fi
+
+    log "this is a manager and $KERNEL_REPO_DIR holds $(kernelrepo_rpm_count) rpm(s)"
+    sync_kernel_repo || die "could not populate $KERNEL_REPO_DIR"
+
+    [ -n "$(uek8_available)" ] \
+        || die "so-repo-sync completed but $KERNEL_REPO still offers no UEK8 $KERNEL_PKG"
+}
+
+reboot_notice() {
+    [ "$(uname -r)" = "$(basename "$1" | sed 's/^vmlinuz-//')" ] \
+        || log "REBOOT REQUIRED to start using the UEK8 kernel (currently running $(uname -r))."
+}
+
+# Keep future kernel updates on the UEK line rather than falling back to RHCK. Oracle ships
+# /etc/sysconfig/kernel; only rewrite it when it's actually pointing somewhere else.
+set_default_kernel_conf() {
+    if [ -f /etc/sysconfig/kernel ] && ! grep -q '^DEFAULTKERNEL=kernel-uek-core$' /etc/sysconfig/kernel; then
+        log "setting DEFAULTKERNEL=kernel-uek-core in /etc/sysconfig/kernel"
+        sed -i 's/^DEFAULTKERNEL=.*/DEFAULTKERNEL=kernel-uek-core/' /etc/sysconfig/kernel
+    fi
+}
+
+# Make sure a UEK8 kernel is installed, leaving its boot entry in INSTALLED_UEK8. If one is
+# already present we leave the repo alone -- it may be disabled or empty and we don't need it
+# just to flip the boot default. Otherwise install the explicit NEVRA, not the bare package
+# name: on a UEK7 node 'dnf install kernel-uek' sees 5.15 already present, prints "Nothing to
+# do" and exits 0 without installing 6.x.
+ensure_uek8_installed() {
+    INSTALLED_UEK8="$(find_uek8)"
+    if [ -n "$INSTALLED_UEK8" ]; then
+        log "UEK8 kernel already installed: $INSTALLED_UEK8"
+        return 0
+    fi
+
+    ensure_kernel_repo
+    local nevra; nevra="$(uek8_available)"
+    log "installing $nevra from $KERNEL_REPO"
+    dnf -y install "$nevra" || die "failed to install $nevra"
+
+    INSTALLED_UEK8="$(find_uek8)"
+    [ -n "$INSTALLED_UEK8" ] || die "$nevra installed but no 6.x UEK boot entry appeared -- check 'grubby --info=ALL'"
+    log "installed UEK8 kernel: $INSTALLED_UEK8"
+}
+
+case "$(running_flavor)" in
+uek8)
+    # Already on the 6.x UEK line. A plain 'dnf update' keeps this node current within the
+    # lineage and auto-promotes newer builds, so there is nothing for this tool to do.
+    log "already running a UEK8 kernel ($(uname -r)); nothing to do."
     exit 0
-fi
+    ;;
 
-current="$(grubby --default-kernel 2>/dev/null)"
+uek7)
-if [ "$current" = "$target" ]; then
+    # On a 5.x UEK kernel. Installing UEK8 stays inside the kernel-uek lineage, so dnf/grubby
-    log "UEK8 kernel is already the boot default: $target"
+    # (UPDATEDEFAULT=yes) auto-promote it and we do NOT touch grubby. A node still on UEK7
-    exit 0
+    # usually means the kernel repo was empty when it last updated, so populate it and install.
-fi
+    log "running UEK7 kernel ($(uname -r)); the kernel repo was likely not yet populated when"
+    log "this node last updated. Populating it and installing UEK8 -- the update stays on the"
+    log "kernel-uek line, so it becomes the boot default automatically (no grubby change needed)."
+    set_default_kernel_conf
+    ensure_uek8_installed
 
-log "current default kernel: ${current:-unknown}"
+    now="$(grubby --default-kernel 2>/dev/null)"
-log "switching boot default to UEK8 kernel: $target"
+    if [ "$now" = "$INSTALLED_UEK8" ]; then
-grubby --set-default="$target" || { log "ERROR: grubby --set-default failed for $target"; exit 1; }
+        log "boot default auto-promoted to UEK8 kernel: $INSTALLED_UEK8"
+    else
+        log "WARNING: expected the UEK8 kernel to auto-promote but the default is still"
+        log "'${now:-unknown}'. Run 'grubby --set-default=$INSTALLED_UEK8' to force it."
+    fi
+    reboot_notice "$INSTALLED_UEK8"
+    ;;
 
-# Verify the change actually took before claiming success.
+rhck)
-now="$(grubby --default-kernel 2>/dev/null)"
+    # On the stock EL9 kernel (5.14, no UEK installed). Crossing from RHCK into the UEK flavor
-if [ "$now" != "$target" ]; then
+    # does NOT auto-promote -- kernel-install/grubby only auto-promote within the running
-    log "ERROR: default kernel is still '${now:-unknown}' after set-default"
+    # kernel's flavor lineage -- so after installing we must set the boot default explicitly.
-    exit 1
+    log "running stock EL9 (RHCK) kernel ($(uname -r)); installing UEK8 and setting it as the"
-fi
+    log "boot default explicitly (a RHCK->UEK flavor change does not auto-promote)."
+    set_default_kernel_conf
+    ensure_uek8_installed
+    target="$INSTALLED_UEK8"
 
-log "boot default is now $target"
+    current="$(grubby --default-kernel 2>/dev/null)"
-log "REBOOT REQUIRED to start using the UEK8 kernel (currently running $(uname -r))."
+    if [ "$current" = "$target" ]; then
+        log "UEK8 kernel is already the boot default: $target"
+        reboot_notice "$target"
+        exit 0
+    fi
+
+    log "current default kernel: ${current:-unknown}"
+    log "switching boot default to UEK8 kernel: $target"
+    grubby --set-default="$target" || die "grubby --set-default failed for $target"
+
+    # Verify the change actually took before claiming success.
+    now="$(grubby --default-kernel 2>/dev/null)"
+    [ "$now" = "$target" ] || die "default kernel is still '${now:-unknown}' after set-default"
+
+    log "boot default is now $target"
+    reboot_notice "$target"
+    ;;
+esac

salt/elasticfleet/enabled.sls

@@ -111,6 +111,17 @@ so-elastic-fleet:
       - file: eastatedir
       - x509: etc_elasticfleet_key
       - x509: etc_elasticfleet_crt
+
+wait_for_so-elastic-fleet:
+  http.wait_for_successful_query:
+    - name: "https://localhost:8220/api/status"
+    - ssl: True
+    - verify_ssl: False
+    - status: 200
+    - wait_for: 300
+    - request_interval: 15
+    - require:
+      - docker_container: so-elastic-fleet
 {%   endif %}
 
 delete_so-elastic-fleet_so-status.disabled:

salt/manager/tools/sbin/soup

@@ -12,7 +12,17 @@
 UPDATE_DIR=/tmp/sogh/securityonion
 DEFAULT_SALT_DIR=/opt/so/saltstack/default
 INSTALLEDVERSION=$(cat /etc/soversion)
-POSTVERSION=$INSTALLEDVERSION
+# /etc/sopostversion is a soup-owned marker (no salt state manages it) tracking how
+# far the post-upgrade walk has progressed. Its presence means a prior upgrade did
+# not finish its post-upgrade steps; its contents are the resume point. It is read
+# here before preupgrade_changes mutates INSTALLEDVERSION and before any highstate
+# stamps /etc/soversion from the pillar.
+POSTVERSION_FILE=/etc/sopostversion
+if [ -f "$POSTVERSION_FILE" ]; then
+  POSTVERSION=$(cat "$POSTVERSION_FILE")
+else
+  POSTVERSION=$INSTALLEDVERSION
+fi
 INSTALLEDSALTVERSION=$(salt --versions-report | grep Salt: | awk '{print $2}')
 BATCHSIZE=5
 SOUP_LOG=/root/soup.log
@@ -23,6 +33,10 @@ NOTIFYCUSTOMELASTICCONFIG=false
 TOPFILE=/opt/so/saltstack/default/salt/top.sls
 BACKUPTOPFILE=/opt/so/saltstack/default/salt/top.sls.backup
 SALTUPGRADED=false
+# Set true once soup begins modifying the system (past the pre-flight checks), so the
+# EXIT trap can tell the user the update did not finish and must be re-run. Only the
+# pre-flight gates (ES compatibility, disk, network) fail before this is set.
+SOUP_UPGRADE_STARTED=false
 SALT_CLOUD_INSTALLED=false
 SALT_CLOUD_CONFIGURED=false
 # Check if salt-cloud is installed
@@ -123,6 +137,28 @@ check_err() {
 
     echo "SOUP XTRACE debug log (if enabled) at $SOUP_DEBUG_LOG. Re-run soup with SOUP_DEBUG=1 to create $SOUP_DEBUG_LOG"
 
+    # If soup had already started modifying the system, make it unmistakable that the
+    # update is incomplete and must be re-run. soup is resumable: a version upgrade
+    # picks up from the /etc/sopostversion marker, and a hotfix re-applies because
+    # /etc/sohotfix is only advanced after a successful highstate.
+    if [[ "$SOUP_UPGRADE_STARTED" == "true" ]]; then
+      echo ""
+      echo "=============================================================================="
+      echo " UPGRADE INCOMPLETE"
+      echo "=============================================================================="
+      echo " This soup run did NOT finish. Your Security Onion installation may be in a"
+      echo " partially-updated state and is not yet fully upgraded."
+      echo ""
+      echo " Review the error above and $SOUP_LOG, resolve the underlying problem, then"
+      echo " run soup again to resume and complete the update:"
+      echo ""
+      echo "     sudo soup"
+      echo ""
+      echo " soup is resumable -- re-running it continues from where this run stopped."
+      echo "=============================================================================="
+      echo ""
+    fi
+
     exit $exit_code
   fi
 
@@ -291,6 +327,30 @@ check_pillar_items() {
   fi
 }
 
+check_cluster_health() {
+  echo "Checking Elasticsearch cluster health."
+  # Require a 'green' cluster before upgrading; anything less (yellow, red, or
+  # unreachable) blocks. Modeled on the wait used in so-elasticsearch-roles-load.
+  if so-elasticsearch-query "_cluster/health?wait_for_status=green&timeout=120s" --fail > /dev/null 2>&1; then
+    printf "\nThe Elasticsearch cluster is healthy (green). We can proceed with SOUP.\n\n"
+  else
+    printf "\nThe Elasticsearch cluster is not green. Please resolve the cluster health issue so the cluster is green before running SOUP again.\n\n"
+    exit 0
+  fi
+}
+
+check_fleet_server() {
+  echo "Checking that Elastic Fleet Server is responding."
+  # Modeled on the wait_for_so-elastic-fleet state check in elasticfleet/enabled.sls,
+  # which waits for HTTP 200 from the Fleet Server status API.
+  if curl -sk --fail --retry 3 --retry-delay 10 --max-time 30 "https://localhost:8220/api/status" > /dev/null 2>&1; then
+    printf "\nElastic Fleet Server is responding. We can proceed with SOUP.\n\n"
+  else
+    printf "\nElastic Fleet Server is not responding at https://localhost:8220/api/status. Please ensure Elastic Fleet is healthy before running SOUP again.\n\n"
+    exit 0
+  fi
+}
+
 check_saltmaster_status() {
   set +e
   echo "Waiting on the Salt Master service to be ready."
@@ -414,6 +474,13 @@ preupgrade_changes() {
     true
 }
 
+set_postversion() {
+    # Persist post-upgrade walk progress so an interrupted upgrade can resume the
+    # remaining steps on the next soup run (see /etc/sopostversion handling).
+    POSTVERSION="$1"
+    echo "$POSTVERSION" > "$POSTVERSION_FILE"
+}
+
 postupgrade_changes() {
     # This function is to add any new pillar items if needed.
     echo "Running post upgrade processes."
@@ -421,6 +488,8 @@ postupgrade_changes() {
     [[ "$POSTVERSION" =~ ^2\.4\.21[0-9]+$ ]] && post_to_3.0.0
     [[ "$POSTVERSION" == "3.0.0" ]] && post_to_3.1.0
     [[ "$POSTVERSION" == "3.1.0" ]] && post_to_3.2.0
+    # All applicable post-upgrade steps completed; clear the resume marker.
+    rm -f "$POSTVERSION_FILE"
     true
 }
 
@@ -513,7 +582,7 @@ post_to_3.0.0() {
   # convert yes/no in suricata pillars to true/false
   convert_suricata_yes_no
 
-  POSTVERSION=3.0.0
+  set_postversion 3.0.0
 }
 
 ### 3.0.0 End ###
@@ -776,7 +845,7 @@ post_to_3.1.0() {
   # Check for unhealthy / unauthorized integration transform jobs and attempt reauthorizations
   check_transform_health_and_reauthorize || true
 
-  POSTVERSION=3.1.0
+  set_postversion 3.1.0
 }
 
 ### 3.1.0 End ###
@@ -911,7 +980,7 @@ post_to_3.2.0() {
 
   update_kafka_metadata "4.3"
 
-  POSTVERSION=3.2.0
+  set_postversion 3.2.0
 }
 
 ### 3.2.0 End ###
@@ -1063,8 +1132,20 @@ upgrade_check() {
   fi
   [[ -f /etc/sohotfix ]] && CURRENTHOTFIX=$(cat /etc/sohotfix)
   if [ "$INSTALLEDVERSION" == "$NEWVERSION" ]; then
+    # A leftover post-version marker means a previous upgrade to this version
+    # advanced /etc/soversion (the highstate stamps it from the pillar) but did not
+    # finish its post-upgrade steps. Resume the upgrade instead of reporting "latest".
+    if [ -f "$POSTVERSION_FILE" ] && [ "$(cat "$POSTVERSION_FILE")" != "$NEWVERSION" ]; then
+      echo "A previous upgrade to $NEWVERSION did not complete its post-upgrade steps; resuming."
+      is_hotfix=false
+      return 0
+    fi
     echo "Checking to see if there are hotfixes needed"
     if [ "$HOTFIXVERSION" == "$CURRENTHOTFIX" ]; then
+      # Reaching here means we are at the target version and NOT resuming (the resume
+      # check above returned otherwise). Clear any stale resume marker so a completed
+      # upgrade is never mistaken for a partial one and re-run on a later invocation.
+      rm -f "$POSTVERSION_FILE"
       echo "You are already running the latest version of Security Onion."
       exit 0
     else
@@ -1760,6 +1841,15 @@ main() {
   set_minionid
   MINION_ROLE=$(lookup_role)
   echo "Found that Security Onion $INSTALLEDVERSION is currently installed."
+  # /etc/soversion is stamped to the target version before the upgrade fully
+  # completes, so a lingering resume marker means this grid is only partially
+  # upgraded even though the line above shows the target version. Make that explicit
+  # so it is not mistaken for a finished upgrade.
+  if [ -f "$POSTVERSION_FILE" ] && [ "$(cat "$POSTVERSION_FILE")" != "$INSTALLEDVERSION" ]; then
+    echo ""
+    echo "NOTE: A previous upgrade to $INSTALLEDVERSION did not finish. This grid is"
+    echo "      partially upgraded and this soup run will resume and complete it."
+  fi
   echo ""
   check_minimum_version
 
@@ -1788,6 +1878,12 @@ main() {
   echo "Verifying Elasticsearch version compatibility across the grid before upgrading."
   verify_es_version_compatibility
 
+  # Pre-flight health checks: confirm the grid is in a good state before we change
+  # anything. These run before any modifications, so a failure exits cleanly and the
+  # operator can fix the issue and re-run soup.
+  check_cluster_health
+  check_fleet_server
+
   echo "Checking for Salt Master and Minion updates."
   upgrade_check_salt
   set -e
@@ -1804,6 +1900,7 @@ main() {
   fi
 
   if [ "$is_hotfix" == "true" ]; then
+    SOUP_UPGRADE_STARTED=true
     echo "Applying $HOTFIXVERSION hotfix"
     # since we don't run the backup.config_backup state on import we wont snapshot previous version states and pillars
     if [[ ! "$MINION_ROLE" == "import" ]]; then
@@ -1814,10 +1911,16 @@ main() {
     create_local_directories "/opt/so/saltstack/default"
     apply_hotfix
     echo "Hotfix applied"
-    update_version
     enable_highstate
     highstate
+    # Record the hotfix only after the highstate succeeds. /etc/sohotfix is written
+    # solely by soup (no salt state manages it), so deferring the write means a failed
+    # hotfix highstate leaves the old hotfix value and re-running soup re-applies it,
+    # rather than reporting "already latest". The soversion/pillar writes in
+    # update_version are no-ops here since the version is unchanged for a hotfix.
+    update_version
   else
+    SOUP_UPGRADE_STARTED=true
     echo ""
     echo "Performing upgrade from Security Onion $INSTALLEDVERSION to Security Onion $NEWVERSION."
     echo ""
@@ -1873,6 +1976,10 @@ main() {
     copy_new_files
     echo ""
     create_local_directories "/opt/so/saltstack/default"
+    # Seed the resume marker before the highstate stamps /etc/soversion to the new
+    # version, so an interrupted upgrade is detectable as "not finished" on re-run.
+    # POSTVERSION still holds the pre-upgrade (or prior resume) version here.
+    [ -f "$POSTVERSION_FILE" ] || echo "$POSTVERSION" > "$POSTVERSION_FILE"
     update_version
 
     echo ""