Merge pull request #15907 from Security-Onion-Solutions/reyesj2/es-verify-compat

Verify compatibility for all ES nodes in the cluster

salt/manager/tools/sbin/soup

@@ -971,6 +971,9 @@ verify_es_version_compatibility() {
     local is_active_intermediate_upgrade=1
     # supported upgrade paths for SO-ES versions
     declare -A es_upgrade_map=(
+        ["8.18.4"]="8.18.6 8.18.8 9.0.8"
+	    ["8.18.6"]="8.18.8 9.0.8"
+	    ["8.18.8"]="9.0.8"
         ["9.0.8"]="9.3.3"
     )
 
@@ -994,6 +997,171 @@ verify_es_version_compatibility() {
         exit 160
     fi
 
+    compatible_es_versions="$target_es_version"
+    for current_version in "${!es_upgrade_map[@]}"; do
+        # shellcheck disable=SC2076
+        if [[ " ${es_upgrade_map[$current_version]} " =~ " $target_es_version " ]]; then
+            compatible_es_versions+=" $current_version"
+        fi
+    done
+
+    # Check if the given ES version can directly upgrade to the target ES version. Used to assist with catching lagging nodes during the upgrade process
+    es_version_can_upgrade_to_target() {
+        local current_version="$1"
+        # shellcheck disable=SC2076
+        if [[ -n "$current_version" && " $compatible_es_versions " =~ " $current_version " ]]; then
+            return 0
+        fi
+
+        return 1
+    }
+
+    # Gather Elasticsearch cluster version info and verify that each node in the cluster is running a version compatible with the target ES version.
+    verify_searchnodes_es_target_compatibility() {
+        local retries=20
+        local retry_count=0
+        local delay=180
+        local expected_es_nodes searchnode_minions attempt
+        local searchnode_discovery_success=false
+        SEARCHNODE_ES_VERSIONS=""
+
+        for attempt in {1..3}; do
+            if searchnode_minions=$(set -o pipefail; salt-key --out=json --list=accepted 2> /dev/null | jq -r '.minions[]? | select(endswith("searchnode"))'); then
+                searchnode_discovery_success=true
+                break
+            fi
+
+            echo "Failed to retrieve grid searchnodes via salt-key... Retrying in 30 seconds. Attempt $attempt of 3."
+            sleep 30
+        done
+
+        if [[ "$searchnode_discovery_success" != "true" ]]; then
+            echo "Failed to retrieve grid searchnodes via salt-key."
+            return 1
+        fi
+
+        # Always add node running soup to expected es nodes
+        expected_es_nodes="${MINIONID%_*}"
+        while IFS= read -r searchnode_minion; do
+            [[ -z "$searchnode_minion" ]] && continue
+            expected_es_nodes+=$'\n'"${searchnode_minion%_searchnode}"
+        done <<< "$searchnode_minions"
+
+        while [[ $retry_count -lt $retries ]]; do
+            SEARCHNODE_ES_VERSIONS=$(so-elasticsearch-query _nodes/_all/version --retry 5 --retry-delay 10 --fail 2>&1)
+            local exit_status=$?
+
+            if [[ $exit_status -ne 0 ]]; then
+                echo "Failed to retrieve Elasticsearch versions from searchnodes... Retrying in $delay seconds. Attempt $((retry_count + 1)) of $retries."
+                ((retry_count++))
+                sleep $delay
+                continue
+            fi
+
+            local all_searchnodes_compatible=true
+            while IFS=$'\t' read -r node current_version; do
+                [[ -z "$node" ]] && continue
+                if ! es_version_can_upgrade_to_target "$current_version"; then
+                    echo "Searchnode $node is running Elasticsearch $current_version, which is not directly upgradable to Elasticsearch $target_es_version."
+                    all_searchnodes_compatible=false
+                fi
+            done < <(echo "$SEARCHNODE_ES_VERSIONS" | jq -r '.nodes | to_entries[] | [.value.name, .value.version] | @tsv')
+
+            while IFS= read -r expected_es_node; do
+                [[ -z "$expected_es_node" ]] && continue
+                if ! echo "$SEARCHNODE_ES_VERSIONS" | jq -e --arg node "$expected_es_node" '.nodes | to_entries | any(.value.name == $node)' > /dev/null; then
+                    echo "Searchnode $expected_es_node did not report an Elasticsearch version. It may be offline or still upgrading."
+                    all_searchnodes_compatible=false
+                fi
+            done <<< "$expected_es_nodes"
+
+            if [[ "$all_searchnodes_compatible" == true ]]; then
+                echo "All Searchnodes are upgradable to Elasticsearch $target_es_version."
+                return 0
+            fi
+
+            echo "One or more Searchnodes cannot upgrade directly to Elasticsearch $target_es_version. Rechecking in $delay seconds. Attempt $((retry_count + 1)) of $retries."
+            ((retry_count++))
+            sleep $delay
+        done
+
+        return 1
+    }
+
+    # Gather heavynode version info and verify that each node is running a version compatible with the target ES version.
+    verify_heavynodes_es_target_compatibility() {
+        local heavynode_minions attempt
+        local retries=20
+        local retry_count=0
+        local delay=180
+        local heavynode_discovery_success=false
+        HEAVYNODE_ES_VERSIONS=""
+
+        for attempt in {1..3}; do
+            if heavynode_minions=$(set -o pipefail; salt-key --out=json --list=accepted 2> /dev/null | jq -r '.minions[]? | select(endswith("heavynode"))'); then
+                heavynode_discovery_success=true
+                break
+            fi
+
+            echo "Failed to retrieve grid heavynodes via salt-key... Retrying in 30 seconds. Attempt $attempt of 3."
+            sleep 30
+        done
+
+        if [[ "$heavynode_discovery_success" != "true" ]]; then
+            echo "Failed to retrieve grid heavynodes via salt-key."
+            return 1
+        fi
+
+        if [[ -z "$heavynode_minions" ]]; then
+            echo "No heavynodes detected. Skipping heavynode Elasticsearch version compatibility check."
+            return 0
+        fi
+
+        while [[ $retry_count -lt $retries ]]; do
+            HEAVYNODE_ES_VERSIONS=$(salt -C 'G@role:so-heavynode' cmd.run 'set -o pipefail; so-elasticsearch-query / --retry 5 --retry-delay 10 | jq -er ".version.number"' shell=/bin/bash --out=json 2> /dev/null)
+            local exit_status=$?
+
+            if [[ $exit_status -ne 0 ]]; then
+                echo "Failed to retrieve Elasticsearch version from one or more heavynodes... Retrying in $delay seconds. Attempt $((retry_count + 1)) of $retries."
+                ((retry_count++))
+                sleep $delay
+                continue
+            fi
+
+            local all_heavynodes_compatible=true
+            while IFS=$'\t' read -r node current_version; do
+                [[ -z "$node" ]] && continue
+                if ! es_version_can_upgrade_to_target "$current_version"; then
+                    echo "Heavynode $node is running Elasticsearch $current_version, which is not directly upgradable to Elasticsearch $target_es_version."
+                    all_heavynodes_compatible=false
+                fi
+            done < <(echo "$HEAVYNODE_ES_VERSIONS" | jq -r 'to_entries[] | [.key, .value] | @tsv')
+
+            while IFS= read -r heavynode_minion; do
+                [[ -z "$heavynode_minion" ]] && continue
+                if ! echo "$HEAVYNODE_ES_VERSIONS" | jq -e --arg minion "$heavynode_minion" 'has($minion)' > /dev/null; then
+                    echo "Heavynode $heavynode_minion did not report an Elasticsearch version. It may be offline or still upgrading."
+                    all_heavynodes_compatible=false
+                fi
+            done <<< "$heavynode_minions"
+
+            if [[ "$all_heavynodes_compatible" == true ]]; then
+                echo -e "\nAll heavynodes can upgrade to Elasticsearch $target_es_version."
+                return 0
+            fi
+
+            echo "One or more heavynodes cannot upgrade directly to Elasticsearch $target_es_version. Rechecking in $delay seconds. Attempt $((retry_count + 1)) of $retries."
+            ((retry_count++))
+            sleep $delay
+        done
+
+        return 1
+    }
+
+    if [[ ! -f "$es_verification_script" ]]; then
+        create_intermediate_upgrade_verification_script "$es_verification_script"
+    fi
+
     for statefile in "${es_required_version_statefile_base}"-*; do
         [[ -f $statefile ]] || continue
 
@@ -1012,10 +1180,6 @@ verify_es_version_compatibility() {
             continue
         fi
 
-        if [[ ! -f "$es_verification_script" ]]; then
-            create_intermediate_upgrade_verification_script "$es_verification_script"
-        fi
-
         echo -e "\n##############################################################################################################################\n"
         echo "A previously required intermediate Elasticsearch upgrade was detected. Verifying that all Searchnodes/Heavynodes have successfully upgraded Elasticsearch to $es_required_version_statefile_value before proceeding with soup to avoid potential data loss! This command can take up to an hour to complete."
         if ! timeout --foreground 4000 bash "$es_verification_script" "$es_required_version_statefile_value" "$statefile"; then
@@ -1037,6 +1201,26 @@ verify_es_version_compatibility() {
 
     # shellcheck disable=SC2076 # Do not want a regex here eg usage " 8.18.8 9.0.8 " =~ " 9.0.8 "
     if [[ " ${es_upgrade_map[$es_version]} " =~ " $target_es_version " || "$es_version" == "$target_es_version" ]]; then
+        if ! verify_searchnodes_es_target_compatibility || ! verify_heavynodes_es_target_compatibility; then
+            echo -e "\n!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n"
+
+            echo "One or more Searchnode(s)/Heavynode(s) cannot upgrade directly to Elasticsearch $target_es_version. This can happen with soups that include Elasticsearch upgrades being run in quick succession. Typically, this will resolve itself as the grid synchronizes. Please allow time for all Searchnodes/Heavynodes to have upgraded Elasticsearch to a compatible version with $target_es_version before running soup again to avoid potential data loss!"
+
+            if [[ -n "$HEAVYNODE_ES_VERSIONS" ]]; then
+                echo "Current heavynode Elasticsearch versions:"
+                echo "$HEAVYNODE_ES_VERSIONS" | jq '.'
+            fi
+
+            if [[ -n "$SEARCHNODE_ES_VERSIONS" ]]; then
+                echo "Current searchnode Elasticsearch versions:"
+                echo "$SEARCHNODE_ES_VERSIONS" | jq '.nodes | to_entries | map({(.value.name): .value.version}) | sort | add'
+            fi
+
+            echo -e "\n!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n"
+
+            exit 161
+        fi
+
         # supported upgrade
         return 0
     else
@@ -1394,7 +1578,7 @@ main() {
   echo "Verifying we have the latest soup script."
   verify_latest_update_script
 
-  echo "Verifying Elasticsearch version compatibility before upgrading."
+  echo "Verifying Elasticsearch version compatibility across the grid before upgrading."
   verify_es_version_compatibility
 
   echo "Let's see if we need to update Security Onion."