Refactored and complete

.gitignore

@@ -1 +1,2 @@
 dockerize
+data

Dockerfile

@@ -2,8 +2,12 @@ FROM alpine:latest
 
 ENV DOCKERIZE_VERSION v0.7.0
 
+ENV DATA_DIR=/data
+
 WORKDIR /app
 
+RUN mkdir -p "$DATA_DIR"
+
 COPY *.sh ./
 COPY integrations/ integrations/
 

docker-cmd.sh

@@ -1,7 +1,9 @@
 #!/usr/bin/env bash
 
-source ./config.sh
-source ./logger.sh
+script_path=$(readlink -f $0)
+script_dir=${script_path%/*}
+source "$script_dir/config.sh"
+source "$script_dir/logger.sh"
 
 LOOP_SLEEP=${LOOP_SLEEP:-10s}
 
@@ -13,47 +15,67 @@ fi
 test -z "$ALERT_SCRIPT" && log_warn "Env ALERT_SCRIPT not defined - alerting disabled"
 test -z "$SWARM_NAME" && log_warn "Env SWARM_NAME not defined using default"
 
-SWARM_NAME="${SWARM_NAME:-Swarm}"
+swarm_name="${SWARM_NAME:-Swarm}"
+DATA_DIR=${DATA_DIR:-$script_dir/data}
+mkdir -p $DATA_DIR
 
 if [[ -n $ALERT_SCRIPT && ! -f $ALERT_SCRIPT ]]; then
-    log_error "Alert script defined but not accessible on $ALERT_SCRIPT path"
-    ALERT_SCRIPT=""
+    log_error "Alert script defined but not accessible on $ALERT_SCRIPT path!"
+    ALERT_SCRIPT="jq ."
 fi
 
 function check_services() {
-    local swarm_name=$SWARM_NAME
-    while read service network_alias port; do
-        read unique_id rest < <(echo "${swarm_name}_${service}_${network_alias}_${port}" | md5sum)
-        prefix="/tmp/alerter-${unique_id}"
-        alert_file=${prefix}.pending
-        log_file=${prefix}.log
+    while read service_name network_alias port; do
+        unique_name=$(echo "${swarm_name} ${service_name} ${network_alias} ${port}" )
+        unique_code=$(echo "${unique_name,,}" | sed -e 's/ /_/g' -e 's/[^a-zA-Z0-9_-]/_/g')
+        random_str=$(tr -dc 'a-zA-Z0-9' < /dev/urandom | head -c 10)
+        read unique_id _ < <(echo -n "$unique_name $random_str" | md5sum)
+        prefix="$DATA_DIR/${unique_code}"
+        pending_file="${prefix}.pending"
+        log_file="${prefix}.log"
         # used for testing
-        real_port=$port
-        if [[ -f test-change-port-$port ]]; then
-            read real_port < test-change-port-$port
+        real_port="$port"
+        if [[ -f "$DATA_DIR/test-change-port-$port" ]]; then
+            real_port=$(< "$DATA_DIR/test-change-port-$port")
         fi
+        action=""
         ./dockerize -timeout 5s -wait tcp://$network_alias:$real_port true 2>$log_file
         if [ $? -ne 0 ]; then
-            if [[ -f $alert_file ]]; then
-                log_warn "$service|$network_alias:$port|Pending alert"
+            if [[ -f $pending_file ]]; then
+                log_warn "$service_name|$network_alias:$port|Pending alert"
             else
-                log_error "$service|$network_alias:$port|Creating alert"
-                echo "$service $network_alias:$port"> $alert_file
-                if [[ -n $ALERT_SCRIPT ]]; then
-                    cat $log_file | /bin/bash -c "$ALERT_SCRIPT CREATE '$swarm_name' '$service' '$network_alias' '$port'"
-                fi
-                cat $log_file
+                log_error "$service_name|$network_alias:$port|Creating alert"
+                echo "$unique_id" > $pending_file
+                action="create"
             fi
         else
-            if [[ -f $alert_file ]]; then
-                log_info "$service|$network_alias:$port|Resolving alert"
-                if [[ -n $ALERT_SCRIPT ]]; then
-                    cat $log_file | /bin/bash -c "$ALERT_SCRIPT RESOLVE '$SWARM_NAME' '$service' '$network_alias' '$port'"
-                fi
-                rm -f $alert_file
+            if [[ -f $pending_file ]]; then
+                log_info "$service_name|$network_alias:$port|Resolving alert"
+                action="resolve"
+                unique_id=$(cat $pending_file)
+                rm -f $pending_file
             fi
-            rm -f $log_file
         fi
+        if [[ -n $action ]]; then
+            jq -n \
+                --arg action        "create" \
+                --arg unique_id     "$unique_id" \
+                --arg swarm_name    "$swarm_name" \
+                --arg service_name  "$service_name" \
+                --arg network_alias "$network_alias" \
+                --arg port          "$port" \
+                --arg log           "$(jq -R -s @json $log_file)" \
+                '{
+                  "action": $action,
+                  "unique_id": $unique_id,
+                  "swarm_name": $swarm_name,
+                  "service_name": $service_name,
+                  "network_alias": $network_alias,
+                  "port": $port,
+                  "log": $log
+                }' | /bin/bash -c "$ALERT_SCRIPT"
+        fi
+        rm -f $log_file
     done < <(./services.sh)
 }
 

integrations/zenduty.sh

@@ -1,61 +1,61 @@
 #!/usr/bin/env bash
 
-echo "Reading log file from stdin ..." >&2
-summary=$(cat | jq -Rs .)
-
 script_path=$(readlink -f $0)
 script_dir=${script_path%/*}
-source "$script_dir/../config.sh"
 source "$script_dir/../logger.sh"
 
-if [[ -z $ZENDUTY_API_KEY ]]; then
-    log_error "Expecting ZENDUTY_API_KEY env"
+DATA_DIR=${DATA_DIR:-$script_dir/../data}
+
+input_file=$(mktemp $DATA_DIR/zenduty-input.XXXXXX)
+trap "rm -f $input_file" EXIT
+
+if ! timeout 2s cat > $input_file; then
+    log_error "Reading from stdin timed out."
     exit 1
 fi
 
-if [[ $# != 5 ]]; then
-    log_error "Expecting parameters: ACTION SWARM_NAME SERVICE_NAME NETWORK_ALIAS PORT"
+if [[ -z $ZENDUTY_API_KEY ]]; then
+    log_error "Expecting ZENDUTY_API_KEY env"
     exit 1
 fi
 
-ACTION=$1
-SWARM_NAME=$2
-SERVICE_NAME=$3
-NETWORK_ALIAS=$4
-PORT=$5
+action=$(jq -r .action $input_file)
+entity_id=$(jq -r .unique_id $input_file)
 
 alert_type=""
-case $ACTION in
-    CREATE )
+appendix=""
+case $action in
+    create )
         alert_type="critical"
+        appendix="not available"
         ;;
-    RESOLVE )
+    resolve )
         alert_type="resolved"
+        appendix="is available"
         ;;
     *)
-        log_error "Action must be one of: CREATE RESOLVE. Received: '$ACTION'"
+        log_error "Action must be one of: create resolve. Received: '$action'"
         ;;
 esac
 
-read entity_id rest < <(echo "${SWARM_NAME}_${SERVICE_NAME}_${NETWORK_ALIAS}_${PORT}" | md5sum)
-
-request_file=/tmp/zenduty-request-${entity_id}.json
-response_file=/tmp/zenduty-response-${entity_id}.json
+request_file=$DATA_DIR/${entity_id}-zenduty-request.json
+response_file=$DATA_DIR/${entity_id}-zenduty-response.json
 
-cat << __PAYLOAD > $request_file
-{
-    "alert_type": "$alert_type",
-    "entity_id": "$entity_id",
-    "message":"$SWARM_NAME service $SERVICE_NAME ($NETWORK_ALIAS:$PORT) not available",
-    "summary": $summary
-}
-__PAYLOAD
+jq -r \
+    --arg alert_type "$alert_type" \
+    --arg appendix "$appendix" \
+    '{
+        "alert_type": $alert_type,
+        "entity_id": .unique_id,
+        "message": "\(.swarm_name) service \(.service_name) (\(.network_alias):\(.port)) \($appendix)",
+        "summary": .log
+    }' $input_file > $request_file
 
-# log_info "Request file:"
-# jq . $request_file 2>/dev/null || cat $request_file
+log_info "Request file:"
+jq . $request_file 2>/dev/null || cat $request_file
 
 url="https://www.zenduty.com/api/events/${ZENDUTY_API_KEY}/"
-curl -s -X POST "$url" -H 'Content-Type: application/json' -d @$request_file >$response_file 2>&1
+curl -s -X POST "$url" -H 'Content-Type: application/json' -d @$request_file >$response_file
 return_code=$?
 
 if [ $return_code -ne 0 ]; then
@@ -65,7 +65,7 @@ fi
 log_info "Response file:"
 jq . $response_file 2>/dev/null || cat $response_file
 
-if [[ $ACTION == "RESOLVE" ]]; then
+if [[ $action == "resolve" ]]; then
     rm -f $request_file
     rm -f $response_file
 fi

test.sh

@@ -13,6 +13,7 @@ docker run --rm \
     --env ALERT_SCRIPT="$ALERT_SCRIPT" \
     --env SWARM_NAME="$SWARM_NAME" \
     --env ZENDUTY_API_KEY="$ZENDUTY_API_KEY" \
+    --env DATA_DIR=/app/data \
     --volume /var/run/docker.sock:/var/run/docker.sock \
     --volume .:/app/ \
     brablc/swarm-health-alerter:dev "$@"