fix: update the make stop script logic and fix the make stop don't stop all process's bug

cmd/openim-api/main.go

@@ -119,7 +119,7 @@ func run(port int, proPort int) error {
 	defer cancel()
 	select {
 	case <-sigs:
-		util.SIGUSR1Exit()
+		util.SIGTERMExit()
 		err := server.Shutdown(ctx)
 		if err != nil {
 			return errs.Wrap(err, "shutdown err")

internal/msggateway/init.go

@@ -46,9 +46,7 @@ func RunWsAndServer(rpcPort, wsPort, prometheusPort int) error {
 	netDone := make(chan error)
 	go func() {
 		err = hubServer.Start()
-		if err != nil {
-			netDone <- err
-		}
+		netDone <- err
 	}()
 	return hubServer.LongConnServer.Run(netDone)
 }

internal/msgtransfer/init.go

@@ -18,12 +18,11 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	util "github.com/openimsdk/open-im-server/v3/pkg/util/genutil"
 
 	"github.com/OpenIMSDK/tools/errs"
 	"github.com/OpenIMSDK/tools/log"
 
-	util "github.com/openimsdk/open-im-server/v3/pkg/util/genutil"
-
 	"net/http"
 	"os"
 	"os/signal"
@@ -144,7 +143,7 @@ func (m *MsgTransfer) Start(prometheusPort int) error {
 	signal.Notify(sigs, syscall.SIGTERM)
 	select {
 	case <-sigs:
-		util.SIGUSR1Exit()
+		util.SIGTERMExit()
 		// graceful close kafka client.
 		m.cancel()
 		m.historyCH.historyConsumerGroup.Close()

pkg/common/startrpc/start.go

@@ -140,7 +140,7 @@ func Start(
 	signal.Notify(sigs, syscall.SIGTERM)
 	select {
 	case <-sigs:
-		util.SIGUSR1Exit()
+		util.SIGTERMExit()
 		ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
 		defer cancel()
 		if err := gracefulStopWithCtx(ctx, srv.GracefulStop); err != nil {
@@ -152,7 +152,7 @@ func Start(
 		if err != nil {
 			return errs.Wrap(err, "shutdown err")
 		}
-		return errors.New("SIGTERM EXIT")
+		return nil
 	case <-netDone:
 		close(netDone)
 		return netErr

pkg/util/genutil/genutil.go

@@ -15,7 +15,6 @@
 package genutil
 
 import (
-	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
@@ -42,15 +41,12 @@ func OutDir(path string) (string, error) {
 }
 
 func ExitWithError(err error) {
-	if errors.Is(err, errors.New("SIGTERM EXIT")) {
-		os.Exit(-1)
-	}
 	progName := filepath.Base(os.Args[0])
 	fmt.Fprintf(os.Stderr, "\n\n%s exit -1: \n%+v\n\n", progName, err)
 	os.Exit(-1)
 }
 
-func SIGUSR1Exit() {
+func SIGTERMExit() {
 	progName := filepath.Base(os.Args[0])
-	fmt.Printf("\n\n%s receive process terminal SIGTERM exit 0\n\n", progName)
+	fmt.Fprintf(os.Stderr, "Warning %s receive process terminal SIGTERM exit 0\n", progName)
 }

scripts/install/openim-api.sh

@@ -82,6 +82,7 @@ function openim::api::start_service() {
   echo "Starting service with command: $cmd"
 
   nohup $cmd >> "${LOG_FILE}" 2> >(tee -a "${STDERR_LOG_FILE}" "$TMP_LOG_FILE" >&2) &
+
 
   if [ $? -ne 0 ]; then
     openim::log::error_exit "Failed to start ${binary_name} on port ${service_port}."

scripts/install/openim-msgtransfer.sh

@@ -23,7 +23,6 @@ set -o pipefail
 OPENIM_ROOT=$(cd "$(dirname "${BASH_SOURCE[0]}")"/../.. && pwd -P)
 [[ -z ${COMMON_SOURCED} ]] && source "${OPENIM_ROOT}"/scripts/install/common.sh
 
-openim::util::set_max_fd 200000
 
 SERVER_NAME="openim-msgtransfer"
 
@@ -88,6 +87,32 @@ function openim::msgtransfer::check() {
   fi
 }
 
+function openim::msgtransfer::check_by_signal() {
+  PIDS=$(pgrep -f "${OPENIM_OUTPUT_HOSTBIN}/openim-msgtransfer") || PIDS="0"
+  if [  "$PIDS" = "0" ]; then
+      return 0
+  fi
+
+  NUM_PROCESSES=$(echo "$PIDS" | wc -l | xargs)
+
+  if [ "$NUM_PROCESSES" -gt 0 ]; then
+    openim::log::error "Found $NUM_PROCESSES processes for $OPENIM_OUTPUT_HOSTBIN/openim-msgtransfer"
+    for PID in $PIDS; do
+      if [[ "$OSTYPE" == "linux-gnu"* ]]; then
+        echo -e "\033[31m$(ps -p $PID -o pid,cmd)\033[0m"
+      elif [[ "$OSTYPE" == "darwin"* ]]; then
+        echo -e "\033[31m$(ps -p $PID -o pid,comm)\033[0m"
+      else
+        openim::log::error "Unsupported OS type: $OSTYPE"
+      fi
+    done
+    openim::log::error "Processes have not been stopped properly."
+  else
+    openim::log::success "All openim-msgtransfer processes have been stopped properly."
+  fi
+  return 0
+}
+
 ###################################### Linux Systemd ######################################
 SYSTEM_FILE_PATH="/etc/systemd/system/${SERVER_NAME}.service"
 

scripts/lib/util.sh

@@ -370,6 +370,81 @@ openim::util::check_ports() {
   fi
 }
 
+
+openim::util::check_ports_by_signal() {
+  # An array to collect ports of processes that are not running.
+  local not_started=()
+
+  # An array to collect information about processes that are running.
+  local started=()
+
+  # Iterate over each given port.
+  for port in "$@"; do
+    # Initialize variables
+    # Check the OS and use the appropriate command
+    if [[ "$OSTYPE" == "linux-gnu"* ]]; then
+      if command -v ss > /dev/null 2>&1; then
+        info=$(ss -ltnp | grep ":$port" || true)
+      else
+        info=$(netstat -ltnp | grep ":$port" || true)
+      fi
+      elif [[ "$OSTYPE" == "darwin"* ]]; then
+      # For macOS, use lsof
+      info=$(lsof -P -i:"$port" | grep "LISTEN" || true)
+    fi
+
+    # Check if any process is using the port
+    if [[ -z $info ]]; then
+      not_started+=($port)
+    else
+      if [[ "$OSTYPE" == "linux-gnu"* ]]; then
+        # Extract relevant details for Linux: Process Name, PID, and FD.
+        details=$(echo $info | sed -n 's/.*users:(("\([^"]*\)",pid=\([^,]*\),fd=\([^)]*\))).*/\1 \2 \3/p')
+        command=$(echo $details | awk '{print $1}')
+        pid=$(echo $details | awk '{print $2}')
+        fd=$(echo $details | awk '{print $3}')
+        elif [[ "$OSTYPE" == "darwin"* ]]; then
+        # Handle extraction for macOS
+        pid=$(echo $info | awk '{print $2}' | cut -d'/' -f1)
+        command=$(ps -p $pid -o comm= | xargs basename)
+        fd=$(echo $info | awk '{print $4}' | cut -d'/' -f1)
+      fi
+
+      # Get the start time of the process using the PID
+      if [[ -z $pid ]]; then
+        start_time="N/A"
+      else
+        start_time=$(ps -p $pid -o lstart=)
+      fi
+
+      started+=("Port $port - Command: $command, PID: $pid, FD: $fd, Started: $start_time")
+    fi
+  done
+
+#  # Print information about ports whose processes are not running.
+#  if [[ ${#not_started[@]} -ne 0 ]]; then
+#    openim::log::info "\n### Not started ports:"
+#    for port in "${not_started[@]}"; do
+#      openim::log::error "Port $port is not started."
+#    done
+#  fi
+
+  # Print information about ports whose processes are running.
+  if [[ ${#started[@]} -ne 0 ]]; then
+    openim::log::error "\n### No stop ports:"
+    for info in "${started[@]}"; do
+      openim::log::error "$info"
+    done
+  fi
+
+  # If any of the processes is not running, return a status of 1.
+  if [[ ${#not_started[@]} -ne 0 ]]; then
+     openim::log::success "All specified processes are stop."
+  else
+    openim::log::error "Have processes no stop."
+  fi
+}
+
 # set +o errexit
 # Sample call for testing:
 # openim::util::check_ports 10002 1004 12345 13306
@@ -476,24 +551,22 @@ openim::util::stop_services_on_ports() {
   openim::log::info "Stopping services on ports: $*"
   # Iterate over each given port.
   for port in "$@"; do
-    # Use the `lsof` command to find process information related to the given port.
-    info=$(lsof -i :$port -n -P | grep LISTEN || true)
-
-    # If there's process information, it means the process associated with the port is running.
+    local info=$(lsof -i :$port -n -P | grep LISTEN || true)
     if [[ -n $info ]]; then
-      # Extract the Process ID.
+      local stopped_this_port=false
       while read -r line; do
         local pid=$(echo $line | awk '{print $2}')
-
-        # Try to stop the service by killing its process.
-        if kill -15 $pid; then
-          stopped+=($port)
-        else
-          not_stopped+=($port)
+        if kill -15 "$pid" &> /dev/null; then
+          stopped+=("$port")
+          stopped_this_port=true
+          break  # Jump out of loop after successfully sending SIGTERM
         fi
-            done <<< "$info"
-        fi
-    done
+      done <<< "$info"
+      if ! $stopped_this_port; then
+        not_stopped+=("$port")
+      fi
+    fi
+  done
 
     # Print information about ports whose processes couldn't be stopped.
     if [[ ${#not_stopped[@]} -ne 0 ]]; then
@@ -519,6 +592,35 @@ openim::util::stop_services_on_ports() {
         return 0
     fi
 }
+
+openim::util::stop_services_by_signal() {
+  # An array to collect ports of processes that couldn't be stopped.
+  local not_stopped=()
+
+  # An array to collect information about processes that were stopped.
+  local stopped=()
+
+  # Iterate over each given port.
+  for port in "$@"; do
+    # Use the `lsof` command to find process information related to the given port.
+    info=$(lsof -i :$port -n -P | grep LISTEN || true)
+
+    # If there's process information, it means the process associated with the port is running.
+    if [[ -n $info ]]; then
+      # Extract the Process ID.
+      while read -r line; do
+        local pid=$(echo $line | awk '{print $2}')
+
+        # Try to stop the service by killing its process.
+        if kill -15 $pid; then
+          stopped+=($port)
+        else
+          not_stopped+=($port)
+        fi
+            done <<< "$info"
+        fi
+    done
+}
 # nc -l -p 12345
 # nc -l -p 123456
 # ps -ef | grep "nc -l"
@@ -594,6 +696,48 @@ openim::util::stop_services_with_name() {
     openim::log::success "All specified services were stopped."
     echo ""
 }
+
+openim::util::stop_services_by_name_signal() {
+    # An array to collect names of processes that couldn't be stopped.
+    local not_stopped=()
+
+    # An array to collect information about processes that were stopped.
+    local stopped=()
+
+    # Iterate over each given service name.
+    for server_name in "$@"; do
+        # Use the `pgrep` command to find process IDs related to the given service name.
+        local pids=$(pgrep -f "$server_name")
+
+        # If no process was found with the name, add it to the not_stopped list
+        if [[ -z $pids ]]; then
+            not_stopped+=("$server_name")
+            continue
+        fi
+        local stopped_this_time=false
+        for pid in $pids; do
+
+            # Exclude the PID of the current script
+            if [[ "$pid" == "$$" ]]; then
+                continue
+            fi
+
+            # If there's a Process ID, it means the service with the name is running.
+            if [[ -n $pid ]]; then
+                # Try to stop the service by killing its process.
+                if kill -15 $pid 2>/dev/null; then
+                    stopped_this_time=true
+                fi
+            fi
+        done
+
+        if $stopped_this_time; then
+            stopped+=("$server_name")
+        else
+            not_stopped+=("$server_name")
+        fi
+    done
+}
 # sleep 333333&
 # sleep 444444&
 # ps -ef | grep "sleep"
@@ -1708,25 +1852,22 @@ openim::util::stop_services_on_ports() {
     local stopped=()
 
     openim::log::info "Stopping services on ports: $*"
-    # Iterate over each given port.
     for port in "$@"; do
-        # Use the `lsof` command to find process information related to the given port.
-        info=$(lsof -i :$port -n -P | grep LISTEN || true)
-
-        # If there's process information, it means the process associated with the port is running.
-        if [[ -n $info ]]; then
-            # Extract the Process ID.
-            while read -r line; do
-                local pid=$(echo $line | awk '{print $2}')
-
-                # Try to stop the service by killing its process.
-                if kill -10 $pid; then
-                    stopped+=($port)
-                else
-                    not_stopped+=($port)
-                fi
-            done <<< "$info"
+      local info=$(lsof -i :$port -n -P | grep LISTEN || true)
+      if [[ -n $info ]]; then
+        local stopped_this_port=false
+        while read -r line; do
+          local pid=$(echo $line | awk '{print $2}')
+          if kill -15 "$pid" &> /dev/null; then
+            stopped+=("$port")
+            stopped_this_port=true
+            break  # Jumping out of cycle m after successfully sending SIGTERM
+          fi
+        done <<< "$info"
+        if ! $stopped_this_port; then
+          not_stopped+=("$port")
         fi
+      fi
     done
 
     # Print information about ports whose processes couldn't be stopped.