[PLT-5465/APIV4] GET /system/health - Improve ping health check to have limits (#6331)

* implement PLT-5465 - Improve ping health check to have limits

* update /ping and delete /health

* remove permission check

2 files changed, 32 insertions, 6 deletions

diff --git a/api4/system.go b/api4/system.go
index 1f4589bf5..3a077283c 100644
--- a/api4/system.go
+++ b/api4/system.go
@@ -5,6 +5,7 @@ package api4
 
 import (
 	"net/http"
+	"runtime"
 	"strconv"
 
 	l4g "github.com/alecthomas/log4go"
@@ -17,6 +18,7 @@ func InitSystem() {
 	l4g.Debug(utils.T("api.system.init.debug"))
 
 	BaseRoutes.System.Handle("/ping", ApiHandler(getSystemPing)).Methods("GET")
+
 	BaseRoutes.ApiRoot.Handle("/config", ApiSessionRequired(getConfig)).Methods("GET")
 	BaseRoutes.ApiRoot.Handle("/config", ApiSessionRequired(updateConfig)).Methods("PUT")
 	BaseRoutes.ApiRoot.Handle("/config/reload", ApiSessionRequired(configReload)).Methods("POST")
@@ -34,7 +36,19 @@ func InitSystem() {
 }
 
 func getSystemPing(c *Context, w http.ResponseWriter, r *http.Request) {
-	ReturnStatusOK(w)
+
+	actualGoroutines := runtime.NumGoroutine()
+	if *utils.Cfg.ServiceSettings.GoroutineHealthThreshold <= 0 || actualGoroutines <= *utils.Cfg.ServiceSettings.GoroutineHealthThreshold {
+		ReturnStatusOK(w)
+	} else {
+		rdata := map[string]string{}
+		rdata["status"] = "unhealthy"
+
+		l4g.Warn(utils.T("api.system.go_routines"), actualGoroutines, *utils.Cfg.ServiceSettings.GoroutineHealthThreshold)
+
+		w.WriteHeader(http.StatusInternalServerError)
+		w.Write([]byte(model.MapToJson(rdata)))
+	}
 }
 
 func testEmail(c *Context, w http.ResponseWriter, r *http.Request) {
diff --git a/api4/system_test.go b/api4/system_test.go
index a6b12d326..a46e14782 100644
--- a/api4/system_test.go
+++ b/api4/system_test.go
@@ -12,13 +12,26 @@ import (
 )
 
 func TestGetPing(t *testing.T) {
-	th := Setup().InitBasic()
+	th := Setup().InitBasic().InitSystemAdmin()
 	defer TearDown()
 	Client := th.Client
 
-	b, _ := Client.GetPing()
-	if b == false {
-		t.Fatal()
+	goRoutineHealthThreshold := *utils.Cfg.ServiceSettings.GoroutineHealthThreshold
+	defer func() {
+		*utils.Cfg.ServiceSettings.GoroutineHealthThreshold = goRoutineHealthThreshold
+	}()
+
+	status, resp := Client.GetPing()
+	CheckNoError(t, resp)
+	if status != "OK" {
+		t.Fatal("should return OK")
+	}
+
+	*utils.Cfg.ServiceSettings.GoroutineHealthThreshold = 10
+	status, resp = th.SystemAdminClient.GetPing()
+	CheckInternalErrorStatus(t, resp)
+	if status != "unhealthy" {
+		t.Fatal("should return unhealthy")
 	}
 }
 
@@ -342,5 +355,4 @@ func TestPostLog(t *testing.T) {
 	if len(logMessage) == 0 {
 		t.Fatal("should return the log message")
 	}
-
 }