Merge pull request #2990 from overleaf/bg-add-health-check-on-active-handles

add health check limit on active handles (optional)

GitOrigin-RevId: c26071c96b220935178012d881bb53013df64155

services/web/app/src/Features/HealthCheck/HealthCheckController.js

@@ -57,6 +57,26 @@ module.exports = HealthCheckController = {
     })
   },
 
+  checkActiveHandles(req, res, next) {
+    if (!(settings.maxActiveHandles > 0) || !process._getActiveHandles) {
+      return next()
+    }
+    const activeHandlesCount = (process._getActiveHandles() || []).length
+    if (activeHandlesCount > settings.maxActiveHandles) {
+      logger.err(
+        { activeHandlesCount, maxActiveHandles: settings.maxActiveHandles },
+        'exceeded max active handles, failing health check'
+      )
+      return res.sendStatus(500)
+    } else {
+      logger.debug(
+        { activeHandlesCount, maxActiveHandles: settings.maxActiveHandles },
+        'active handles are below maximum'
+      )
+      next()
+    }
+  },
+
   checkApi(req, res, next) {
     rclient.healthCheck(err => {
       if (err) {

services/web/app/src/router.js

@@ -981,8 +981,16 @@ function initialize(webRouter, privateApiRouter, publicApiRouter) {
   // used by kubernetes health-check and acceptance tests
   webRouter.get('/dev/csrf', (req, res) => res.send(res.locals.csrfToken))
 
-  publicApiRouter.get('/health_check', HealthCheckController.check)
-  privateApiRouter.get('/health_check', HealthCheckController.checkApi)
+  publicApiRouter.get(
+    '/health_check',
+    HealthCheckController.checkActiveHandles,
+    HealthCheckController.check
+  )
+  privateApiRouter.get(
+    '/health_check',
+    HealthCheckController.checkActiveHandles,
+    HealthCheckController.checkApi
+  )
 
   publicApiRouter.get('/health_check/redis', HealthCheckController.checkRedis)
   privateApiRouter.get('/health_check/redis', HealthCheckController.checkRedis)

services/web/config/settings.defaults.coffee

@@ -231,6 +231,9 @@ module.exports = settings =
 	
 	maxUploadSize: 50 * 1024 * 1024 # 50 MB
 
+	# start failing the health check if active handles exceeds this limit
+	maxActiveHandles: if process.env['MAX_ACTIVE_HANDLES'] then parseInt(process.env['MAX_ACTIVE_HANDLES'], 10)
+
 	# Security
 	# --------
 	security: