Merge pull request #92 from overleaf/bg-flush-queue-prod-fixes

add continuous background flush

services/document-updater/app.coffee

@@ -14,6 +14,7 @@ if Settings.sentry?.dsn?
 
 RedisManager = require('./app/js/RedisManager')
 DispatchManager = require('./app/js/DispatchManager')
+DeleteQueueManager = require('./app/js/DeleteQueueManager')
 Errors = require "./app/js/Errors"
 HttpController = require "./app/js/HttpController"
 mongojs = require "./app/js/mongojs"
@@ -142,7 +143,12 @@ host = Settings.internal.documentupdater.host or "localhost"
 if !module.parent # Called directly
 	app.listen port, host, ->
 		logger.info "Document-updater starting up, listening on #{host}:#{port}"
+		if Settings.continuousBackgroundFlush
+			logger.info "Starting continuous background flush"
+			DeleteQueueManager.startBackgroundFlush()
+
 module.exports = app
 
 for signal in ['SIGINT', 'SIGHUP', 'SIGQUIT', 'SIGUSR1', 'SIGUSR2', 'SIGTERM', 'SIGABRT']
 	process.on signal, shutdownCleanly(signal)
+

services/document-updater/app/coffee/DeleteQueueManager.coffee

@@ -1,3 +1,4 @@
+Settings = require('settings-sharelatex')
 RedisManager = require "./RedisManager"
 ProjectManager = require "./ProjectManager"
 logger = require "logger-sharelatex"
@@ -22,7 +23,7 @@ async = require "async"
 module.exports = DeleteQueueManager =
     flushAndDeleteOldProjects: (options, callback) ->
         startTime = Date.now()
-        cutoffTime = startTime - options.min_delete_age 
+        cutoffTime = startTime - options.min_delete_age + 100 * (Math.random() - 0.5)
         count = 0
 
         flushProjectIfNotModified = (project_id, flushTimestamp, cb) ->
@@ -60,4 +61,19 @@ module.exports = DeleteQueueManager =
                     count++ if flushed
                     flushNextProject()
 
-        flushNextProject()
+        flushNextProject()
+
+    startBackgroundFlush: () ->
+        SHORT_DELAY = 10
+        LONG_DELAY = 1000
+        doFlush = () ->
+            if Settings.shuttingDown
+                logger.warn "discontinuing background flush due to shutdown"
+                return
+            DeleteQueueManager.flushAndDeleteOldProjects {
+                timeout:1000,
+                min_delete_age:3*60*1000,
+                limit:1000 # high value, to ensure we always flush enough projects
+            }, (err, flushed) ->
+                setTimeout doFlush, (if flushed > 10 then SHORT_DELAY else LONG_DELAY)
+        doFlush()

services/document-updater/app/coffee/RedisManager.coffee

@@ -295,8 +295,9 @@ module.exports = RedisManager =
 		, callback
 
 	queueFlushAndDeleteProject: (project_id, callback) ->
-		# store the project id in a sorted set ordered by time
-		rclient.zadd keys.flushAndDeleteQueue(), Date.now(), project_id, callback
+		# store the project id in a sorted set ordered by time with a random offset to smooth out spikes
+		SMOOTHING_OFFSET = if Settings.smoothingOffset > 0 then Math.round(Settings.smoothingOffset * Math.random()) else 0
+		rclient.zadd keys.flushAndDeleteQueue(), Date.now() + SMOOTHING_OFFSET, project_id, callback
 
 	getNextProjectToFlushAndDelete: (cutoffTime, callback = (error, key, timestamp)->) ->
 		# find the oldest queued flush that is before the cutoff time

services/document-updater/config/settings.defaults.coffee

@@ -95,3 +95,7 @@ module.exports =
 		dsn: process.env.SENTRY_DSN
 
 	publishOnIndividualChannels: process.env['PUBLISH_ON_INDIVIDUAL_CHANNELS'] or false
+
+	continuousBackgroundFlush: process.env['CONTINUOUS_BACKGROUND_FLUSH'] or false
+
+	smoothingOffset: process.env['SMOOTHING_OFFSET'] or 1000 # milliseconds

services/document-updater/test/acceptance/coffee/DeletingAProjectTests.coffee

@@ -147,9 +147,8 @@ describe "Deleting a project", ->
 						@statusCode = res.statusCode
 						# after deleting the project and putting it in the queue, flush the queue
 						setTimeout () ->
-							DocUpdaterClient.flushOldProjects (error, res, body) =>
-							setTimeout done, 1000 # allow time for the flush to complete
-						, 100
+							DocUpdaterClient.flushOldProjects done
+						, 2000
 				, 200
 
 		after ->