feat(jobserver): Enable/disable jar caching on upload (spark-jobserver#950)

* feat(jobserver): Enable/disable jar caching on upload

Currently as soon as the jar is upload using /binaries
endpoint, it is cached on filesystem. On a high usage
SJS, disk is important and a trend has been seen where
users upload jars but don't start jobs. This change
gives the option to admin to cache jars on job start
only.

The property "spark.jobserver.cache-on-upload" is
configurable and can have the value true/false.

* Add "cache-on-upload" parameter to application.conf

* Remove ";" from application.conf

* Add cache-on-upload to test configurations

* Add test case

Check if jar can be saved and loaded without creating a cache

Author: ecandreev

job-server/src/main/resources/application.conf

@@ -28,6 +28,10 @@ spark {
     #   test = /path/to/my/test.jar
     # }
 
+    # Cache binary on upload.
+    # If set to false, the binary will be cached on job start only.
+    cache-on-upload = true
+
     filedao {
       rootdir = /tmp/spark-jobserver/filedao/data
     }

job-server/src/main/scala/spark/jobserver/io/JobSqlDAO.scala

@@ -137,8 +137,11 @@ class JobSqlDAO(config: Config) extends JobDAO with FileCacher {
                           binaryType: BinaryType,
                           uploadTime: DateTime,
                           binBytes: Array[Byte]) {
-    // The order is important. Save the jar file first and then log it into database.
-    cacheBinary(appName, binaryType, uploadTime, binBytes)
+    val cacheOnUploadEnabled = config.getBoolean("spark.jobserver.cache-on-upload")
+    if (cacheOnUploadEnabled) {
+      // The order is important. Save the jar file first and then log it into database.
+      cacheBinary(appName, binaryType, uploadTime, binBytes)
+    }
 
     // log it into database
     if (Await.result(insertBinaryInfo(

job-server/src/test/resources/local.test.jobsqldao.conf

@@ -3,6 +3,8 @@
 spark.jobserver {
   jobdao = spark.jobserver.io.JobSqlDAO
 
+  cache-on-upload = true
+
   cassandra {
     hosts = ["localhost:9142"]
     chunk-size-in-kb = 8

job-server/src/test/resources/local.test.jobsqldao_dbcp.conf

@@ -3,6 +3,8 @@
 spark.jobserver {
   jobdao = spark.jobserver.io.JobSqlDAO
 
+  cache-on-upload = true
+
   sqldao {
     rootdir = /tmp/spark-job-server-test/sqldao/data
     # https://coderwall.com/p/a2vnxg

job-server/src/test/scala/spark/jobserver/io/JobSqlDAOSpec.scala

@@ -122,6 +122,25 @@ class JobSqlDAOSpec extends JobSqlDAOSpecBase with TestJarFinder with FunSpecLik
       apps(jarInfo.appName) should equal ((BinaryType.Jar, jarInfo.uploadTime))
     }
 
+    it("should be able to save one jar and get it back without creating a cache") {
+      val configNoCache = config.withValue("spark.jobserver.cache-on-upload", ConfigValueFactory.fromAnyRef(false))
+      val daoNoCache = new JobSqlDAO(configNoCache)
+      // check the pre-condition
+      jarFile.exists() should equal (false)
+
+      // save
+      daoNoCache.saveBinary(jarInfo.appName, BinaryType.Jar, jarInfo.uploadTime, jarBytes)
+
+      // read it back
+      val apps: Map[String, (BinaryType, DateTime)] =
+        Await.result(daoNoCache.getApps, timeout).filter(_._2._1 == BinaryType.Jar)
+
+      // test
+      jarFile.exists() should equal (false)
+      apps.keySet should equal (Set(jarInfo.appName))
+      apps(jarInfo.appName) should equal ((BinaryType.Jar, jarInfo.uploadTime))
+    }
+
     it("should be able to retrieve the jar file") {
       // check the pre-condition
       jarFile.exists() should equal (false)
@@ -376,6 +395,7 @@ class JobSqlDAOSpec extends JobSqlDAOSpecBase with TestJarFinder with FunSpecLik
       apps.keys should not contain (jarInfo.appName)
     }
   }
+
 }
 
 class JobSqlDAODBCPSpec extends JobSqlDAOSpec {