fix: unit test failed and update rdsn (#384)

.travis.yml

@@ -20,7 +20,7 @@ cache:
     - apt
 
 before_install:
-    - wget https://media.githubusercontent.com/media/XiaoMi/pegasus-common/master/build-depends.tar.gz
+    - wget https://github.com/XiaoMi/pegasus-common/releases/download/deps/build-depends.tar.gz
     - tar xfz build-depends.tar.gz
     - rm -f build-depends.tar.gz
     - cd packages
@@ -35,8 +35,8 @@ install:
 
 before_script:
     - cd rdsn/thirdparty
-    - wget https://media.githubusercontent.com/media/XiaoMi/pegasus-common/master/pegasus-thirdparty-prebuild.tar.gz
-    - tar xfz pegasus-thirdparty-prebuild.tar.gz
+    - wget https://github.com/XiaoMi/pegasus-common/releases/download/deps/pegasus-thirdparty-prebuild.tar.gz
+    - tar xf pegasus-thirdparty-prebuild.tar.gz
     - rm -f pegasus-thirdparty-prebuild.tar.gz
     - cd ../..
     - ulimit -c unlimited -S

rdsn @ c3c13269

-Subproject commit fb3ed3bc5866bb3a7435e6f6fff31ff0dc5fa565
+Subproject commit c3c13269083506ec28a7d3b931261f51b6c5f741

run.sh

@@ -308,11 +308,13 @@ function usage_test()
     echo "   -h|--help         print the help info"
     echo "   -m|--modules      set the test modules: pegasus_unit_test pegasus_function_test"
     echo "   -k|--keep_onebox  whether keep the onebox after the test[default false]"
+    echo "   --on_travis       run tests on travis without some time-cosuming function tests"
 }
 function run_test()
 {
     local test_modules=""
     local clear_flags="1"
+    local on_traivs=""
     while [[ $# > 0 ]]; do
         key="$1"
         case $key in
@@ -327,6 +329,9 @@ function run_test()
             -k|--keep_onebox)
                 clear_flags=""
                 ;;
+            --on_travis)
+                on_travis="--on_travis"
+                ;;
             *)
                 echo "Error: unknow option \"$key\""
                 echo
@@ -349,7 +354,7 @@ function run_test()
 
     for module in `echo $test_modules`; do
         pushd $ROOT/src/builder/bin/$module
-        REPORT_DIR=$REPORT_DIR ./run.sh
+        REPORT_DIR=$REPORT_DIR ./run.sh $on_travis
         if [ $? != 0 ]; then
             echo "run test \"$module\" in `pwd` failed"
             exit 1

scripts/start_zk.sh

@@ -30,7 +30,7 @@ if [ ! -f zookeeper-3.4.6.tar.gz ]; then
     download_url="http://git.n.xiaomi.com/pegasus/packages/raw/master/zookeeper-3.4.6.tar.gz"
     wget -T 5 -t 1 $download_url
     if [ $? -ne 0 ]; then
-        download_url="https://github.com/xiaomi/pegasus-common/raw/master/zookeeper-3.4.6.tar.gz"
+        download_url="https://github.com/XiaoMi/pegasus-common/releases/download/deps/zookeeper-3.4.6.tar.gz"
         wget -T 5 -t 1 $download_url
         if [ $? -ne 0 ]; then
             echo "ERROR: download zookeeper failed"
@@ -41,7 +41,7 @@ fi
 
 if [ ! -d zookeeper-3.4.6 ]; then
     echo "Decompressing zookeeper..."
-    tar xfz zookeeper-3.4.6.tar.gz
+    tar xf zookeeper-3.4.6.tar.gz
     if [ $? -ne 0 ]; then
         echo "ERROR: decompress zookeeper failed"
         exit 1

scripts/travis.sh

@@ -16,7 +16,7 @@ if [ "$modified" ]; then
     exit 1
 fi
 
-"${root}"/run.sh build -c --skip_thirdparty --disable_gperf && ./run.sh test
+"${root}"/run.sh build -c --skip_thirdparty --disable_gperf && ./run.sh test --on_travis
 
 if [ $? ]; then
     echo "travis failed with exit code $?"

src/server/config.min.ini

@@ -6,7 +6,7 @@
   type = meta
   name = meta
   ports = @META_PORT@
-  pools = THREAD_POOL_DEFAULT,THREAD_POOL_META_SERVER,THREAD_POOL_META_STATE,THREAD_POOL_FD,THREAD_POOL_DLOCK
+  pools = THREAD_POOL_DEFAULT,THREAD_POOL_META_SERVER,THREAD_POOL_META_STATE,THREAD_POOL_FD,THREAD_POOL_DLOCK,THREAD_POOL_LOCAL_SERVICE
 
 [apps.replica]
   type = replica
@@ -34,7 +34,7 @@
   stderr_start_level = LOG_LEVEL_ERROR
 
 [threadpool..default]
-  worker_count = 1
+  worker_count = 4
   worker_priority = THREAD_xPRIORITY_NORMAL
   partitioned = false
 
@@ -87,6 +87,7 @@
   server_load_balancer_type = greedy_load_balancer
   min_live_node_count_for_unfreeze = 1
   cold_backup_disabled = false
+  recover_from_replica_server = false
 
 [replication]
   allow_non_idempotent_write = true

src/server/test/pegasus_write_service_test.cpp

@@ -80,6 +80,8 @@ public:
 
     void test_multi_remove()
     {
+        dsn::fail::setup();
+
         dsn::apps::multi_remove_request request;
         dsn::apps::multi_remove_response response;
 
@@ -123,6 +125,8 @@ public:
             ASSERT_EQ(err, 0);
             verify_response(response, 0, decree);
         }
+
+        dsn::fail::teardown();
     }
 
     void test_batched_writes()

src/shell/CMakeLists.txt

@@ -38,7 +38,7 @@ set(MY_PROJ_LIBS
 
 set(MY_BINPLACES "${CMAKE_CURRENT_SOURCE_DIR}/config.ini")
 
-set(MY_BOOST_LIBS Boost::system Boost::filesystem)
+set(MY_BOOST_LIBS Boost::system Boost::filesystem Boost::regex)
 
 SET(CMAKE_INSTALL_RPATH ".")
 SET(CMAKE_BUILD_WITH_INSTALL_RPATH TRUE)

src/test/function_test/CMakeLists.txt

@@ -25,7 +25,7 @@ else()
     set(MY_PROJ_LIBS ${MY_PROJ_LIBS} gtest)
 endif()
 
-set(MY_BOOST_LIBS Boost::system Boost::filesystem)
+set(MY_BOOST_LIBS Boost::system Boost::filesystem Boost::regex)
 
 set(MY_BINPLACES "config.ini" "run.sh")
 

src/test/function_test/run.sh

-#!/bin/sh
+#!/bin/bash
 
 exit_if_fail() {
     if [ $1 != 0 ]; then
@@ -11,6 +11,23 @@ if [ -z $REPORT_DIR ]; then
     REPORT_DIR="./"
 fi
 
+# If run function tests on traivs, we exclude some time-consuming tests
+# incluing restore test, recovery test
+on_travis="NO"
+while [ $# -gt 0 ]; do
+    key="$1"
+    case $key in
+        --on_travis)
+            on_travis="YES"
+            ;;
+        *)
+            echo "Error: unknow option \"$key\""
+            exit 1
+            ;;
+    esac
+    shift
+done
+
 test_case=pegasus_function_test
 config_file=config.ini
 table_name=temp
@@ -31,7 +48,9 @@ GTEST_OUTPUT="xml:$REPORT_DIR/slog_log.xml" GTEST_FILTER="lost_log.*" ./$test_ca
 exit_if_fail $? "run test slog_lost failed: $test_case $config_file $table_name"
 GTEST_OUTPUT="xml:$REPORT_DIR/recall.xml" GTEST_FILTER="drop_and_recall.*" ./$test_case $config_file $table_name
 exit_if_fail $? "run test recall failed: $test_case $config_file $table_name"
-GTEST_OUTPUT="xml:$REPORT_DIR/restore.xml" GTEST_FILTER="restore_test.*" ./$test_case $config_file $table_name
-exit_if_fail $? "run test restore_test failed: $test_case $config_file $table_name"
-GTEST_OUTPUT="xml:$REPORT_DIR/recovery.xml" GTEST_FILTER="recovery_test.*" ./$test_case $config_file $table_name
-exit_if_fail $? "run test recovery failed: $test_case $config_file $table_name"
+if [ $on_travis == "NO" ]; then
+    GTEST_OUTPUT="xml:$REPORT_DIR/restore.xml" GTEST_FILTER="restore_test.*" ./$test_case $config_file $table_name
+    exit_if_fail $? "run test restore_test failed: $test_case $config_file $table_name"
+    GTEST_OUTPUT="xml:$REPORT_DIR/recovery.xml" GTEST_FILTER="recovery_test.*" ./$test_case $config_file $table_name
+    exit_if_fail $? "run test recovery failed: $test_case $config_file $table_name"
+fi

src/test/function_test/test_recovery.cpp

@@ -28,18 +28,24 @@ class recovery_test : public testing::Test
 protected:
     virtual void SetUp()
     {
+        // THREAD_POOL_META_SERVER worker count should be greater than 1
+        // This function test update 'distributed_lock_service_type' to
+        // 'distributed_lock_service_simple', which executes in threadpool THREAD_POOL_META_SERVER
+        // As a result, failure detection lock executes in this pool
+        // if worker count = 1, it will lead to ERR_TIMEOUT when execute 'ddl_client->do_recovery'
+
         // 1. restart onebox, modify the config
         chdir(global_env::instance()._pegasus_root.c_str());
         system("./run.sh clear_onebox");
 
-        system("cp src/server/config-server.ini config-server-test-recovery.ini");
-        system("sed -i \"/^meta_state_service_type/c meta_state_service_type = "
+        system("cp src/server/config.min.ini config-server-test-recovery.ini");
+        system("sed -i \"/^\\s*meta_state_service_type/c meta_state_service_type = "
                "meta_state_service_simple\" config-server-test-recovery.ini");
-        system("sed -i \"/^distributed_lock_service_type/c distributed_lock_service_type = "
+        system("sed -i \"/^\\s*distributed_lock_service_type/c distributed_lock_service_type = "
                "distributed_lock_service_simple\" config-server-test-recovery.ini");
-        system("sed -i \"/^server_list/c server_list = @LOCAL_IP@:34601\" "
+        system("sed -i \"/^\\s*server_list/c server_list = @LOCAL_IP@:34601\" "
                "config-server-test-recovery.ini");
-        system("sed -i \"/^perf_counter_enable_logging/c perf_counter_enable_logging = false\" "
+        system("sed -i \"/^\\s*perf_counter_enable_logging/c perf_counter_enable_logging = false\" "
                "config-server-test-recovery.ini");
 
         system("./run.sh start_onebox -m 1 -r 3 --config_path config-server-test-recovery.ini");
@@ -158,11 +164,12 @@ public:
     void config_meta_to_do_cold_recovery()
     {
         char command[512];
-        snprintf(command,
-                 512,
-                 "cd %s && sed -i \"/^recover_from_replica_server/c recover_from_replica_server = "
-                 "true\" onebox/meta1/config.ini",
-                 global_env::instance()._pegasus_root.c_str());
+        snprintf(
+            command,
+            512,
+            "cd %s && sed -i \"/^\\s*recover_from_replica_server/c recover_from_replica_server = "
+            "true\" onebox/meta1/config.ini",
+            global_env::instance()._pegasus_root.c_str());
         system(command);
     }
 

src/test/function_test/test_restore.cpp

@@ -27,10 +27,10 @@ public:
 
         // modify the config to enable backup, and restart onebox
         system("./run.sh clear_onebox");
-        system("cp src/server/config-server.ini config-server-test-restore.ini");
-        system("sed -i \"/^cold_backup_disabled/c cold_backup_disabled = false\" "
+        system("cp src/server/config.min.ini config-server-test-restore.ini");
+        system("sed -i \"/^\\s*cold_backup_disabled/c cold_backup_disabled = false\" "
                "config-server-test-restore.ini");
-        std::string cmd = "sed -i \"/^cold_backup_root/c cold_backup_root = " + cluster_name;
+        std::string cmd = "sed -i \"/^\\s*cold_backup_root/c cold_backup_root = " + cluster_name;
         cmd = cmd + std::string("\" config-server-test-restore.ini");
         system(cmd.c_str());
         system("./run.sh start_onebox --config_path config-server-test-restore.ini");