Merge pull request #9223 from emqx/av/parallelCTruns

ci(test): test parallel CT runs on self-hosted runners
2022-11-04 12:11:54 +01:00 · 2022-11-04 12:11:54 +01:00 · 7dfa83f3c6
parent 57cc0366e0 b7a04c50ff
commit 7dfa83f3c6
12 changed files with 426 additions and 195 deletions
--- a/.github/workflows/run_test_cases.yaml
+++ b/.github/workflows/run_test_cases.yaml
@ -10,26 +10,103 @@ on:
  pull_request:

 jobs:
-    run_proper_test:
+  prepare:
    runs-on: ubuntu-20.04
    container: emqx/build-env:erl23.3.4.9-3-ubuntu20.04
-
+    outputs:
+      ct_apps: ${{ steps.run_find_apps.outputs.ct_apps }}
    steps:
-        - uses: actions/checkout@v2
-        - name: set git credentials
+      - uses: actions/checkout@v3
+        with:
+          path: source
+          fetch-depth: 0
+      - name: git credentials
        run: |
          if make emqx-ee --dry-run > /dev/null 2>&1; then
            echo "https://ci%40emqx.io:${{ secrets.CI_GIT_TOKEN }}@github.com" > $HOME/.git-credentials
            git config --global credential.helper store
          fi
-        - name: proper
-          run: make proper
+      - name: find_ct_apps
+        working-directory: source
+        id: run_find_apps
+        # emqx_plugin_libs doesn't have a test suite -> excluded from app list
+        # emqx ct is run independently -> exclude it from the app list
+        run: |
+          ct_apps="$(./scripts/find-apps.sh --json | jq -c 'del (.[] | select (. == "apps/emqx_plugin_libs" or . ==  "emqx"))')"
+          echo "ct-apps: $ct_apps"
+          echo "ct_apps=$ct_apps" >> $GITHUB_OUTPUT
+      - name: get_all_deps
+        working-directory: source
+        run: |
+          make deps-all
+          ./rebar3 as test compile
+          cd ..
+          zip -ryq source.zip source/* source/.[^.]*
+      - uses: actions/upload-artifact@v3
+        with:
+          name: source
+          path: source.zip

-    run_common_test:
-        runs-on: ${{ matrix.runs-on }}
+  eunit_and_proper:
+    needs: prepare
+    runs-on: ubuntu-20.04
+    container: emqx/build-env:erl23.3.4.9-3-ubuntu20.04
    strategy:
      fail-fast: false
      matrix:
+        task:
+          - eunit
+          - proper
+    steps:
+      - uses: AutoModality/action-clean@v1
+      - uses: actions/download-artifact@v3
+        with:
+          name: source
+          path: .
+      - name: unzip source code
+        run: unzip -o -q source.zip
+      # produces eunit.coverdata and proper.coverdata
+      - name: eunit and proper
+        working-directory: source
+        run: make ${{ matrix.task }}
+      - uses: actions/upload-artifact@v3
+        with:
+          name: cover
+          path: source/_build/test/cover
+          if-no-files-found: warn
+
+  emqx_ct:
+    needs: prepare
+    runs-on: ubuntu-20.04
+    container: emqx/build-env:erl23.3.4.9-3-ubuntu20.04
+    strategy:
+      fail-fast: false
+    steps:
+      - uses: AutoModality/action-clean@v1
+      - uses: actions/download-artifact@v3
+        with:
+          name: source
+          path: .
+      - name: unzip source code
+        run: unzip -o -q source.zip
+      # produces emqx-emqx.coverdata
+      - name: emqx-ct-pipeline
+        working-directory: source
+        run: make emqx-ct-pipeline
+      - uses: actions/upload-artifact@v3
+        with:
+          name: cover
+          path: source/_build/test/cover
+          if-no-files-found: warn
+
+  ct:
+    needs: prepare
+    runs-on: ${{ matrix.runs-on }}
+    strategy:
+      max-parallel: 12
+      fail-fast: false
+      matrix:
+        app_name: ${{ fromJson(needs.prepare.outputs.ct_apps) }}
        runs-on:
          - aws-amd64
          - ubuntu-20.04
@ -41,13 +118,20 @@ jobs:
          - runs-on: aws-amd64
            use-self-hosted: false
    steps:
-        - uses: actions/checkout@v2
+      - uses: AutoModality/action-clean@v1
+      - uses: actions/download-artifact@v3
+        with:
+          name: source
+          path: .
+      - name: unzip source code
+        run: unzip -q source.zip
        # to avoid dirty self-hosted runners
      - name: stop containers
        run: |
          docker rm -f $(docker ps -qa) || true
          docker network rm $(docker network ls -q) || true
      - name: docker compose up
+        working-directory: source
        if: endsWith(github.repository, 'emqx')
        env:
          MYSQL_TAG: 8
@ -66,7 +150,9 @@ jobs:
              -f .ci/docker-compose-file/docker-compose-pgsql-tcp.yaml \
              -f .ci/docker-compose-file/docker-compose-redis-single-tcp.yaml \
              up -d --build
+          docker exec -i erlang bash -c "git config --global --add safe.directory /emqx"
      - name: docker compose up
+        working-directory: source
        if: endsWith(github.repository, 'emqx-enterprise')
        env:
          MYSQL_TAG: 8
@ -111,33 +197,62 @@ jobs:
               != $(docker ps -a --filter name=client | wc -l) ]; do
            sleep 5
          done
-        - name: run eunit
-          run: |
-            docker exec -i erlang bash -c "make eunit"
      - name: run common test
-          run: |
-            docker exec -i erlang bash -c "make ct"
-        - name: run cover
-          run: |
-            printenv > .env
-            docker exec -i erlang bash -c "git config --global --add safe.directory /emqx"
-            docker exec -i erlang bash -c "make cover"
-            docker exec --env-file .env -i erlang bash -c "make coveralls"
+        run: docker exec -i erlang bash -c "make ${{ matrix.app_name }}-ct-pipeline"
      - name: cat rebar.crashdump
        if: failure()
+        working-directory: source
        run: if [ -f 'rebar3.crashdump' ];then cat 'rebar3.crashdump'; fi
-        - uses: actions/upload-artifact@v1
+      - name: set log file name
+        if: failure()
+        run: echo "LOGFILENAME=logs-$(echo ${{ matrix.app_name }} | tr '/' '_')" >> $GITHUB_ENV
+      - uses: actions/upload-artifact@v3
        if: failure()
        with:
-            name: logs
-            path: _build/test/logs
-        - uses: actions/upload-artifact@v1
+          name: ${{ env.LOGFILENAME }}
+          path: source/_build/test/logs
+          if-no-files-found: warn
+      - uses: actions/upload-artifact@v3
        with:
          name: cover
-            path: _build/test/cover
+          path: source/_build/test/cover
+          if-no-files-found: warn
+
+  make_cover:
+    needs:
+      - eunit_and_proper
+      - emqx_ct
+      - ct
+    runs-on: ubuntu-20.04
+    container: emqx/build-env:erl23.3.4.9-3-ubuntu20.04
+    steps:
+      - uses: AutoModality/action-clean@v1
+      - uses: actions/download-artifact@v3
+        with:
+          name: source
+          path: .
+      - name: unzip source code
+        run: unzip -q source.zip
+      - uses: actions/download-artifact@v3
+        name: download cover data
+        with:
+          name: cover
+          path: source/_build/test/cover
+      - name: make cover
+        working-directory: source
+        run: make cover
+      - name: send to coveralls
+        working-directory: source
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: make coveralls
+      - name: get coveralls logs
+        working-directory: source
+        if: failure()
+        run: cat rebar3.crashdump

  finish:
-      needs: run_common_test
+    needs: make_cover
    runs-on: ubuntu-20.04
    steps:
      - name: Coveralls Finished
--- a/8
+++ b/8
@ -61,6 +61,14 @@ $1-ct: $(REBAR)
 endef
 $(foreach app,$(APPS),$(eval $(call gen-app-ct-target,$(app))))

+## app/name-ct-pipeline targets are used in pipeline -> make cover data for each app
+.PHONY: $(APPS:%=%-ct-pipeline)
+define gen-app-ct-target-pipeline
+$1-ct-pipeline: $(REBAR)
+	$(REBAR) ct --name 'test@127.0.0.1' -c -v --cover_export_name $(PROFILE)-$(subst /,-,$1) --suite $(shell $(CURDIR)/scripts/find-suites.sh $1)
+endef
+$(foreach app,$(APPS),$(eval $(call gen-app-ct-target-pipeline,$(app))))
+
 ## apps/name-prop targets
 .PHONY: $(APPS:%=%-prop)
 define gen-app-prop-target
--- a/apps/emqx_management/test/emqx_bridge_mqtt_data_export_import_SUITE.erl
+++ b/apps/emqx_management/test/emqx_bridge_mqtt_data_export_import_SUITE.erl
@ -32,8 +32,11 @@ init_per_suite(Cfg) ->
    ok = emqx_dashboard_admin:mnesia(boot),
    application:load(emqx_modules),
    application:load(emqx_bridge_mqtt),
+    ekka_mnesia:start(),
+    emqx_dashboard_admin:mnesia(boot),
    emqx_ct_helpers:start_apps([emqx_rule_engine, emqx_management]),
    application:ensure_all_started(emqx_dashboard),
+    ok = emqx_rule_engine:load_providers(),
    Cfg.

 end_per_suite(Cfg) ->
--- a/apps/emqx_management/test/emqx_mgmt_bootstrap_app_SUITE.erl
+++ b/apps/emqx_management/test/emqx_mgmt_bootstrap_app_SUITE.erl
@ -41,6 +41,8 @@ init_per_suite(Config) ->
    Config.

 end_per_suite(_) ->
+    ok = application:unset_env(emqx_management, bootstrap_apps_file),
+    _ = mnesia:clear_table(mqtt_app),
    emqx_ct_helpers:stop_apps([]),
    ok.

--- a/lib-ce/emqx_modules/test/emqx_mod_sup_SUITE.erl
+++ b/lib-ce/emqx_modules/test/emqx_mod_sup_SUITE.erl
@ -23,6 +23,18 @@

 all() -> emqx_ct:all(?MODULE).

+init_per_suite(Config) ->
+    %% do not start the application
+    %% only testing the root supervisor in this suite
+    application:stop(emqx_modules),
+    {ok, Pid} = emqx_mod_sup:start_link(),
+    unlink(Pid),
+    Config.
+
+end_per_suite(_Config) ->
+    exit(whereis(emqx_mod_sup), kill),
+    ok.
+
 %%--------------------------------------------------------------------
 %% Test cases
 %%--------------------------------------------------------------------
--- a/scripts/find-apps.sh
+++ b/scripts/find-apps.sh
@ -5,6 +5,29 @@ set -euo pipefail
 # ensure dir
 cd -P -- "$(dirname -- "$0")/.."

+help() {
+    echo
+    echo "-h|--help:        To display this usage info"
+    echo "--json:           Print apps in json"
+}
+
+WANT_JSON='no'
+while [ "$#" -gt 0 ]; do
+    case $1 in
+        -h|--help)
+            help
+            exit 0
+            ;;
+        --json)
+            WANT_JSON='yes'
+            shift 1
+            ;;
+        *)
+            echo "unknown option $1"
+            exit 1
+            ;;
+    esac
+done
 if [ "$(./scripts/get-distro.sh)" = 'windows' ]; then
    # Otherwise windows may resolve to find.exe
    FIND="/usr/bin/find"
@ -17,17 +40,26 @@ find_app() {
    "$FIND" "${appdir}" -mindepth 1 -maxdepth 1 -type d
 }

-# append emqx application first
-echo 'emqx'
+EM="emqx"
+CE="$(find_app 'apps')"

-find_app 'apps'
 if [ -f 'EMQX_ENTERPRISE' ]; then
-    find_app 'lib-ee'
+    LIB="$(find_app 'lib-ee')"
 else
-    find_app 'lib-ce'
+    LIB="$(find_app 'lib-ce')"
 fi

 ## find directories in lib-extra
-find_app 'lib-extra'
+LIBE="$(find_app 'lib-extra')"
+
 ## find symlinks in lib-extra
-"$FIND" 'lib-extra' -mindepth 1 -maxdepth 1 -type l -exec test -e {} \; -print
+LIBES="$("$FIND" 'lib-extra' -mindepth 1 -maxdepth 1 -type l -exec test -e {} \; -print)"
+
+APPS_ALL="$(echo -e "${EM}\n${CE}\n${LIB}\n${LIBE}\n${LIBES}")"
+
+if [ "$WANT_JSON" = 'yes' ]; then
+    echo "${APPS_ALL}" | xargs | tr -d '\n' | jq -R -s -c 'split(" ")'
+else
+    echo "${APPS_ALL}"
+fi
+
--- a/test/emqx_channel_SUITE.erl
+++ b/test/emqx_channel_SUITE.erl
@ -822,6 +822,7 @@ t_enrich_connack_caps(_) ->
                          wildcard_subscription => true
                         }
                     end),
+    try
        AckProps = emqx_channel:enrich_connack_caps(#{}, channel()),
        ?assertMatch(#{'Retain-Available' := 1,
                    'Maximum-Packet-Size' := 1024,
@ -829,8 +830,10 @@ t_enrich_connack_caps(_) ->
                    'Wildcard-Subscription-Available' := 1,
                    'Subscription-Identifier-Available' := 1,
                    'Shared-Subscription-Available' := 1
-                  }, AckProps),
-    ok = meck:unload(emqx_mqtt_caps).
+                    }, AckProps)
+    after
+        ok = meck:unload(emqx_mqtt_caps)
+    end.

 %%--------------------------------------------------------------------
 %% Test cases for terminate
--- a/test/emqx_mqtt_caps_SUITE.erl
+++ b/test/emqx_mqtt_caps_SUITE.erl
@ -52,7 +52,7 @@ t_check_sub(_) ->
                wildcard_subscription => false
               },
    emqx_zone:set_env(zone, '$mqtt_sub_caps', SubCaps),
-    timer:sleep(50),
+    try
        ClientInfo = #{zone => zone},
        ok = emqx_mqtt_caps:check_sub(ClientInfo, <<"topic">>, SubOpts),
        ?assertEqual({error, ?RC_TOPIC_FILTER_INVALID},
@ -60,5 +60,7 @@ t_check_sub(_) ->
        ?assertEqual({error, ?RC_WILDCARD_SUBSCRIPTIONS_NOT_SUPPORTED},
                    emqx_mqtt_caps:check_sub(ClientInfo, <<"+/#">>, SubOpts)),
        ?assertEqual({error, ?RC_SHARED_SUBSCRIPTIONS_NOT_SUPPORTED},
-                 emqx_mqtt_caps:check_sub(ClientInfo, <<"topic">>, SubOpts#{share => true})),
-    emqx_zone:unset_env(zone, '$mqtt_pub_caps').
+                    emqx_mqtt_caps:check_sub(ClientInfo, <<"topic">>, SubOpts#{share => true}))
+    after
+        emqx_zone:unset_env(zone, '$mqtt_pub_caps')
+    end.
--- a/test/emqx_mqtt_protocol_v5_SUITE.erl
+++ b/test/emqx_mqtt_protocol_v5_SUITE.erl
@ -468,8 +468,8 @@ t_connack_max_qos_allowed(_) ->

    %% max_qos_allowed = 0
    emqx_zone:set_env(external, max_qos_allowed, 0),
-    persistent_term:erase({emqx_zone, external, '$mqtt_caps'}),
-    persistent_term:erase({emqx_zone, external, '$mqtt_pub_caps'}),
+    emqx_zone:unset_env(external, '$mqtt_caps'),
+    emqx_zone:unset_env(external, '$mqtt_pub_caps'),

    {ok, Client1} = emqtt:start_link([{proto_ver, v5}]),
    {ok, Connack1} = emqtt:connect(Client1),
@ -496,8 +496,8 @@ t_connack_max_qos_allowed(_) ->

    %% max_qos_allowed = 1
    emqx_zone:set_env(external, max_qos_allowed, 1),
-    persistent_term:erase({emqx_zone, external, '$mqtt_caps'}),
-    persistent_term:erase({emqx_zone, external, '$mqtt_pub_caps'}),
+    emqx_zone:unset_env(external, '$mqtt_caps'),
+    emqx_zone:unset_env(external, '$mqtt_pub_caps'),

    {ok, Client3} = emqtt:start_link([{proto_ver, v5}]),
    {ok, Connack3} = emqtt:connect(Client3),
@ -524,8 +524,8 @@ t_connack_max_qos_allowed(_) ->

    %% max_qos_allowed = 2
    emqx_zone:set_env(external, max_qos_allowed, 2),
-    persistent_term:erase({emqx_zone, external, '$mqtt_caps'}),
-    persistent_term:erase({emqx_zone, external, '$mqtt_pub_caps'}),
+    emqx_zone:unset_env(external, '$mqtt_caps'),
+    emqx_zone:unset_env(external, '$mqtt_pub_caps'),

    {ok, Client5} = emqtt:start_link([{proto_ver, v5}]),
    {ok, Connack5} = emqtt:connect(Client5),
--- a/test/emqx_router_SUITE.erl
+++ b/test/emqx_router_SUITE.erl
@ -32,6 +32,7 @@ init_per_suite(Config) ->
    Config.

 end_per_suite(_Config) ->
+    emqx_ct_helpers:boot_modules(all),
    emqx_ct_helpers:stop_apps([]).

 init_per_testcase(_TestCase, Config) ->
--- a/test/emqx_shared_sub_SUITE.erl
+++ b/test/emqx_shared_sub_SUITE.erl
@ -51,7 +51,7 @@ init_per_suite(Config) ->
    PortDiscovery = application:get_env(gen_rpc, port_discovery),
    application:set_env(gen_rpc, port_discovery, stateless),
    application:ensure_all_started(gen_rpc),
-    %% ensure emqx_moduels' app modules are loaded
+    %% ensure emqx_modules app modules are loaded
    %% so the mnesia tables are created
    ok = load_app(emqx_modules),
    emqx_ct_helpers:start_apps([]),
--- a/test/emqx_takeover_SUITE.erl
+++ b/test/emqx_takeover_SUITE.erl
@ -32,25 +32,44 @@
 all() -> emqx_ct:all(?MODULE).

 init_per_suite(Config) ->
+    emqx_ct_helpers:boot_modules(all),
    emqx_ct_helpers:start_apps([]),
    Config.

 end_per_suite(_Config) ->
    emqx_ct_helpers:stop_apps([]),
    ok.
+
+init_per_testcase(Case, Config) ->
+    ?MODULE:Case({'init', Config}).
+
+end_per_testcase(Case, Config) ->
+    ?MODULE:Case({'end', Config}).
+
 %%--------------------------------------------------------------------
 %% Testcases

-t_takeover(_) ->
-    process_flag(trap_exit, true),
+t_takeover({init, Config}) when is_list(Config) ->
+    Config;
+t_takeover({'end', Config}) when is_list(Config) ->
+    ok;
+t_takeover(Config) when is_list(Config) ->
    AllMsgs = messages(?CNT),
    Pos = rand:uniform(?CNT),
-
-    ClientId = <<"clientid">>,
-    {ok, C1} = emqtt:start_link([{clientid, ClientId}, {clean_start, false}]),
-    {ok, _} = emqtt:connect(C1),
+    ClientId = random_clientid(),
+    ClientOpts = [{clientid, ClientId},
+                  {clean_start, false},
+                  {host, "127.0.0.1"},
+                  {port, 1883}
+                 ],
+    C1 =
+        with_retry(
+          fun() ->
+              {ok, C} = emqtt:start_link(ClientOpts),
+              {ok, _} = emqtt:connect(C),
+              C
+          end, 5),
    emqtt:subscribe(C1, <<"t">>, 1),
-
    spawn(fun() ->
            [begin
                emqx:publish(lists:nth(I, AllMsgs)),
@ -59,31 +78,65 @@ t_takeover(_) ->
          end),
    emqtt:pause(C1),
    timer:sleep(?CNT*10),
-
    load_meck(ClientId),
+    try
        spawn(fun() ->
                [begin
                    emqx:publish(lists:nth(I, AllMsgs)),
                    timer:sleep(rand:uniform(10))
                end || I <- lists:seq(Pos+1, ?CNT)]
            end),
-    {ok, C2} = emqtt:start_link([{clientid, ClientId}, {clean_start, false}]),
+        {ok, C2} = emqtt:start_link(ClientOpts),
+        %% C1 is going down, unlink it so the test can continue to run
+        _ = monitor(process, C1),
+        ?assert(erlang:is_process_alive(C1)),
+        unlink(C1),
        {ok, _} = emqtt:connect(C2),
-
+        receive
+            {'DOWN', _, process, C1, _} ->
+                ok
+        after 1000 ->
+                  ct:fail("timedout_waiting_for_old_connection_shutdown")
+        end,
        Received = all_received_publishs(),
        ct:pal("middle: ~p, received: ~p", [Pos, [P || {publish, #{payload := P}} <- Received]]),
        assert_messages_missed(AllMsgs, Received),
        assert_messages_order(AllMsgs, Received),
-
-    emqtt:disconnect(C2),
-    unload_meck(ClientId).
-
-t_takover_in_cluster(_) ->
-    todo.
+        kill_process(C2, fun emqtt:stop/1)
+    after
+        unload_meck(ClientId)
+    end.

 %%--------------------------------------------------------------------
 %% Helpers

+random_clientid() ->
+    iolist_to_binary(["clientid", "-", integer_to_list(erlang:system_time())]).
+
+kill_process(Pid, WithFun) ->
+    _ = unlink(Pid),
+    _ = monitor(process, Pid),
+    try WithFun(Pid)
+    catch _:_ -> ok
+    end,
+    receive
+        {'DOWN', _, process, Pid, _} ->
+            ok
+    after 10_000 ->
+              exit(Pid, kill),
+              error(timeout)
+    end.
+
+with_retry(Fun, 1) -> Fun();
+with_retry(Fun, N) when N > 1 ->
+    try
+        Fun()
+    catch
+        _ : _ ->
+            ct:sleep(1000),
+            with_retry(Fun, N - 1)
+    end.
+
 load_meck(ClientId) ->
    meck:new(fake_conn_mod, [non_strict]),
    HookTakeover = fun(Pid, Msg = {takeover, 'begin'}) ->