Linux CI: Retry failed XOpenDisplay calls and only log errors in coverage run (#1368)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Bart Louwers <bart@emeel.net>
Co-authored-by: Bart Louwers <bart.louwers@gmail.com>

.github/workflows/linux-ci.yml

@@ -72,6 +72,8 @@ jobs:
             libwebp-dev
 
       - name: Build MapLibre Native Core
+        env:
+          CI: 1
         run: |
           cmake --version
           cmake -B build -GNinja -DCMAKE_BUILD_TYPE=Debug -DMLN_WITH_CLANG_TIDY=ON -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DMLN_WITH_COVERAGE=ON
@@ -182,7 +184,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install -y \
+          DEBIAN_FRONTEND=noninteractive sudo apt-get install -y \
             libcurl4-openssl-dev \
             libuv1-dev \
             libjpeg-dev \
@@ -190,8 +192,8 @@ jobs:
             libwebp-dev \
             libglfw3-dev \
             libsqlite3-dev \
-            xserver-xorg-video-dummy \
-            xinit
+            xvfb \
+            x11-xserver-utils
 
       - name: Cache Bazel
         uses: actions/cache@v3
@@ -208,10 +210,14 @@ jobs:
 
       - name: Generate coverage report
         run: |
-          bazel coverage --test_output=all --combined_report=lcov --nobuild_runfile_links --local_test_jobs=1 --instrumentation_filter="//:mbgl-core" //test:core //render-test:render-test --//:maplibre_platform=linux --run_under=//platform/linux:startxwrapper
-          echo coverage_report="$(bazel info output_path)"/_coverage/_coverage_report.dat >> "$GITHUB_ENV"
+          xvfb-run -a \
+            bazel coverage --combined_report=lcov  --instrumentation_filter="//:mbgl-core" --//:maplibre_platform=linux \
+            --test_output=errors --local_test_jobs=1 \
+            --test_env=DISPLAY --test_env=XAUTHORITY  --define=CI_BUILD=1 \
+            //test:core
 
       - name: Upload coverage report
+        if: '!cancelled()'
         uses: codecov/codecov-action@v3
         with:
           files: ${{ env.coverage_report }}

platform/linux/BUILD.bazel

@@ -10,13 +10,13 @@ cc_library(
     name = "impl",
     srcs = [
         "src/gl_functions.cpp",
-        "src/headless_backend_egl.cpp",
+        "src/headless_backend_glx.cpp",
     ],
     copts = CPP_FLAGS + MAPLIBRE_FLAGS,
     linkopts = [
         "-lGL",
         "-lglfw",
-        "-lEGL",
+        "-lX11",
         "-lsqlite3",
         "-luv",
         "-lz",

platform/linux/linux.cmake

@@ -93,6 +93,15 @@ else()
     )
 endif()
 
+if (DEFINED ENV{CI})
+    message("Building for CI")
+    target_compile_definitions(
+        mbgl-core
+        PRIVATE
+            CI_BUILD=1
+    )
+endif()
+
 # FIXME: Should not be needed, but now needed by node because of the headless frontend.
 target_include_directories(
     mbgl-core
@@ -153,9 +162,19 @@ add_executable(
 
 target_compile_definitions(
     mbgl-test-runner
-    PRIVATE WORK_DIRECTORY=${PROJECT_SOURCE_DIR}
+    PRIVATE
+        WORK_DIRECTORY=${PROJECT_SOURCE_DIR}
 )
 
+if (DEFINED ENV{CI})
+    message("Building for CI")
+    target_compile_definitions(
+        mbgl-test-runner
+        PRIVATE
+            CI_BUILD=1
+    )
+endif()
+
 target_link_libraries(
     mbgl-test-runner
     PRIVATE

platform/linux/src/headless_backend_glx.cpp

@@ -4,6 +4,11 @@
 
 #include <cassert>
 
+#ifdef CI_BUILD
+#include <chrono>
+#include <thread>
+#endif
+
 #include <GL/glx.h>
 
 namespace mbgl {
@@ -24,7 +29,20 @@ public:
             throw std::runtime_error("Failed to XInitThreads.");
         }
 
-        xDisplay = XOpenDisplay(nullptr);
+#ifdef CI_BUILD
+        // XOpenDisplay has proven very unreliable on CI, perform some retries if it fails
+        constexpr auto CI_RETRIES = 10;
+        for (int i = 0; i < CI_RETRIES; i++) {
+#endif
+            xDisplay = XOpenDisplay(nullptr);
+#ifdef CI_BUILD
+            if (xDisplay != nullptr) {
+                break;
+            }
+            Log::Debug(Event::OpenGL, "[CI] Failed to open X display, retrying...");
+            std::this_thread::sleep_for(std::chrono::milliseconds(500));
+        }
+#endif
         if (xDisplay == nullptr) {
             throw std::runtime_error("Failed to open X display.");
         }

test/BUILD.bazel

@@ -42,7 +42,7 @@ cc_test(
         ],
     ),
     args = [
-        "--gtest_filter=-Map.StyleNetworkErrorRetry:Map.StyleNotFound",
+        "--gtest_filter=-Map.StyleNetworkErrorRetry:Map.StyleNotFound:MainResourceLoader.CachedResourceLowPriority",
     ],
     copts = CPP_FLAGS + MAPLIBRE_FLAGS,
     data = glob(["fixtures/**/*"]) + [

test/storage/main_resource_loader.test.cpp

@@ -715,9 +715,8 @@ TEST(MainResourceLoader, TEST_REQUIRES_SERVER(RespondToStaleMustRevalidate)) {
 }
 
 // Test that requests for expired resources have lower priority than requests for new resources
+// Flaky https://github.com/maplibre/maplibre-native/issues/1071
 TEST(MainResourceLoader, TEST_REQUIRES_SERVER(CachedResourceLowPriority)) {
-    GTEST_SKIP() << "Skipping MainResourceLoader.CachedResourceLowPriority. "
-                 << "See https://github.com/maplibre/maplibre-native/issues/1071";
     util::RunLoop loop;
     MainResourceLoader fs(ResourceOptions{}, ClientOptions{});