Use Google Breakpad in ydbd (#12994)

Author: iddqdex

ydb/apps/ydbd/ya.make

@@ -69,7 +69,8 @@ PEERDIR(
     yql/essentials/udfs/common/url_base
     yql/essentials/udfs/common/yson2
     yql/essentials/udfs/logs/dsv
-    ydb/public/sdk/cpp/src/client/persqueue_public/codecs
+    ydb/library/breakpad
+    ydb/public/sdk/cpp/client/ydb_persqueue_public/codecs
 )
 
 YQL_LAST_ABI_VERSION()

ydb/deploy/docker/Dockerfile

@@ -24,30 +24,20 @@ COPY --chmod=0644 /liblibaio-dynamic.so /lib/liblibaio-dynamic.so
 ###
 # Base image with google brekpad assets
 ###
-FROM ${BREAKPAD_INIT_IMAGE}:${BREAKPAD_INIT_IMAGE_TAG} AS breakpad_init
-
-
-FROM base AS breakpad-setuid
-COPY --from=breakpad_init /usr/lib/libbreakpad_init.so /usr/lib/libbreakpad_init.so
-# workaround for old docker versions
-# https://github.com/moby/buildkit/issues/3920
-RUN /usr/bin/chmod 4644 /usr/lib/libbreakpad_init.so
 
+FROM ${BREAKPAD_INIT_IMAGE}:${BREAKPAD_INIT_IMAGE_TAG} AS breakpad_init
 FROM base AS base-breakpad
 RUN \
     apt-get -yqq update && \
     apt-get -yqq install --no-install-recommends binutils gdb strace linux-tools-generic && \
     apt-get clean && rm -rf /var/lib/apt/lists/*
-ENV LD_PRELOAD=libbreakpad_init.so
 ENV BREAKPAD_MINIDUMPS_PATH=/opt/ydb/volumes/coredumps
 ENV BREAKPAD_MINIDUMPS_SCRIPT=/opt/ydb/bin/minidump_script.py
 # breakpad binaries
 COPY --chmod=0755 --from=breakpad_init /usr/bin/minidump_stackwalk /usr/bin/minidump_stackwalk
 COPY --chmod=0755 --from=breakpad_init /usr/bin/minidump-2-core /usr/bin/minidump-2-core
 # minidump callback script
 COPY --chmod=0755 --chown=ydb /minidump_script.py /opt/ydb/bin/minidump_script.py
-# minidump init library
-COPY --link --from=breakpad-setuid /usr/lib/libbreakpad_init.so /usr/lib/libbreakpad_init.so
 
 FROM base AS ydbd-setcap
 COPY --chmod=0755 --chown=ydb /ydbd /opt/ydb/bin/ydbd

ydb/library/breakpad/minidumps.cpp

@@ -0,0 +1,36 @@
+#include <util/generic/ptr.h>
+#include <contrib/libs/breakpad/src/client/linux/handler/exception_handler.h>
+#include <unistd.h>
+#include <sys/wait.h>
+
+
+class TMinidumper {
+public:
+    TMinidumper() {
+        if(const char* path = getenv("BREAKPAD_MINIDUMPS_PATH")) {
+            using namespace google_breakpad;
+            Handler = MakeHolder<ExceptionHandler>(MinidumpDescriptor(path), nullptr, DumpCallback, nullptr, true, -1, true);
+        }
+    }
+
+private:
+    static bool DumpCallback(const google_breakpad::MinidumpDescriptor& descriptor, void* /*context*/, bool succeeded) {
+        if (char* script = getenv("BREAKPAD_MINIDUMPS_SCRIPT")) {
+            if (auto pid = fork()) {
+                waitpid(pid, 0, 0);
+            } else {
+                char* dumpSucceded = succeeded ? (char *)"true" : (char *)"false";  
+                char* descriptorPath = succeeded ? (char *)descriptor.path() : (char *)"\0";  
+                char* cmd[] = {script, dumpSucceded, descriptorPath, NULL};  
+                if (execve(cmd[0], &cmd[0], NULL)) {
+                    fprintf(stderr, "Error while process BREAKPAD MINIDUMPS SCRIPT: execve(%s '%s' '%s' '%s'): %s\n", cmd[0], cmd[1], cmd[2], strerror(errno));
+                }
+            }
+        }
+        return succeeded;
+    }
+
+    THolder<google_breakpad::ExceptionHandler> Handler;
+};
+
+TMinidumper Minidumper;

ydb/library/breakpad/ya.make

@@ -0,0 +1,14 @@
+LIBRARY()
+
+IF (OS_LINUX)
+    PEERDIR(
+        contrib/libs/breakpad/src
+        contrib/libs/breakpad/src/client/linux
+    )
+
+    SRCS(
+        GLOBAL minidumps.cpp
+    )
+ENDIF()
+
+END()

ydb/library/ya.make

@@ -7,6 +7,7 @@ RECURSE(
     arrow_parquet
     backup
     benchmarks
+    breakpad
     chunks_limiter
     folder_service
     formats

ydb/tests/functional/minidumps/test_break.py

@@ -0,0 +1,48 @@
+import yatest.common
+import os
+from ydb.tests.library.harness.kikimr_runner import KiKiMR
+
+
+def test_create_minidump():
+    dump_path = os.path.join(yatest.common.tempfile.gettempdir(), 'dumps1')
+    os.makedirs(dump_path, exist_ok=True)
+    os.environ['BREAKPAD_MINIDUMPS_PATH'] = dump_path
+    cluster = KiKiMR()
+    cluster.start()
+    for node in cluster.nodes.values():
+        node.send_signal(6)
+    try:
+        cluster.stop()
+    except RuntimeError:
+        pass
+    assert len(os.listdir(dump_path)) == len(cluster.nodes)
+
+
+def test_minidump_script():
+    temp_dir = yatest.common.tempfile.gettempdir()
+    dump_path = os.path.join(temp_dir, 'dumps2')
+    script_path = os.path.join(temp_dir, 'minidump_script.sh')
+    os.makedirs(dump_path, exist_ok=True)
+    with open(script_path, 'w') as f:
+        f.write(
+            '#!/usr/bin/bash\n'
+            'SUCCESS=$1\n'
+            'PATH=$2\n'
+            'echo $SUCCESS >${PATH}.success\n'
+        )
+    os.chmod(script_path, 0o777)
+    os.environ['BREAKPAD_MINIDUMPS_SCRIPT'] = script_path
+    os.environ['BREAKPAD_MINIDUMPS_PATH'] = dump_path
+    cluster = KiKiMR()
+    cluster.start()
+    for node in cluster.nodes.values():
+        node.send_signal(6)
+    try:
+        cluster.stop()
+    except RuntimeError:
+        pass
+    files = os.listdir(dump_path)
+    dmps = list(filter(lambda x: x.endswith('.dmp'), files))
+    successes = list(filter(lambda x: x.endswith('.dmp.success'), files))
+    assert len(dmps) == len(cluster.nodes)
+    assert len(successes) == len(cluster.nodes)

ydb/tests/functional/minidumps/ya.make

@@ -0,0 +1,24 @@
+IF (OS_LINUX)
+
+PY3TEST()
+
+TEST_SRCS(
+    test_break.py
+)
+
+SIZE(MEDIUM)
+
+ENV(YDB_DRIVER_BINARY="ydb/apps/ydbd/ydbd")
+
+PEERDIR(
+    ydb/tests/library
+)
+
+DEPENDS(
+    ydb/apps/ydbd
+)
+
+
+END()
+
+ENDIF()

ydb/tests/functional/ya.make

@@ -15,6 +15,7 @@ RECURSE(
     kqp
     large_serializable
     limits
+    minidumps
     postgresql
     query_cache
     rename