Debug option to collect diagnostic information from PHP worker process (

elastic#1050) (elastic#1065)
intuibase · Sep 21, 2023 · b80bba4 · b80bba4
1 parent 1c6f09c
commit b80bba4
Show file tree

Hide file tree

Showing 20 changed files with 422 additions and 20 deletions.
diff --git a/agent/native/CMakeLists.txt b/agent/native/CMakeLists.txt
@@ -89,7 +89,7 @@ set(dependencies
     "zlib/1.3"
     "libcurl/8.0.1"
     "libunwind/1.6.2"
-#    "boost/1.82.0"
+    "boost/1.82.0"
     "gtest/1.13.0"
 )
 
@@ -104,7 +104,7 @@ endforeach()
 
 conan_cmake_run(REQUIRES ${dependencies}
                 OPTIONS Pkg/*:shared=False
-#                        boost:header_only=True
+                        boost:header_only=True
                         libcurl:shared=True
                         libcurl:with_libssh2=True
                 BUILD missing

diff --git a/agent/native/ext/ConfigManager.cpp b/agent/native/ext/ConfigManager.cpp
@@ -844,6 +844,8 @@ ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS( stringValue, transactionMaxSpans )
 ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS( stringValue, transactionSampleRate )
 ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS( stringValue, urlGroups )
 ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS( boolValue, verifyServerCert )
+ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS( stringValue, debugDiagnosticsFile )
+
 
 #undef ELASTIC_APM_DEFINE_FIELD_ACCESS_FUNCS
 #undef ELASTIC_APM_DEFINE_ENUM_FIELD_ACCESS_FUNCS
@@ -1225,6 +1227,12 @@ static void initOptionsMetadata( OptionMetadata* optsMeta )
             ELASTIC_APM_CFG_OPT_NAME_VERIFY_SERVER_CERT,
             /* defaultValue: */ true );
 
+    ELASTIC_APM_INIT_METADATA(
+            buildStringOptionMetadata,
+            debugDiagnosticsFile,
+            ELASTIC_APM_CFG_OPT_NAME_DEBUG_DIAGNOSTICS_FILE,
+            /* defaultValue: */ nullptr );
+
     ELASTIC_APM_ASSERT_EQ_UINT64( i, numberOfOptions );
 }
 

diff --git a/agent/native/ext/ConfigManager.h b/agent/native/ext/ConfigManager.h
@@ -117,6 +117,7 @@ enum OptionId
     optionId_transactionSampleRate,
     optionId_urlGroups,
     optionId_verifyServerCert,
+    optionId_debugDiagnosticsFile,
 
     numberOfOptions
 };
@@ -336,4 +337,7 @@ const ConfigSnapshot* getGlobalCurrentConfigSnapshot();
 #define ELASTIC_APM_CFG_OPT_NAME_URL_GROUPS "url_groups"
 #define ELASTIC_APM_CFG_OPT_NAME_VERIFY_SERVER_CERT "verify_server_cert"
 
+#define ELASTIC_APM_CFG_OPT_NAME_DEBUG_DIAGNOSTICS_FILE "debug_diagnostic_file"
+
+
 #define ELASTIC_APM_CFG_CONVERT_OPT_NAME_TO_INI_NAME( optNameStringLiteral ) ( "elastic_apm." optNameStringLiteral )
diff --git a/agent/native/ext/ConfigSnapshot.h b/agent/native/ext/ConfigSnapshot.h
@@ -88,4 +88,5 @@ struct ConfigSnapshot
     String transactionSampleRate = nullptr;
     String urlGroups = nullptr;
     bool verifyServerCert = false;
+    String debugDiagnosticsFile = nullptr;
 };
diff --git a/agent/native/ext/elastic_apm.cpp b/agent/native/ext/elastic_apm.cpp
@@ -194,6 +194,8 @@ PHP_INI_BEGIN()
     ELASTIC_APM_INI_ENTRY( ELASTIC_APM_CFG_OPT_NAME_TRANSACTION_SAMPLE_RATE )
     ELASTIC_APM_INI_ENTRY( ELASTIC_APM_CFG_OPT_NAME_URL_GROUPS )
     ELASTIC_APM_INI_ENTRY( ELASTIC_APM_CFG_OPT_NAME_VERIFY_SERVER_CERT )
+
+    ELASTIC_APM_INI_ENTRY( ELASTIC_APM_CFG_OPT_NAME_DEBUG_DIAGNOSTICS_FILE )
 PHP_INI_END()
 
 #undef ELASTIC_APM_INI_ENTRY_IMPL
@@ -273,7 +275,7 @@ static PHP_GINIT_FUNCTION(elastic_apm)
     });
 
     try {
-        elastic_apm_globals->globals = new elasticapm::php::AgentGlobals(std::move(phpBridge), {}, std::move(inferredSpans));
+        elastic_apm_globals->globals = new elasticapm::php::AgentGlobals(std::move(phpBridge), {}, std::move(inferredSpans), std::make_shared<elasticapm::php::SharedMemoryState>());
     } catch (std::exception const &e) {
         ELASTIC_APM_LOG_DIRECT_CRITICAL( "Unable to allocate AgentGlobals. '%s'", e.what());
     }

diff --git a/agent/native/ext/lifecycle.cpp b/agent/native/ext/lifecycle.cpp
@@ -44,6 +44,7 @@
 #include "AST_instrumentation.h"
 #include "Hooking.h"
 #include "CommonUtils.h"
+#include "Diagnostics.h"
 
 #define ELASTIC_APM_CURRENT_LOG_CATEGORY ELASTIC_APM_LOG_CATEGORY_LIFECYCLE
 
@@ -658,6 +659,22 @@ void elasticApmRequestInit()
 {
     requestCounter++;
 
+    Tracer* const tracer = getGlobalTracer();
+    const ConfigSnapshot* config = getTracerCurrentConfigSnapshot( tracer );
+
+    enableAccessToServerGlobal();
+    bool preloadDetected = requestCounter == 1 ? detectOpcachePreload() : false;
+
+    if (config && config->debugDiagnosticsFile && !preloadDetected && requestCounter <= 2) {
+        if (ELASTICAPM_G(globals)->sharedMemory_->shouldExecuteOneTimeTaskAmongWorkers()) {
+            try {
+                elasticapm::utils::storeDiagnosticInformation(elasticapm::utils::getParameterizedString(config->debugDiagnosticsFile), *(ELASTICAPM_G(globals)->bridge_));
+            } catch (std::exception const &e) {
+                ELASTIC_APM_LOG_WARNING( "Unable to write agent diagnostics: %s", e.what() );
+            }
+        }
+    }
+
     tracerPhpPartOnRequestInitSetInitialTracerState();
 
     TimePoint requestInitStartTime;
@@ -672,8 +689,6 @@ void elasticApmRequestInit()
     ELASTIC_APM_LOG_DEBUG_FUNCTION_ENTRY_MSG( "parent PID: %d", (int)(getParentProcessId()) );
 
     ResultCode resultCode;
-    Tracer* const tracer = getGlobalTracer();
-    const ConfigSnapshot* config = getTracerCurrentConfigSnapshot( tracer );
 
     if ( ! tracer->isInited )
     {
@@ -707,13 +722,10 @@ void elasticApmRequestInit()
 
     enableAccessToServerGlobal();
 
-    if (requestCounter == 1) {
-        bool preloadDetected = detectOpcachePreload();
-        if (preloadDetected) {
-            ELASTIC_APM_LOG_DEBUG( "opcache.preload request detected on init" );
-            resultCode = resultSuccess;
-            goto finally;
-        }
+    if (requestCounter == 1 && preloadDetected) {
+        ELASTIC_APM_LOG_DEBUG( "opcache.preload request detected on init" );
+        resultCode = resultSuccess;
+        goto finally;
     }
 
     if (!config->captureErrors) {

diff --git a/agent/native/libcommon/code/AgentGlobals.h b/agent/native/libcommon/code/AgentGlobals.h
@@ -3,21 +3,24 @@
 #include "InferredSpans.h"
 #include "PeriodicTaskExecutor.h"
 #include "PhpBridgeInterface.h"
+#include "SharedMemoryState.h"
 #include <memory>
 
 namespace elasticapm::php {
 
 class AgentGlobals {
 public:
-    AgentGlobals(std::shared_ptr<PhpBridgeInterface> bridge, std::unique_ptr<PeriodicTaskExecutor> periodicTaskExecutor, std::shared_ptr<InferredSpans> inferredSpans) :
+    AgentGlobals(std::shared_ptr<PhpBridgeInterface> bridge, std::unique_ptr<PeriodicTaskExecutor> periodicTaskExecutor, std::shared_ptr<InferredSpans> inferredSpans, std::shared_ptr<SharedMemoryState> sharedMemory) :
         bridge_(std::move(bridge)),
         periodicTaskExecutor_(std::move(periodicTaskExecutor)),
-        inferredSpans_(std::move(inferredSpans)) {
+        inferredSpans_(std::move(inferredSpans)),
+        sharedMemory_(std::move(sharedMemory)) {
     }
 
     std::shared_ptr<PhpBridgeInterface> bridge_;
     std::unique_ptr<PeriodicTaskExecutor> periodicTaskExecutor_;
     std::shared_ptr<InferredSpans> inferredSpans_;
+    std::shared_ptr<SharedMemoryState> sharedMemory_;
 };
 
 

diff --git a/agent/native/libcommon/code/CMakeLists.txt b/agent/native/libcommon/code/CMakeLists.txt
@@ -7,5 +7,9 @@ set (_Target  libcommon)
 add_library (${_Target}
     STATIC ${SrcFiles}
 )
-target_include_directories(${_Target} PUBLIC "./")
+
+
+target_include_directories(${_Target} PUBLIC "./"
+                                            "${CONAN_INCLUDE_DIRS_BOOST}"
+                                            )
 
diff --git a/agent/native/libcommon/code/CommonUtils.cpp b/agent/native/libcommon/code/CommonUtils.cpp
@@ -5,6 +5,8 @@
 #include <string_view>
 #include <signal.h>
 #include <stddef.h>
+#include <sys/types.h>
+#include <unistd.h>
 
 
 namespace elasticapm::utils {
@@ -53,5 +55,36 @@ std::chrono::milliseconds convertDurationWithUnit(std::string timeWithUnit) {
     throw std::invalid_argument("Invalid time unit.");
 }
 
+std::string getParameterizedString(std::string_view format) {
+
+    std::string out;
+
+    for (auto c = format.begin(); c < format.end(); ++c) {
+        if (*c == '%') {
+            c++;
+            if (c == format.end()) {
+                out.append(1, '%');
+                break;
+            }
+
+            switch (*c) {
+                case 'p':
+                    out.append(std::to_string(getpid()));
+                    break;
+                case 't':
+                    out.append(std::to_string(std::chrono::milliseconds(std::time(NULL)).count()));
+                    break;
+                default:
+                    out.append(1, '%');
+                    out.append(1, *c);
+                }
+        } else {
+            out.append({*c});
+        }
+    }
+
+    return out;
+}
+
 
 }
diff --git a/agent/native/libcommon/code/CommonUtils.h b/agent/native/libcommon/code/CommonUtils.h
@@ -10,4 +10,6 @@ namespace elasticapm::utils {
 
 std::chrono::milliseconds convertDurationWithUnit(std::string timeWithUnit); // default unit - ms, handles ms, s, m, throws std::invalid_argument if unit is unknown
 
+std::string getParameterizedString(std::string_view format);
+
 }
diff --git a/agent/native/libcommon/code/Diagnostics.cpp b/agent/native/libcommon/code/Diagnostics.cpp
@@ -0,0 +1,84 @@
+
+#include "Diagnostics.h"
+
+#include <sys/types.h>
+#include <unistd.h>
+
+#include <chrono>
+#include <fstream>
+#include <sstream>
+#include <iomanip>
+
+namespace elasticapm::utils {
+
+namespace detail {
+    static constexpr int separatorWidth = 60;
+    static constexpr char separator = '=';
+}
+
+static void getProcessDiags(std::ostream &out, std::string_view name) {
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+    out << "Process " << name << ":" << std::endl;
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+    try {
+        std::stringstream mapsname;
+        mapsname << "/proc/self/" << name;
+        std::ifstream maps;
+        maps.exceptions(std::ios_base::failbit);
+        maps.open(mapsname.str());
+        out << maps.rdbuf();
+        maps.close();
+    } catch (std::exception const &e) {
+        out << "Unable to get process " << name << ": " << e.what() << std::endl; 
+    }
+}
+
+static void getDiagnosticInformation(std::ostream &out, elasticapm::php::PhpBridgeInterface const &bridge) {
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+    out << "Elastic APM PHP agent diagnostics:" << std::endl;
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+    std::time_t time = std::time({});
+    char timeString[std::size("yyyy-mm-ddThh:mm:ssZ")];
+    std::strftime(std::data(timeString), std::size(timeString), "%FT%TZ", std::gmtime(&time));
+
+    out << "Time: " << timeString << " UTC (" << std::chrono::milliseconds(time).count() << ')' << std::endl;
+    out << "PID: " << getpid() << std::endl;
+    out << "PPID: " << getppid() << std::endl;
+    out << "UID: " << getuid() << std::endl;
+
+    auto extensions = bridge.getExtensionList();
+    if (extensions.size() > 0) {
+        out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+        out << "Loaded extensions:" << std::endl;
+        out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+
+        for (auto const &extension : extensions) {
+            out << extension.name << " " << extension.version << std::endl;
+        }
+    }
+
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl << std::endl;
+    out << "phpinfo() output:" << std::endl;
+    out << std::setfill(detail::separator) << std::setw(detail::separatorWidth) << detail::separator << std::endl;
+    out << bridge.getPhpInfo() << std::endl;
+
+    getProcessDiags(out, "maps");
+    getProcessDiags(out, "smaps_rollup");
+    getProcessDiags(out, "status");
+    getProcessDiags(out, "limits");
+    getProcessDiags(out, "cgroup");
+    getProcessDiags(out, "cmdline");
+    getProcessDiags(out, "mountinfo");
+    getProcessDiags(out, "mounts");
+    getProcessDiags(out, "mountstats");
+    getProcessDiags(out, "stat");
+}
+
+void storeDiagnosticInformation(std::string_view outputFileName, elasticapm::php::PhpBridgeInterface const &bridge) {
+    std::ofstream out;
+    out.exceptions(std::ios_base::failbit);
+    out.open(outputFileName.data());
+    getDiagnosticInformation(out, bridge);
+}
+
+}
diff --git a/agent/native/libcommon/code/Diagnostics.h b/agent/native/libcommon/code/Diagnostics.h
@@ -0,0 +1,10 @@
+
+#pragma once
+
+#include "PhpBridgeInterface.h"
+
+namespace elasticapm::utils {
+
+void storeDiagnosticInformation(std::string_view outputFileName, elasticapm::php::PhpBridgeInterface const &bridge); //throws
+
+}
diff --git a/agent/native/libcommon/code/PhpBridgeInterface.h b/agent/native/libcommon/code/PhpBridgeInterface.h
@@ -1,15 +1,27 @@
 #pragma once
 
 #include <chrono>
+#include <string>
+#include <vector>
 
 namespace elasticapm::php {
 
 
 class PhpBridgeInterface {
 public:
+
+    struct phpExtensionInfo_t {
+        std::string name;
+        std::string version;
+    };
+
     virtual ~PhpBridgeInterface() = default;
 
-    virtual bool callInferredSpans(std::chrono::milliseconds duration) = 0;
+    virtual bool callInferredSpans(std::chrono::milliseconds duration) const = 0;
+    virtual std::vector<phpExtensionInfo_t> getExtensionList() const = 0;
+    virtual std::string getPhpInfo() const = 0;
+
+
 };
 
 }
diff --git a/agent/native/libcommon/code/SharedMemoryState.h b/agent/native/libcommon/code/SharedMemoryState.h
@@ -0,0 +1,42 @@
+
+
+#include <boost/interprocess/anonymous_shared_memory.hpp>
+#include <boost/interprocess/mapped_region.hpp>
+#include <boost/interprocess/sync/interprocess_upgradable_mutex.hpp>
+#include <boost/interprocess/sync/scoped_lock.hpp>
+#include <boost/interprocess/sync/sharable_lock.hpp>
+
+namespace elasticapm::php {
+
+class SharedMemoryState {
+public:
+    struct SharedData {
+        boost::interprocess::interprocess_upgradable_mutex mutex;
+        bool oneTimeTaskAmongWorkersExecuted = false;
+    };
+
+    bool shouldExecuteOneTimeTaskAmongWorkers() {
+        {
+            boost::interprocess::sharable_lock< decltype( SharedData::mutex ) > lock( data_->mutex );
+            if ( data_->oneTimeTaskAmongWorkersExecuted )
+            {
+                return false;
+            }
+        }
+
+        boost::interprocess::scoped_lock< decltype( SharedData::mutex ) > ulock( data_->mutex );
+        if ( data_->oneTimeTaskAmongWorkersExecuted )
+        {
+            return false;
+        }
+        data_->oneTimeTaskAmongWorkersExecuted = true;
+        return true;
+    }
+
+
+protected:
+    boost::interprocess::mapped_region region_{ boost::interprocess::anonymous_shared_memory( sizeof( SharedData ) ) };
+    SharedData* data_{ new (region_.get_address()) SharedData };
+};
+
+}