path: root/cpp/src/IceGrid/PlatformInfo.cpp

// **********************************************************************
//
// Copyright (c) 2003-2008 ZeroC, Inc. All rights reserved.
//
// This copy of Ice is licensed to you under the terms described in the
// ICE_LICENSE file included in this distribution.
//
// **********************************************************************

#include <IceUtil/StringUtil.h>
#include <Ice/Communicator.h>
#include <Ice/Properties.h>
#include <Ice/LocalException.h>
#include <Ice/LoggerUtil.h>

#include <IceGrid/PlatformInfo.h>
#include <IceGrid/TraceLevels.h>

#include <IcePatch2/Util.h>
#include <climits>

#if defined(_WIN32)
#   include <direct.h> // For _getcwd
#   include <pdhmsg.h> // For PDH_MORE_DATA
#else
#   include <sys/utsname.h>
#   if defined(__APPLE__)
#      include <sys/sysctl.h>
#   elif defined(__sun)
#      include <sys/loadavg.h>
#   elif defined(__hpux)
#      include <sys/pstat.h>
#   elif defined(_AIX)
#      include <nlist.h>
#      include <fcntl.h>
#   endif
#endif

using namespace std;
using namespace IceGrid;

namespace
{

#ifdef _WIN32

string
pdhErrorToString(PDH_STATUS err)
{
    return IceUtilInternal::errorToString(err, GetModuleHandle(TEXT("PDH.DLL")));
}

static string
getLocalizedPerfName(int idx, const Ice::LoggerPtr& logger)
{
    vector<char> localized;
    unsigned long size = 256;
    localized.resize(size);
    PDH_STATUS err;
    while((err = PdhLookupPerfNameByIndex(0, idx, &localized[0], &size)) == PDH_MORE_DATA)
    {
        size += 256;
        localized.resize(size);
    }

    if(err != ERROR_SUCCESS)
    {
	Ice::Warning out(logger);
	out << "Unable to lookup the performance counter name:\n";
	out << pdhErrorToString(err);
	out << "\nThis usually occurs when you do not have sufficient privileges";
         
	throw Ice::SyscallException(__FILE__, __LINE__, err);
    }
    return string(&localized[0]);
}

class UpdateUtilizationAverageThread : public IceUtil::Thread
{
public:

    UpdateUtilizationAverageThread(PlatformInfo& platform) : _platform(platform)
    { 
    }

    virtual void
    run()
    {
	_platform.runUpdateLoadInfo();
    }

private:
    
    PlatformInfo& _platform;
};
#endif

}

namespace IceGrid
{

RegistryInfo
toRegistryInfo(const InternalReplicaInfoPtr& replica)
{
    RegistryInfo info;
    info.name = replica->name;
    info.hostname = replica->hostname;
    return info;
}

NodeInfo
toNodeInfo(const InternalNodeInfoPtr& node)
{
    NodeInfo info;
    info.name = node->name;
    info.os = node->os;
    info.hostname = node->hostname;
    info.release = node->release;
    info.version = node->version;
    info.machine = node->machine;
    info.nProcessors = node->nProcessors;
    info.dataDir = node->dataDir;
    return info;
}

}

PlatformInfo::PlatformInfo(const string& prefix, 
                           const Ice::CommunicatorPtr& communicator, 
                           const TraceLevelsPtr& traceLevels) : 
    _traceLevels(traceLevels)
{
    //
    // Initialization of the necessary data structures to get the load average.
    //
#if defined(_WIN32)
    _terminated = false;
    _usages1.insert(_usages1.end(), 1 * 60 / 5, 0); // 1 sample every 5 seconds during 1 minutes.
    _usages5.insert(_usages5.end(), 5 * 60 / 5, 0); // 1 sample every 5 seconds during 5 minutes.
    _usages15.insert(_usages15.end(), 15 * 60 / 5, 0); // 1 sample every 5 seconds during 15 minutes.
    _last1Total = 0;
    _last5Total = 0;
    _last15Total = 0;
#elif defined(_AIX)
    struct nlist nl;
    nl.n_name = "avenrun";
    nl.n_value = 0;
    if(knlist(&nl, 1, sizeof(nl)) == 0)
    {
        _kmem = open("/dev/kmem", O_RDONLY);

        //
        // Give up root permissions to minimize security risks, it's
        // only needed to access /dev/kmem.
        //
        setuid(getuid());
        setgid(getgid());
    }
    else
    {
        _kmem = -1;
    }
#endif

    //
    // Get the number of processors.
    //
#if defined(_WIN32)
    SYSTEM_INFO sysInfo;
    GetSystemInfo(&sysInfo);
    _nProcessors = sysInfo.dwNumberOfProcessors;
#elif defined(__APPLE__)
    static int ncpu[2] = { CTL_HW, HW_NCPU };
    size_t sz = sizeof(_nProcessors);
    if(sysctl(ncpu, 2, &_nProcessors, &sz, 0, 0) == -1)
    {
        Ice::SyscallException ex(__FILE__, __LINE__);
        ex.error = getSystemErrno();
        throw ex;
    }
#elif defined(__hpux)
    struct pst_dynamic dynInfo;
    if(pstat_getdynamic(&dynInfo, sizeof(dynInfo), 1, 0) >= 0)
    {
        _nProcessors = dynInfo.psd_proc_cnt;
    }
    else
    {
        _nProcessors = 1;
    }
#else
    _nProcessors = static_cast<int>(sysconf(_SC_NPROCESSORS_ONLN));
#endif

    //
    // Get the rest of the node information.
    //
#ifdef _WIN32
    _os = "Windows";
    char hostname[MAX_COMPUTERNAME_LENGTH + 1];
    unsigned long size = sizeof(hostname);
    if(GetComputerName(hostname, &size))
    {
        _hostname = hostname;
    }
    OSVERSIONINFO osInfo;
    osInfo.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
    GetVersionEx(&osInfo);
    ostringstream os;
    os << osInfo.dwMajorVersion << "." << osInfo.dwMinorVersion;
    _release = os.str();
    _version = osInfo.szCSDVersion;

    switch(sysInfo.wProcessorArchitecture)
    {
#if defined(_MSC_VER) && (_MSC_VER < 1300)
    case 9: // PROCESSOR_ARCHITECTURE_AMD64
#else
    case PROCESSOR_ARCHITECTURE_AMD64:
#endif
        _machine = "x64";
        break;
    case PROCESSOR_ARCHITECTURE_IA64:
        _machine = "IA64";
        break;
    case PROCESSOR_ARCHITECTURE_INTEL:
        _machine = "x86";
        break;
    default:
        _machine = "unknown";
        break;
    };
#else
    struct utsname utsinfo;
    uname(&utsinfo);
    _os = utsinfo.sysname;
    _hostname = utsinfo.nodename;
    _release = utsinfo.release;
    _version = utsinfo.version;
    _machine = utsinfo.machine;
#endif

    Ice::PropertiesPtr properties = communicator->getProperties();
    string endpointsPrefix;
    if(prefix == "IceGrid.Registry")
    {
        _name = properties->getPropertyWithDefault("IceGrid.Registry.ReplicaName", "Master");
        endpointsPrefix = prefix + ".Client";
    }
    else
    {
        _name = properties->getProperty(prefix + ".Name");
        endpointsPrefix = prefix;
    }

    Ice::PropertyDict props = properties->getPropertiesForPrefix(endpointsPrefix);
    Ice::PropertyDict::const_iterator p = props.find(endpointsPrefix + ".PublishedEndpoints");
    if(p != props.end())
    {
        _endpoints = p->second;
    }
    else
    {
        _endpoints = properties->getProperty(endpointsPrefix + ".Endpoints");
    }

#ifdef _WIN32
    char cwd[_MAX_PATH];
    if(_getcwd(cwd, _MAX_PATH) == NULL)
#else
    char cwd[PATH_MAX];
    if(getcwd(cwd, PATH_MAX) == NULL)
#endif
    {
        throw "cannot get the current directory:\n" + IceUtilInternal::lastErrorToString();
    }
    _cwd = string(cwd);

    _dataDir = properties->getProperty(prefix + ".Data");    
    if(!IcePatch2::isAbsolute(_dataDir))
    {
        _dataDir = _cwd + '/' + _dataDir;
    }
    if(_dataDir[_dataDir.length() - 1] == '/')
    {
        _dataDir = _dataDir.substr(0, _dataDir.length() - 1);
    }
}

PlatformInfo::~PlatformInfo()
{
#if defined(_AIX)
    if(_kmem > 0)
    {
        close(_kmem);
    }
#endif
}

void
PlatformInfo::start()
{
#if defined(_WIN32)
    _updateUtilizationThread = new UpdateUtilizationAverageThread(*this);
    _updateUtilizationThread->start();
#endif
}

void
PlatformInfo::stop()
{
#if defined(_WIN32)
    {
	IceUtil::Monitor<IceUtil::Mutex>::Lock sync(_utilizationMonitor);
	_terminated = true;
	_utilizationMonitor.notify();
    }

    assert(_updateUtilizationThread);
    _updateUtilizationThread->getThreadControl().join();
    _updateUtilizationThread = 0;
#endif
}

NodeInfo
PlatformInfo::getNodeInfo() const
{
    return toNodeInfo(getInternalNodeInfo());
}

RegistryInfo
PlatformInfo::getRegistryInfo() const
{
    return toRegistryInfo(getInternalReplicaInfo());
}

InternalNodeInfoPtr
PlatformInfo::getInternalNodeInfo() const
{
    InternalNodeInfoPtr info = new InternalNodeInfo();
    info->name = _name;
    info->os = _os;
    info->hostname = _hostname;
    info->release = _release;
    info->version = _version;
    info->machine = _machine;
    info->nProcessors = _nProcessors;
    info->dataDir = _dataDir;
    return info;
}

InternalReplicaInfoPtr
PlatformInfo::getInternalReplicaInfo() const
{
    InternalReplicaInfoPtr info = new InternalReplicaInfo();
    info->name = _name;
    info->hostname = _hostname;
    return info;
}

LoadInfo
PlatformInfo::getLoadInfo()
{
    LoadInfo info;
    info.avg1 = -1.0f;
    info.avg5 = -1.0f;
    info.avg15 = -1.0f;

#if defined(_WIN32)
    IceUtil::Monitor<IceUtil::Mutex>::Lock sync(_utilizationMonitor);
    info.avg1 = static_cast<float>(_last1Total) / _usages1.size() / 100.0f;
    info.avg5 = static_cast<float>(_last5Total) / _usages5.size() / 100.0f;
    info.avg15 = static_cast<float>(_last15Total) / _usages15.size() / 100.0f;
#elif defined(__sun) || defined(__linux) || defined(__APPLE__)
    //
    // We use the load average divided by the number of
    // processors to figure out if the machine is busy or
    // not. The result is capped at 1.0f.
    //
    double loadAvg[3];
    if(getloadavg(loadAvg, 3) != -1)
    {
        info.avg1 = static_cast<float>(loadAvg[0]);
        info.avg5 = static_cast<float>(loadAvg[1]);
        info.avg15 = static_cast<float>(loadAvg[2]);
    }
#elif defined(__hpux)
    struct pst_dynamic dynInfo;
    if(pstat_getdynamic(&dynInfo, sizeof(dynInfo), 1, 0) >= 0)
    {
        info.avg1 = dynInfo.psd_avg_1_min;
        info.avg5 = dynInfo.psd_avg_5_min;
        info.avg15 = dynInfo.psd_avg_15_min;
    }
#elif defined(_AIX)
    if(_kmem > 1)
    {
        long long avenrun[3];
        struct nlist nl;
        nl.n_name = "avenrun";
        nl.n_value = 0;
        if(knlist(&nl, 1, sizeof(nl)) == 0)
        {
            if(pread(_kmem, avenrun, sizeof(avenrun), nl.n_value) >= sizeof(avenrun))
            {
                info.avg1 = avenrun[0] / 65536.0f;
                info.avg5 = avenrun[1] / 65536.0f;
                info.avg15 = avenrun[2] / 65536.0f;
            }
        }
    }
#endif
    return info;
}

std::string
PlatformInfo::getHostname() const
{
    return _hostname;
}

std::string
PlatformInfo::getDataDir() const
{
    return _dataDir;
}

std::string
PlatformInfo::getCwd() const
{
    return _cwd;
}

#ifdef _WIN32
void
PlatformInfo::runUpdateLoadInfo()
{
    //
    // NOTE: We shouldn't initialize the performance counter from the
    // PlatformInfo constructor because it might be called when
    // IceGrid is started on boot as a Windows service with the
    // Windows service control manager (SCM) locked. The query
    // initialization would fail (hang) because it requires to start
    // the "WMI Windows Adapter" service (which can't be started
    // because the SCM is locked...).
    //

    //
    // Open the query.
    //
    HQUERY query;
    PDH_STATUS err = PdhOpenQuery(0, 0, &query);
    if(err != ERROR_SUCCESS)
    {
        Ice::Warning out(_traceLevels->logger);
        out << "Cannot open performance data query:\n" << pdhErrorToString(err);
        return;
    }

    //
    // Add the counter for \\Processor(_Total)\\"%Processor Time".
    //
    // We have to look up the localized names for these.  "Processor"
    // is index 238 and "%Processor Time" is index 6.
    //
    // If either lookup fails, close the query system, and we're done.
    //
    
    string processor;
    string percentProcessorTime;
    try
    {
        processor = getLocalizedPerfName(238, _traceLevels->logger);
        percentProcessorTime = getLocalizedPerfName(6, _traceLevels->logger);
    }
    catch(const Ice::LocalException&)
    {
	// No need to print a warning, it's taken care of by getLocalizedPerfName
        PdhCloseQuery(query);
        return;
    }

    const string name = "\\" + processor + "(_Total)\\" + percentProcessorTime;
    HCOUNTER _counter;
    err = PdhAddCounter(query, name.c_str(), 0, &_counter);
    if(err != ERROR_SUCCESS)
    {
        Ice::Warning out(_traceLevels->logger);
        out << "Cannot add performance counter `" + name + "' (expected ";
        out << "if you have insufficient privileges to monitor performance counters):\n";
        out << pdhErrorToString(err);
        PdhCloseQuery(query);
        return;
    }

    while(true)
    {
	IceUtil::Monitor<IceUtil::Mutex>::Lock sync(_utilizationMonitor);
	_utilizationMonitor.timedWait(IceUtil::Time::seconds(5)); // 5 seconds.
	if(_terminated)
	{
	    break;
	}

	int usage = 100;
        PDH_STATUS err = PdhCollectQueryData(query);
        if(err == ERROR_SUCCESS)
        {
            DWORD type;
            PDH_FMT_COUNTERVALUE value;
            PdhGetFormattedCounterValue(_counter, PDH_FMT_LONG, &type, &value);
            usage = static_cast<int>(value.longValue);
        }
        else
        {
            Ice::Warning out(_traceLevels->logger);
            out << "Could not collect performance counter data:\n" << pdhErrorToString(err);
        }
	
	_last1Total += usage - _usages1.back();
	_last5Total += usage - _usages5.back();
	_last15Total += usage - _usages15.back();
	
	_usages1.pop_back();
	_usages5.pop_back();
	_usages15.pop_back();
	_usages1.push_front(usage);
	_usages5.push_front(usage);
	_usages15.push_front(usage);
    }

    PdhCloseQuery(query);
}
#endif