pegasus/src/Pegasus/Common/Monitor.cpp - diff

Return to Monitor.cpp CVS log

Up to [Pegasus] / pegasus / src / Pegasus / Common

Diff for /pegasus/src/Pegasus/Common/Monitor.cpp between version 1.39 and 1.115.14.2

version 1.39, 2003/05/06 16:36:44

version 1.115.14.2, 2006/12/15 10:36:06

Line 1

//%/////////////////////////////////////////////////////////////////////////////

//%2006////////////////////////////////////////////////////////////////////////

// The Open Group, Tivoli Systems

// Company, L.P.; IBM Corp.; The Open Group; Tivoli Systems.

// IBM Corp.; EMC Corporation, The Open Group.

// IBM Corp.; EMC Corporation; VERITAS Software Corporation; The Open Group.

// EMC Corporation; VERITAS Software Corporation; The Open Group.

// EMC Corporation; Symantec Corporation; The Open Group.

// Permission is hereby granted, free of charge, to any person obtaining a copy

// of this software and associated documentation files (the "Software"), to

Line 21

Line 29

//==============================================================================

// Author: Mike Brasher (mbrasher@bmc.com)

// Modified By:

//%/////////////////////////////////////////////////////////////////////////////

#include "Network.h"

#include <Pegasus/Common/Config.h>

#include <cstring>

#include "Monitor.h"

Line 34

Line 40

#include "Socket.h"

#include <Pegasus/Common/Tracer.h>

#include <Pegasus/Common/HTTPConnection.h>

#include <Pegasus/Common/MessageQueueService.h>

#ifdef PEGASUS_OS_TYPE_WINDOWS

#include <Pegasus/Common/Exception.h>

# if defined(FD_SETSIZE) && FD_SETSIZE != 1024

#include "ArrayIterator.h"

# error "FD_SETSIZE was not set to 1024 prior to the last inclusion \

#include <errno.h>

of <winsock.h>. It may have been indirectly included (e.g., by including \

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

<windows.h>). Finthe inclusion of that header which is visible to this \

// Maximum iterations of Pipe processing in Monitor::run

compilation unit and #define FD_SETZIE to 1024 prior to that inclusion; \

const Uint32 maxIterations = 2;

otherwise, less than 64 clients (the default) will be able to connect to the \

CIMOM. PLEASE DO NOT SUPPRESS THIS WARNING; PLEASE FIX THE PROBLEM."

# endif

# define FD_SETSIZE 1024

# include <windows.h>

#else

# include <sys/types.h>

# include <sys/socket.h>

# include <sys/time.h>

# include <netinet/in.h>

# include <netdb.h>

# include <arpa/inet.h>

#endif

PEGASUS_USING_STD;

PEGASUS_NAMESPACE_BEGIN

static AtomicInt _connections(0);

static AtomicInt _connections = 0;

Mutex Monitor::_cout_mut;

// Added for NamedPipe implementation for windows

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

static struct timeval create_time = {0, 1};

#define PIPE_INCREMENT 1

static struct timeval destroy_time = {300, 0};

#endif

static struct timeval deadlock_time = {0, 0};

////////////////////////////////////////////////////////////////////////////////

// MonitorRep

////////////////////////////////////////////////////////////////////////////////

struct MonitorRep

{

fd_set rd_fd_set;

fd_set wr_fd_set;

fd_set ex_fd_set;

fd_set active_rd_fd_set;

fd_set active_wr_fd_set;

fd_set active_ex_fd_set;

};

////////////////////////////////////////////////////////////////////////////////

// Monitor

////////////////////////////////////////////////////////////////////////////////

#define MAX_NUMBER_OF_MONITOR_ENTRIES 32

Monitor::Monitor()

: _module_handle(0), _controller(0), _async(false)

: _stopConnections(0),

_stopConnectionsSem(0),

_solicitSocketCount(0),

_tickle_client_socket(-1),

_tickle_server_socket(-1),

_tickle_peer_socket(-1)

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

,_solicitPipeCount(0)

#endif

{

int numberOfMonitorEntriesToAllocate = MAX_NUMBER_OF_MONITOR_ENTRIES;

Socket::initializeInterface();

_rep = 0;

_entries.reserveCapacity(numberOfMonitorEntriesToAllocate);

_entries.reserveCapacity(32);

for( int i = 0; i < 32; i++ )

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

_entries_pipe.reserveCapacity(numberOfMonitorEntriesToAllocate);

#endif

// setup the tickler

initializeTickler();

// Start the count at 1 because initilizeTickler()

// has added an entry in the first position of the

// _entries array

for( int i = 1; i < numberOfMonitorEntriesToAllocate; i++ )

{

_MonitorEntry entry(0, 0, 0);

_entries.append(entry);

}

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

for( int i = 1; i < numberOfMonitorEntriesToAllocate; i++ )

{

_MonitorEntry entry(0, 0, 0);

_entries_pipe.append(entry);

}

#endif

}

Monitor::Monitor(Boolean async)

Monitor::~Monitor()

: _module_handle(0), _controller(0), _async(async)

{

Socket::initializeInterface();

uninitializeTickler();

_rep = 0;

Socket::uninitializeInterface();

_entries.reserveCapacity(32);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

for( int i = 0; i < 32; i++ )

"returning from monitor destructor");

}

void Monitor::uninitializeTickler(){

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "uninitializing interface");

try{

if(_tickle_peer_socket >= 0)

{

_MonitorEntry entry(0, 0, 0);

Socket::close(_tickle_peer_socket);

}

if(_tickle_client_socket >= 0)

{

Socket::close(_tickle_client_socket);

}

if(_tickle_server_socket >= 0)

{

Socket::close(_tickle_server_socket);

}

catch(...)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Failed to close tickle sockets");

}

void Monitor::initializeTickler(){

NOTE: On any errors trying to

setup out tickle connection,

throw an exception/end the server

/* setup the tickle server/listener */

// try until the tcpip is restarted

{

// get a socket for the server side

if((_tickle_server_socket = Socket::createSocket(PF_INET, SOCK_STREAM, 0)) == PEGASUS_INVALID_SOCKET)

{

//handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_CREATE",

"Received error number $0 while creating the internal socket.",

getSocketError());

throw Exception(parms);

}

// initialize the address

memset(&_tickle_server_addr, 0, sizeof(_tickle_server_addr));

#ifdef PEGASUS_PLATFORM_OS400_ISERIES_IBM

#pragma convert(37)

#endif

_tickle_server_addr.sin_addr.s_addr = inet_addr("127.0.0.1");

#ifdef PEGASUS_PLATFORM_OS400_ISERIES_IBM

#pragma convert(0)

#endif

_tickle_server_addr.sin_family = PF_INET;

_tickle_server_addr.sin_port = 0;

SocketLength _addr_size = sizeof(_tickle_server_addr);

// bind server side to socket

if((::bind(_tickle_server_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_server_addr),

sizeof(_tickle_server_addr))) < 0)

{

// handle error

#ifdef PEGASUS_OS_ZOS

MessageLoaderParms parms("Common.Monitor.TICKLE_BIND_LONG",

"Received error:$0 while binding the internal socket."

,strerror(errno));

#else

MessageLoaderParms parms("Common.Monitor.TICKLE_BIND",

"Received error number $0 while binding the internal socket.",

getSocketError());

#endif

throw Exception(parms);

}

// tell the kernel we are a server

if((::listen(_tickle_server_socket,3)) < 0)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_LISTEN",

"Received error number $0 while listening to the internal socket.",

getSocketError());

throw Exception(parms);

}

// make sure we have the correct socket for our server

int sock = ::getsockname(_tickle_server_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_server_addr),

&_addr_size);

if(sock < 0)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_SOCKNAME",

"Received error number $0 while getting the internal socket name.",

getSocketError());

throw Exception(parms);

}

/* set up the tickle client/connector */

// get a socket for our tickle client

if((_tickle_client_socket = Socket::createSocket(PF_INET, SOCK_STREAM, 0))

== PEGASUS_INVALID_SOCKET)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_CLIENT_CREATE",

"Received error number $0 while creating the internal client socket.",

getSocketError());

throw Exception(parms);

}

// setup the address of the client

memset(&_tickle_client_addr, 0, sizeof(_tickle_client_addr));

#ifdef PEGASUS_PLATFORM_OS400_ISERIES_IBM

#pragma convert(37)

#endif

_tickle_client_addr.sin_addr.s_addr = inet_addr("127.0.0.1");

#ifdef PEGASUS_PLATFORM_OS400_ISERIES_IBM

#pragma convert(0)

#endif

_tickle_client_addr.sin_family = PF_INET;

_tickle_client_addr.sin_port = 0;

// bind socket to client side

if((::bind(_tickle_client_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_client_addr),

sizeof(_tickle_client_addr))) < 0)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_CLIENT_BIND",

"Received error number $0 while binding the internal client socket.",

getSocketError());

throw Exception(parms);

}

// connect to server side

if((::connect(_tickle_client_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_server_addr),

sizeof(_tickle_server_addr))) < 0)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_CLIENT_CONNECT",

"Received error number $0 while connecting the internal client socket.",

getSocketError());

throw Exception(parms);

}

/* set up the slave connection */

memset(&_tickle_peer_addr, 0, sizeof(_tickle_peer_addr));

SocketLength peer_size = sizeof(_tickle_peer_addr);

Threads::sleep(1);

// this call may fail, we will try a max of 20 times to establish this peer connection

if((_tickle_peer_socket = ::accept(_tickle_server_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_peer_addr),

&peer_size)) < 0)

{

#if !defined(PEGASUS_OS_TYPE_WINDOWS)

if(_tickle_peer_socket == PEGASUS_SOCKET_ERROR

&& getSocketError() == PEGASUS_NETWORK_TRYAGAIN)

{

int retries = 0;

{

Threads::sleep(1);

_tickle_peer_socket = ::accept(_tickle_server_socket,

reinterpret_cast<struct sockaddr*>(&_tickle_peer_addr),

&peer_size);

retries++;

} while(_tickle_peer_socket == PEGASUS_SOCKET_ERROR

&& getSocketError() == PEGASUS_NETWORK_TRYAGAIN

&& retries < 20);

}

// TCP/IP is down, destroy sockets and retry again.

if(_tickle_peer_socket == PEGASUS_SOCKET_ERROR &&

getSocketError() == PEGASUS_NETWORK_TCPIP_STOPPED )

{

// destroy everything

uninitializeTickler();

// retry again.

continue;

}

#endif

}

if(_tickle_peer_socket == PEGASUS_SOCKET_ERROR)

{

// handle error

MessageLoaderParms parms("Common.Monitor.TICKLE_ACCEPT",

"Received error number $0 while accepting the internal socket connection.",

getSocketError());

throw Exception(parms);

} else

{

// socket is ok

break;

}

} while(1); // try until TCP/IP is restarted

Socket::disableBlocking(_tickle_peer_socket);

Socket::disableBlocking(_tickle_client_socket);

// add the tickler to the list of entries to be monitored and set to IDLE because Monitor only

// checks entries with IDLE state for events

_MonitorEntry entry(_tickle_peer_socket, 1, INTERNAL);

entry._status = _MonitorEntry::IDLE;

// is the tickler initalized as first socket on startup ?

if (_entries.size()==0)

{

// if yes, append a new entry

_entries.append(entry);

}

if( _async == true )

else

{

_thread_pool = new ThreadPool(0,

// if not, overwrite the tickler entry with new socket

"Monitor",

_entries[0]=entry;

}

create_time,

}

destroy_time,

deadlock_time);

void Monitor::tickle(void)

{

static char _buffer[] =

{

'0','0'

};

AutoMutex autoMutex(_tickle_mutex);

Socket::write(_tickle_client_socket,&_buffer, 2);

}

void Monitor::setState( Uint32 index, _MonitorEntry::entry_status status )

{

// Set the state to requested state

_entries[index]._status = status;

}

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

void Monitor::setPipeState( Uint32 index, _MonitorEntry::entry_status status )

{

// Set the state to requested state

_entries_pipe[index]._status = status;

}

int Monitor::handlePipe()

{

AutoMutex autoEntryMutex(_entry_mut);

ArrayIterator<_MonitorEntry> entries(_entries_pipe);

// Check the stopConnections flag. If set, clear the Acceptor monitor entries

if (_stopConnections.get() == 1)

{

for ( int indx = 0; indx < (int)entries.size(); indx++)

{

if (entries[indx]._type == Monitor::ACCEPTOR)

{

if ( entries[indx]._status.get() != _MonitorEntry::EMPTY)

{

if ( entries[indx]._status.get() == _MonitorEntry::IDLE ||

entries[indx]._status.get() == _MonitorEntry::DYING )

{

// remove the entry

entries[indx]._status = _MonitorEntry::EMPTY;

}

else

_thread_pool = 0;

{

// set status to DYING

entries[indx]._status = _MonitorEntry::DYING;

}

_stopConnections = 0;

_stopConnectionsSem.signal();

}

Monitor::~Monitor()

for( int indx = 0; indx < (int)entries.size(); indx++)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

const _MonitorEntry &entry = entries[indx];

"deregistering with module controller");

if ((entry._status.get() == _MonitorEntry::DYING) &&

(entry._type == Monitor::CONNECTION))

{

MessageQueue *q = MessageQueue::lookup(entry.queueId);

PEGASUS_ASSERT(q != 0);

HTTPConnection &h = *static_cast<HTTPConnection *>(q);

if (h._connectionClosePending == false)

continue;

// NOTE: do not attempt to delete while there are pending responses

// coming thru. The last response to come thru after a

// _connectionClosePending will reset _responsePending to false

// and then cause the monitor to rerun this code and clean up.

// (see HTTPConnection.cpp)

if (h._responsePending == true)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "Monitor::run - "

"Ignoring connection delete request because "

"responses are still pending. "

"connection=0x%p, NamedPipe=%d\n",

(void *)&h, h.getNamedPipe().getPipe());

continue;

}

h._connectionClosePending = false;

MessageQueue &o = h.get_owner();

Message* message = 0;

message= new CloseConnectionMessage(entry.namedPipe);

message->dest = o.getQueueId();

// HTTPAcceptor is responsible for closing the connection.

// The lock is released to allow HTTPAcceptor to call

// unsolicitSocketMessages to free the entry.

// Once HTTPAcceptor completes processing of the close

// connection, the lock is re-requested and processing of

// the for loop continues. This is safe with the current

// implementation of the entries object. Note that the

// loop condition accesses the entries.size() on each

// iteration, so that a change in size while the mutex is

// unlocked will not result in an ArrayIndexOutOfBounds

// exception.

_entry_mut.unlock();

o.enqueue(message);

_entry_mut.lock();

// After enqueue a message and the autoEntryMutex has been released and locked again,

// the array of _entries can be changed. The ArrayIterator has be reset with the original _entries.

entries.reset(_entries_pipe);

}

Uint32 _idleEntries = 0;

We will keep track of the maximum socket number and pass this value

to the kernel as a parameter to SELECT. This loop seems like a good

place to calculate the max file descriptor (maximum socket number)

because we have to traverse the entire array.

SocketHandle maxSocketCurrentPass = 0;

int indx = 0;

Array <Uint32> indexPipeCountAssociator;

int pipeEntryCount=0;

int MaxPipes = PIPE_INCREMENT;

// List of Pipe Handlers

HANDLE * hPipeList = new HANDLE[PIPE_INCREMENT];

for( indx = 0; indx < (int)entries.size() ; indx++)

{

if (!entries[indx].namedPipeConnection)

continue;

if(_module_handle != NULL)

entries[indx].pipeSet = false;

if (pipeEntryCount >= MaxPipes)

{

MaxPipes += PIPE_INCREMENT;

HANDLE* temp_pList = new HANDLE[MaxPipes];

for (Uint32 i =0;i<pipeEntryCount;i++)

{

_controller->deregister_module(PEGASUS_MODULENAME_MONITOR);

temp_pList[i] = hPipeList[i];

_controller = 0;

delete _module_handle;

}

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "deleting rep");

delete [] hPipeList;

hPipeList = temp_pList;

}

hPipeList[pipeEntryCount] = entries[indx].namedPipe.getPipe();

indexPipeCountAssociator.append(indx);

pipeEntryCount++;

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "uninitializing interface");

Socket::uninitializeInterface();

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"returning from monitor destructor");

if(_async == true)

delete _thread_pool;

}

Add 1 then assign maxSocket accordingly. We add 1 to account for

descriptors starting at 0.

maxSocketCurrentPass++;

int pEvents = -1;

int pCount = -1;

BOOL bPeekPipe = 0;

DWORD dwBytesAvail=0;

// The pipe is sniffed and check if there are any data. If available, the

// message is picked from the Queue and appropriate methods are invoked.

// pipeProcessCount records the number of requests that are processed.

// At the end of loop this is verified against the count of request

// on local connection . If there are any pipes which needs to be

// processed we would apply delay and then proceed to iterate.

Uint32 pipeProcessCount =0;

_entry_mut.unlock();

// pipeIndex is used to index into indexPipeCountAssociator to fetch

// index of the _MonitorEntry of Monitor

for (int pipeIndex = 0; pipeIndex < pipeEntryCount; pipeIndex++)

{

dwBytesAvail = 0;

bPeekPipe = ::PeekNamedPipe(hPipeList[pipeIndex],

NULL,

&dwBytesAvail,

NULL

);

// If peek on NamedPipe was successfull and data is available

if (bPeekPipe && dwBytesAvail)

{

// Tracer::trace(TRC_HTTP,Tracer::LEVEL4," PIPE_PEEKING FOUND = %u BYTES", dwBytesAvail);

pEvents = 1;

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "EVENT TRIGGERED in Pipe = %u ",entries[indexPipeCountAssociator[pipeIndex]].namedPipe.getPipe());

entries[indexPipeCountAssociator[pipeIndex]].pipeSet = true;

int pIndx = indexPipeCountAssociator[pipeIndex];

if ((entries[pIndx]._status.get() == _MonitorEntry::IDLE) &&

entries[pIndx].namedPipe.isConnected() &&

(pEvents))

{

MessageQueue *q = 0;

try

{

int Monitor::kill_idle_threads()

q = MessageQueue::lookup (entries[pIndx].queueId);

}

catch (Exception e)

{

static struct timeval now, last;

e.getMessage();

gettimeofday(&now, NULL);

}

int dead_threads = 0;

catch(...)

{

}

if( now.tv_sec - last.tv_sec > 120 )

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::run indx = %d, queueId = %d,\

q = %p",pIndx, entries[pIndx].queueId, q);

try

{

gettimeofday(&last, NULL);

if (entries[pIndx]._type == Monitor::CONNECTION)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"entries[indx].type for indx = \

%d is Monitor::CONNECTION",

pIndx);

static_cast<HTTPConnection *>(q)->_entry_index = pIndx;

HTTPConnection *dst = reinterpret_cast \

<HTTPConnection *>(q);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: entering run() \

for indx = %d, queueId = %d, \

q = %p",\

dst->_entry_index,

dst->_monitor->_entries\

[dst->_entry_index].queueId, dst);

try

{

dead_threads = _thread_pool->kill_dead_threads();

dst->run(1);

// Record that the requested data is read/Written

pipeProcessCount++;

}

catch (...)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: \

exception received");

}

catch(IPCException& )

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: exited \

\run() index %d",

dst->_entry_index);

}

else

{

/* The condition

entries[indx]._type == Monitor::INTERNAL can be

ignored for pipes as the tickler is of

Monitor::INTERNAL type. The tickler is

a socket.

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Non-connection entry, indx = %d,\

has been received.", pIndx);

int events = 0;

Message *msg = 0;

pEvents |= NamedPipeMessage::READ;

msg = new NamedPipeMessage(entries[pIndx].namedPipe, pEvents);

entries[pIndx]._status = _MonitorEntry::BUSY;

q->enqueue(msg);

_entry_mut.lock();

entries.reset(_entries_pipe);

entries[pIndx]._status = _MonitorEntry::IDLE;

delete [] hPipeList;

return 1;

}

return dead_threads;

catch(...)

{

}

_entry_mut.lock();

}

delete [] hPipeList;

return 1;

}

#endif

Boolean Monitor::run(Uint32 milliseconds)

void Monitor::run(Uint32 milliseconds)

{

Boolean handled_events = false;

int i = 0;

#if defined(PEGASUS_OS_OS400) || defined(PEGASUS_OS_HPUX)

struct timeval tv = {milliseconds/1000, milliseconds%1000*1000};

#else

struct timeval tv = {0, 1};

#endif

fd_set fdread;

FD_ZERO(&fdread);

_entry_mut.lock(pegasus_thread_self());

for( int indx = 0; indx < (int)_entries.size(); indx++)

AutoMutex autoEntryMutex(_entry_mut);

ArrayIterator<_MonitorEntry> entries(_entries);

// Check the stopConnections flag. If set, clear the Acceptor monitor entries

if (_stopConnections.get() == 1)

{

for ( int indx = 0; indx < (int)entries.size(); indx++)

{

if (entries[indx]._type == Monitor::ACCEPTOR)

{

if(_entries[indx]._status.value() == _MonitorEntry::IDLE)

if ( entries[indx]._status.get() != _MonitorEntry::EMPTY)

{

FD_SET(_entries[indx].socket, &fdread);

if ( entries[indx]._status.get() == _MonitorEntry::IDLE ||

entries[indx]._status.get() == _MonitorEntry::DYING )

{

// remove the entry

entries[indx]._status = _MonitorEntry::EMPTY;

}

else

{

// set status to DYING

entries[indx]._status = _MonitorEntry::DYING;

}

_stopConnections = 0;

_stopConnectionsSem.signal();

}

for( int indx = 0; indx < (int)entries.size(); indx++)

{

const _MonitorEntry &entry = entries[indx];

if ((entry._status.get() == _MonitorEntry::DYING) &&

(entry._type == Monitor::CONNECTION))

{

MessageQueue *q = MessageQueue::lookup(entry.queueId);

PEGASUS_ASSERT(q != 0);

HTTPConnection &h = *static_cast<HTTPConnection *>(q);

if (h._connectionClosePending == false)

continue;

// NOTE: do not attempt to delete while there are pending responses

// coming thru. The last response to come thru after a

// _connectionClosePending will reset _responsePending to false

// and then cause the monitor to rerun this code and clean up.

// (see HTTPConnection.cpp)

if (h._responsePending == true)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4, "Monitor::run - "

"Ignoring connection delete request because "

"responses are still pending. "

"connection=0x%p, socket=%d\n",

(void *)&h, h.getSocket());

continue;

}

h._connectionClosePending = false;

MessageQueue &o = h.get_owner();

Message* message= new CloseConnectionMessage(entry.socket);

message->dest = o.getQueueId();

int events = select(FD_SETSIZE, &fdread, NULL, NULL, &tv);

// HTTPAcceptor is responsible for closing the connection.

// The lock is released to allow HTTPAcceptor to call

// unsolicitSocketMessages to free the entry.

// Once HTTPAcceptor completes processing of the close

// connection, the lock is re-requested and processing of

// the for loop continues. This is safe with the current

// implementation of the entries object. Note that the

// loop condition accesses the entries.size() on each

// iteration, so that a change in size while the mutex is

// unlocked will not result in an ArrayIndexOutOfBounds

// exception.

_entry_mut.unlock();

o.enqueue(message);

_entry_mut.lock();

// After enqueue a message and the autoEntryMutex has been released and locked again,

// the array of _entries can be changed. The ArrayIterator has be reset with the original _entries.

entries.reset(_entries);

}

Uint32 _idleEntries = 0;

We will keep track of the maximum socket number and pass this value

to the kernel as a parameter to SELECT. This loop seems like a good

place to calculate the max file descriptor (maximum socket number)

because we have to traverse the entire array.

SocketHandle maxSocketCurrentPass = 0;

for( int indx = 0; indx < (int)entries.size(); indx++)

{

if(maxSocketCurrentPass < entries[indx].socket)

maxSocketCurrentPass = entries[indx].socket;

if(entries[indx]._status.get() == _MonitorEntry::IDLE)

{

_idleEntries++;

FD_SET(entries[indx].socket, &fdread);

}

Add 1 then assign maxSocket accordingly. We add 1 to account for

descriptors starting at 0.

maxSocketCurrentPass++;

_entry_mut.unlock();

// The first argument to select() is ignored on Windows and it is not

// a socket value. The original code assumed that the number of sockets

// and a socket value have the same type. On Windows they do not.

#ifdef PEGASUS_OS_TYPE_WINDOWS

if(events && events != SOCKET_ERROR )

int events = select(0, &fdread, NULL, NULL, &tv);

#else

if(events && events != -1 )

int events = select(maxSocketCurrentPass, &fdread, NULL, NULL, &tv);

#endif

_entry_mut.lock();

// After enqueue a message and the autoEntryMutex has been released and locked again,

// the array of _entries can be changed. The ArrayIterator has be reset with the original _entries

entries.reset(_entries);

if (events == PEGASUS_SOCKET_ERROR)

{

for( int indx = 0; indx < (int)_entries.size(); indx++)

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

{

"Monitor::run - errorno = %d has occurred on select.", errno);

if(FD_ISSET(_entries[indx].socket, &fdread))

// The EBADF error indicates that one or more or the file

// descriptions was not valid. This could indicate that

// the entries structure has been corrupted or that

// we have a synchronization error.

PEGASUS_ASSERT(errno != EBADF);

}

else if (events)

{

MessageQueue *q = MessageQueue::lookup(_entries[indx].queueId);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

if(q == 0)

"Monitor::run select event received events = %d, monitoring %d idle entries",

events, _idleEntries);

for( int indx = 0; indx < (int)entries.size(); indx++)

{

// The Monitor should only look at entries in the table that are IDLE (i.e.,

// owned by the Monitor).

if((entries[indx]._status.get() == _MonitorEntry::IDLE) &&

(FD_ISSET(entries[indx].socket, &fdread)))

{

MessageQueue *q = MessageQueue::lookup(entries[indx].queueId);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::run indx = %d, queueId = %d, q = %p",

indx, entries[indx].queueId, q);

PEGASUS_ASSERT(q !=0);

try

{

_entries[indx]._status = _MonitorEntry::EMPTY;

if(entries[indx]._type == Monitor::CONNECTION)

}

catch(...)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"entries[indx].type for indx = %d is Monitor::CONNECTION", indx);

static_cast<HTTPConnection *>(q)->_entry_index = indx;

// Do not update the entry just yet. The entry gets updated once

// the request has been read.

//entries[indx]._status = _MonitorEntry::BUSY;

// If allocate_and_awaken failure, retry on next iteration

/* Removed for PEP 183.

if (!MessageQueueService::get_thread_pool()->allocate_and_awaken(

(void *)q, _dispatch))

{

Tracer::trace(TRC_DISCARDED_DATA, Tracer::LEVEL2,

"Monitor::run: Insufficient resources to process request.");

entries[indx]._status = _MonitorEntry::IDLE;

return true;

}

continue;

}

// Added for PEP 183

HTTPConnection *dst = reinterpret_cast<HTTPConnection *>(q);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: entering run() for indx = %d, queueId = %d, q = %p",

dst->_entry_index, dst->_monitor->_entries[dst->_entry_index].queueId, dst);

try

{

if(_entries[indx]._type == Monitor::CONNECTION)

dst->run(1);

}

catch (...)

{

static_cast<HTTPConnection *>(q)->_entry_index = indx;

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

if(static_cast<HTTPConnection *>(q)->_dying.value() > 0 )

"Monitor::_dispatch: exception received");

}

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: exited run() for index %d", dst->_entry_index);

// It is possible the entry status may not be set to busy.

// The following will fail in that case.

// PEGASUS_ASSERT(dst->_monitor->_entries[dst->_entry_index]._status.get() == _MonitorEntry::BUSY);

// Once the HTTPConnection thread has set the status value to either

// Monitor::DYING or Monitor::IDLE, it has returned control of the connection

// to the Monitor. It is no longer permissible to access the connection

// or the entry in the _entries table.

// The following is not relevant as the worker thread or the

// reader thread will update the status of the entry.

//if (dst->_connectionClosePending)

//{

// dst->_monitor->_entries[dst->_entry_index]._status = _MonitorEntry::DYING;

//}

//else

//{

// dst->_monitor->_entries[dst->_entry_index]._status = _MonitorEntry::IDLE;

//}

// end Added for PEP 183

}

else if( entries[indx]._type == Monitor::INTERNAL){

// set ourself to BUSY,

// read the data

// and set ourself back to IDLE

entries[indx]._status = _MonitorEntry::BUSY;

static char buffer[2];

Sint32 amt = Socket::read(entries[indx].socket,&buffer, 2);

if(amt == PEGASUS_SOCKET_ERROR &&

getSocketError() == PEGASUS_NETWORK_TCPIP_STOPPED )

{

_entries[indx]._status = _MonitorEntry::DYING;

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

MessageQueue & o = static_cast<HTTPConnection *>(q)->get_owner();

"Monitor::run: Tickler socket got an IO error. "

Message* message= new CloseConnectionMessage(_entries[indx].socket);

"Going to re-create Socket and wait for TCP/IP restart.");

message->dest = o.getQueueId();

uninitializeTickler();

_entry_mut.unlock();

initializeTickler();

o.enqueue(message);

return true;

} else

{

entries[indx]._status = _MonitorEntry::IDLE;

}

_entries[indx]._status = _MonitorEntry::BUSY;

_thread_pool->allocate_and_awaken((void *)q, _dispatch);

}

else

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Non-connection entry, indx = %d, has been received.", indx);

int events = 0;

events |= SocketMessage::READ;

Message *msg = new SocketMessage(_entries[indx].socket, events);

Message *msg = new SocketMessage(entries[indx].socket, events);

_entries[indx]._status = _MonitorEntry::BUSY;

entries[indx]._status = _MonitorEntry::BUSY;

_entry_mut.unlock();

q->enqueue(msg);

_entries[indx]._status = _MonitorEntry::IDLE;

_entry_mut.lock();

return true;

// After enqueue a message and the autoEntryMutex has been released and locked again,

// the array of entries can be changed. The ArrayIterator has be reset with the original _entries

entries.reset(_entries);

entries[indx]._status = _MonitorEntry::IDLE;

}

catch(...)

{

}

handled_events = true;

}

_entry_mut.unlock();

}

return(handled_events);

void Monitor::stopListeningForConnections(Boolean wait)

{

PEG_METHOD_ENTER(TRC_HTTP, "Monitor::stopListeningForConnections()");

// set boolean then tickle the server to recognize _stopConnections

_stopConnections = 1;

tickle();

if (wait)

{

// Wait for the monitor to notice _stopConnections. Otherwise the

// caller of this function may unbind the ports while the monitor

// is still accepting connections on them.

_stopConnectionsSem.wait();

}

PEG_METHOD_EXIT();

}

int Monitor::solicitSocketMessages(

Sint32 socket,

SocketHandle socket,

Uint32 events,

Uint32 queueId,

int type)

{

PEG_METHOD_ENTER(TRC_HTTP, "Monitor::solicitSocketMessages");

AutoMutex autoMut(_entry_mut);

// Check to see if we need to dynamically grow the _entries array

// We always want the _entries array to 2 bigger than the

// current connections requested

_solicitSocketCount++; // bump the count

int size = (int)_entries.size();

if((int)_solicitSocketCount >= (size-1)){

for(int i = 0; i < ((int)_solicitSocketCount - (size-1)); i++){

_MonitorEntry entry(0, 0, 0);

_entries.append(entry);

}

int index = -1;

int index;

_entry_mut.lock(pegasus_thread_self());

for(index = 1; index < (int)_entries.size(); index++)

for(index = 0; index < (int)_entries.size(); index++)

{

try

{

if(_entries[index]._status.value() == _MonitorEntry::EMPTY)

if(_entries[index]._status.get() == _MonitorEntry::EMPTY)

{

_entries[index].socket = socket;

_entries[index].queueId = queueId;

_entries[index]._type = type;

_entries[index]._status = _MonitorEntry::IDLE;

_entry_mut.unlock();

return index;

}

Line 294

Line 990

catch(...)

{

}

_entry_mut.unlock();

_solicitSocketCount--; // decrease the count, if we are here we didnt do anything meaningful

PEG_METHOD_EXIT();

return index;

return -1;

}

void Monitor::unsolicitSocketMessages(Sint32 socket)

void Monitor::unsolicitSocketMessages(SocketHandle socket)

{

PEG_METHOD_ENTER(TRC_HTTP, "Monitor::unsolicitSocketMessages");

_entry_mut.lock(pegasus_thread_self());

AutoMutex autoMut(_entry_mut);

for(int index = 0; index < (int)_entries.size(); index++)

Start at index = 1 because _entries[0] is the tickle entry which never needs

to be EMPTY;

unsigned int index;

for(index = 1; index < _entries.size(); index++)

{

if(_entries[index].socket == socket)

{

_entries[index]._status = _MonitorEntry::EMPTY;

_entries[index].socket = PEGASUS_INVALID_SOCKET;

_solicitSocketCount--;

break;

}

_entry_mut.unlock();

Dynamic Contraction:

To remove excess entries we will start from the end of the _entries array

and remove all entries with EMPTY status until we find the first NON EMPTY.

This prevents the positions, of the NON EMPTY entries, from being changed.

index = _entries.size() - 1;

while(_entries[index]._status.get() == _MonitorEntry::EMPTY){

if(_entries.size() > MAX_NUMBER_OF_MONITOR_ENTRIES)

_entries.remove(index);

index--;

}

PEG_METHOD_EXIT();

}

PEGASUS_THREAD_RETURN PEGASUS_THREAD_CDECL Monitor::_dispatch(void *parm)

// Note: this is no longer called with PEP 183.

ThreadReturnType PEGASUS_THREAD_CDECL Monitor::_dispatch(void *parm)

{

HTTPConnection *dst = reinterpret_cast<HTTPConnection *>(parm);

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: entering run() for indx = %d, queueId = %d, q = %p",

dst->_entry_index, dst->_monitor->_entries[dst->_entry_index].queueId, dst);

try

{

dst->run(1);

if( dst->_monitor->_entries.size() > (Uint32)dst->_entry_index )

}

dst->_monitor->_entries[dst->_entry_index]._status = _MonitorEntry::IDLE;

catch (...)

{

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: exception received");

}

Tracer::trace(TRC_HTTP, Tracer::LEVEL4,

"Monitor::_dispatch: exited run() for index %d", dst->_entry_index);

PEGASUS_ASSERT(dst->_monitor->_entries[dst->_entry_index]._status.get() == _MonitorEntry::BUSY);

// Once the HTTPConnection thread has set the status value to either

// Monitor::DYING or Monitor::IDLE, it has returned control of the connection

// to the Monitor. It is no longer permissible to access the connection

// or the entry in the _entries table.

if (dst->_connectionClosePending)

{

dst->_monitor->_entries[dst->_entry_index]._status = _MonitorEntry::DYING;

}

else

{

dst->_monitor->_entries[dst->_entry_index]._status = _MonitorEntry::IDLE;

}

return 0;

}

// Added for NamedPipe implementation for windows

#if defined PEGASUS_OS_TYPE_WINDOWS && !defined(PEGASUS_DISABLE_LOCAL_DOMAIN_SOCKET)

//This method is anlogus to solicitSocketMessages. It does the same thing for named Pipes

int Monitor::solicitPipeMessages(

NamedPipe namedPipe,

Uint32 events, //not sure what has to change for this enum

Uint32 queueId,

int type)

{

PEG_METHOD_ENTER(TRC_HTTP, "Monitor::solicitPipeMessages");

AutoMutex autoMut(_entry_pipe_mut);

// Check to see if we need to dynamically grow the _entries array

// We always want the _entries array to 2 bigger than the

// current connections requested

#ifdef PEGASUS_LOCALDOMAINSOCKET_DEBUG

{

AutoMutex automut(Monitor::_cout_mut);

PEGASUS_STD(cout) << "In Monitor::solicitPipeMessages at the begining" << PEGASUS_STD(endl);

}

#endif

_solicitPipeCount++; // bump the count

int size = (int)_entries_pipe.size();

if((int)_solicitPipeCount >= (size-1)){

for(int i = 0; i < ((int)_solicitPipeCount - (size-1)); i++){

_MonitorEntry entry(0, 0, 0);

_entries_pipe.append(entry);

}

int index;

for(index = 1; index < (int)_entries_pipe.size(); index++)

{

try

{

if(_entries_pipe[index]._status.get() == _MonitorEntry::EMPTY)

{

_entries_pipe[index].socket = NULL;

_entries_pipe[index].namedPipe = namedPipe;

_entries_pipe[index].namedPipeConnection = true;

_entries_pipe[index].queueId = queueId;

_entries_pipe[index]._type = type;

_entries_pipe[index]._status = _MonitorEntry::IDLE;

#ifdef PEGASUS_LOCALDOMAINSOCKET_DEBUG

{

AutoMutex automut(Monitor::_cout_mut);

PEGASUS_STD(cout) << "In Monitor::solicitPipeMessages after seting up _entries[index] index = " << index << PEGASUS_STD(endl);

}

#endif

return index;

}

catch(...)

{

}

_solicitPipeCount--; // decrease the count, if we are here we didnt do anything meaningful

PEGASUS_STD(cout) << "In Monitor::solicitPipeMessages nothing happed - it didn't work" << PEGASUS_STD(endl);

PEG_METHOD_EXIT();

return -1;

}

//////////////////////////////////////////////////////////////////////////////

// Method Name : unsolicitPipeMessages

// Input Parameter : namedPipe - type NamedPipe

// Return Type : void

//============================================================================

// This method is invoked from HTTPAcceptor::handleEnqueue for server

// when the CLOSE_CONNECTION_MESSAGE is recieved. This method is also invoked

// from HTTPAcceptor::destroyConnections method when the CIMServer is shutdown.

// For the CIMClient, this is invoked from HTTPConnector::handleEnqueue when the

// CLOSE_CONNECTION_MESSAGE is recieved. This method is also invoked from

// HTTPConnector::disconnect when CIMClient requests a disconnect request.

// The list of _MonitorEntry is searched for the matching pipe.

// The Handle of the identified is closed and _MonitorEntry for the

// requested pipe is removed.

///////////////////////////////////////////////////////////////////////////////

void Monitor::unsolicitPipeMessages(NamedPipe namedPipe)

{

#ifdef PEGASUS_LOCALDOMAINSOCKET_DEBUG

{

AutoMutex automut(Monitor::_cout_mut);

PEGASUS_STD(cout) << "Entering: Monitor::unsolicitPipeMessages(): (tid:" << Uint32(pegasus_thread_self()) << ")" << PEGASUS_STD(endl);

}

#endif

PEG_METHOD_ENTER(TRC_HTTP, "Monitor::unsolicitPipeMessages");

AutoMutex autoMut(_entry_pipe_mut);

Start at index = 1 because _entries[0] is the tickle entry which never needs

to be EMPTY;

unsigned int index;

for (index = 1; index < _entries_pipe.size(); index++)

{

if (_entries_pipe[index].namedPipe.getPipe() == namedPipe.getPipe())

{

_entries_pipe[index]._status = _MonitorEntry::EMPTY;

// Ensure that the client has read the data

::FlushFileBuffers (namedPipe.getPipe());

//Disconnect to release the pipe. This doesn't release Pipe Handle

::DisconnectNamedPipe (_entries_pipe[index].namedPipe.getPipe());

// Must use CloseHandle to Close Pipe

::CloseHandle(_entries_pipe[index].namedPipe.getPipe());

_entries_pipe[index].namedPipe.disconnect();

_solicitPipeCount--;

break;

}

Dynamic Contraction:

To remove excess entries we will start from the end of the _entries array

and remove all entries with EMPTY status until we find the first NON EMPTY.

This prevents the positions, of the NON EMPTY entries, from being changed.

index = _entries_pipe.size() - 1;

while (_entries_pipe[index]._status.get() == _MonitorEntry::EMPTY

&& index > 0)

{

if ((_entries_pipe[index].namedPipe.getPipe() == namedPipe.getPipe()) ||

(_entries_pipe.size() > MAX_NUMBER_OF_MONITOR_ENTRIES))

{

_entries_pipe.remove(index);

}

index--;

}

PEG_METHOD_EXIT();

#ifdef PEGASUS_LOCALDOMAINSOCKET_DEBUG

{

AutoMutex automut(Monitor::_cout_mut);

PEGASUS_STD(cout) << "Exiting: Monitor::unsolicitPipeMessages(): (tid:" << Uint32(pegasus_thread_self()) << ")" << PEGASUS_STD(endl);

}

#endif

}

#endif

Legend:

Removed from v.1.39
changed lines
	Added in v.1.115.14.2

No CVS admin address has been configured