[u/mrichter/AliRoot.git] / MONITOR / AliOnlineReco.cxx

// @(#)root/eve:$Id$
// Author: Matevz Tadel 2007

/**************************************************************************
 * Copyright(c) 1998-2008, ALICE Experiment at CERN, all rights reserved. *)
 * See http://aliceinfo.cern.ch/Offline/AliRoot/License.html for          *
 * full copyright notice.                                                 *
 **************************************************************************/

#include "AliOnlineReco.h"
#include "AliChildProcTerminator.h"
#include "AliDimIntNotifier.h"
#include "AliCDBManager.h"
#include "AliGRPPreprocessor.h"

#include <TTimer.h>

#include <TGListBox.h>
#include <TGButton.h>

#include <TInterpreter.h>
#include <TROOT.h>

#include <unistd.h>
#include <signal.h>

//______________________________________________________________________________
// Full description of AliOnlineReco
//

ClassImp(AliOnlineReco)

AliOnlineReco::AliOnlineReco() :
  TGMainFrame(gClient->GetRoot(), 400, 400),

  fRunList(0), fAutoRun(0), fStartButt(0), fStopButt(0), fExitButt(0),
  fAutoRunTimer(0), fAutoRunScheduled(0), fAutoRunRunning(0),
  fRun2PidMap(),
  fTestMode(kFALSE),
  fDoExit(kFALSE)
{
  // Constructor.

  // GUI components.
  fRunList = new TGListBox(this);
  AddFrame(fRunList, new TGLayoutHints(kLHintsNormal | kLHintsExpandX | kLHintsExpandY));

  TGHorizontalFrame *hf = new TGHorizontalFrame(this, 1, 20);

  fAutoRun = new TGCheckButton(hf, "AutoRun");
  hf->AddFrame(fAutoRun, new TGLayoutHints(kLHintsNormal | kLHintsExpandX | kLHintsExpandY));
  fAutoRun->Connect("Clicked()", "AliOnlineReco", this, "DoAutoRun()");

  fStartButt = new TGTextButton(hf, "Start");
  hf->AddFrame(fStartButt, new TGLayoutHints(kLHintsNormal | kLHintsExpandX | kLHintsExpandY));
  fStartButt->Connect("Clicked()", "AliOnlineReco", this, "DoStart()");

  fStopButt = new TGTextButton(hf, "Stop");
  hf->AddFrame(fStopButt, new TGLayoutHints(kLHintsNormal | kLHintsExpandX | kLHintsExpandY));
  fStopButt->Connect("Clicked()", "AliOnlineReco", this, "DoStop()");

  fExitButt = new TGTextButton(hf, "Exit");
  hf->AddFrame(fExitButt, new TGLayoutHints(kLHintsNormal | kLHintsExpandX | kLHintsExpandY));
  fExitButt->Connect("Clicked()", "AliOnlineReco", this, "DoExit()");

  AddFrame(hf, new TGLayoutHints(kLHintsNormal | kLHintsExpandX));

  MapSubwindows();
  Layout();
  SetWindowName("Alice Online Reconstruction");

  // DIM interface.  
  for (Int_t i = 0; i < 5; ++i)
  {
    if (i == 0)
    {
      fSOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS");
      fEOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS");
    }
    else
    {
      fSOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS_%d", i));
      fEOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS_%d", i));
    }

    fSOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "StartOfRun(Int_t)");
    fEOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "EndOfRun(Int_t)");
  }

  const Int_t autoRunDelay = 10; // should go to config
  fAutoRunTimer = new TTimer(autoRunDelay * 1000l);
  fAutoRunTimer->Connect("Timeout()", "AliOnlineReco", this, "AutoRunTimerTimeout()");

  // OS Signal handlers
  // ROOT's TSignalHAndler works not SIGCHLD ...
  AliChildProcTerminator::Instance()->Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ChildProcTerm(Int_t,Int_t)");
  
  // we need this by OnExit() to kill next process child after another
  Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ExitLoopChildProcTerm()");
}

AliOnlineReco::~AliOnlineReco()
{
  // Destructor.

  delete fAutoRunTimer;
}

Int_t AliOnlineReco::GetLastRun() const
{
  // Returns the last started run.

  return fRun2PidMap.empty() ? 0 : fRun2PidMap.rbegin()->first;
}

Bool_t AliOnlineReco::GetAutoRunMode() const
{
  // Return state of auto-run flag.

  return fAutoRun->IsOn();
}

void AliOnlineReco::SetAutoRunMode(Bool_t ar)
{
  // Set auto-run flag.

  if (ar == fAutoRun->IsOn())
    return;

  fAutoRun->SetState(ar ? kButtonDown : kButtonUp, kTRUE);
}

//------------------------------------------------------------------------------
// Private methods
//------------------------------------------------------------------------------

AliOnlineReco::mIntInt_i AliOnlineReco::FindMapEntryByPid(Int_t pid)
{
  // Find run-to-pid map iterator by pid.
  // Requires iteration over map.

  for (mIntInt_i i = fRun2PidMap.begin(); i != fRun2PidMap.end(); ++i)
  {
    if (i->second == pid)
      return i;
  }

  return fRun2PidMap.end();
}

void AliOnlineReco::StartAliEve(mIntInt_i& mi)
{
  // Start alieve to process run given my the run-pid entry.

  Int_t run = mi->first;

  if (mi->second == 0)
  {
    pid_t pid = fork();
    if (pid == -1)
    {
      perror("DoStart -- Fork failed");
      return;
    }

    if (pid)
    {
      mi->second = pid;
      fRunList->RemoveEntry(run);
      fRunList->AddEntrySort(TString::Format("%-20d -- RUNNING", run), run);
      fRunList->Layout();
    }
    else
    {
      gCINTMutex = 0;

      struct sigaction sac;
      memset(&sac, 0, sizeof(sac));
      sac.sa_handler = NULL;
      sigemptyset(&sac.sa_mask);
      sac.sa_flags = 0;
      
      // The sa_restorer field is Not POSIX and obsolete.
      // This is for compilation on other systems
      #if defined(__linux) && \
          (defined(__i386__) || defined(__x86_64__)) && \
           defined(__GNUC__)
      sac.sa_restorer= NULL;
      #endif
      sigaction(SIGCHLD, &sac, NULL);
  
      int s;
      if (fTestMode)
      {
        s = execlp("alitestproc", "alitestproc", TString::Format("%d", run).Data(), (char*) 0);
      }
      else
      {
        Int_t procPID = gSystem->GetPid();
        TString logFile = Form("%s/reco/log/run%d_%d.log",
             gSystem->Getenv("ONLINERECO_BASE_DIR"),
             run,
            (Int_t)procPID);
        Info("DoStart","Reconstruction log will be written to %s",logFile.Data());
        gSystem->RedirectOutput(logFile.Data());

        gSystem->cd(Form("%s/reco",gSystem->Getenv("ONLINERECO_BASE_DIR")));

        TString gdcs;
        if (RetrieveGRP(run,gdcs) <= 0 || gdcs.IsNull()) 
          gSystem->Exit(1);

        gSystem->Setenv("DATE_RUN_NUMBER", Form("%d", run));
        // Setting CDB
        // 	AliCDBManager * man = AliCDBManager::Instance();
        // 	man->SetDefaultStorage("local:///local/cdb");
        // 	man->SetSpecificStorage("GRP/GRP/Data",
        // 			      Form("local://%s",gSystem->pwd()));
        // 	man->SetSpecificStorage("GRP/CTP/Config",
        // 			      Form("local://%s",gSystem->pwd()));
        // 	man->SetSpecificStorage("ACORDE/Align/Data",
        // 				"local://$ALICE_ROOT/OCDB");

        gSystem->mkdir(Form("run%d_%d", run, (Int_t)procPID));
        gSystem->cd(Form("run%d_%d", run, (Int_t)procPID));

        TString recMacroPath(gSystem->Getenv("ONLINERECO_MACRO"));
        if (recMacroPath.IsNull()) {
          recMacroPath = "$ALICE_ROOT/MONITOR/rec.C";
        }
        
        s = execlp("alieve",
             "alieve",
             "-q",
             Form("%s(\"mem://@*:\")", gSystem->ExpandPathName(recMacroPath.Data())),
             (char*) 0);

        gSystem->Exec(Form("rm -rf %s/reco/run%d_%d",gSystem->Getenv("ONLINERECO_BASE_DIR"),run,(Int_t)procPID));
      }

      if (s == -1)
      {
        perror("execlp failed - this will not end well");
        gSystem->Exit(1);
      }
    }
  }
  else
  {
    Error("DoStart", "Process already running.");
  }
}

void AliOnlineReco::KillPid(Int_t pid)
{
  // Terminate process given by pid.

  // Send terminate signal to process ...

  if (fTestMode)
  {
    kill(pid, SIGTERM);
  }
  else
  {
    // alieve will auto-destruct on SIGUSR1
    kill(pid, SIGUSR1);
  }
}

void AliOnlineReco::StartAutoRunTimer(Int_t run)
{
  // Start timer for given run.
  // If an auto-started run is already active, this call is ignored.
  // If timer is already active, it is restarted.

  if (fAutoRunRunning)
    return;

  fAutoRunTimer->Reset();
  fAutoRunTimer->TurnOn();
  fAutoRunScheduled = run;

  Info("StartAutoRunTimer", "Scheduling run %d for auto-display.", run);
}

void AliOnlineReco::StopAutoRunTimer()
{
  // Stop auto-run timer.

  fAutoRunTimer->TurnOff();
  fAutoRunScheduled = 0;
}

void AliOnlineReco::AutoRunTimerTimeout()
{
  // Slot called on auto-timer timeout.

  Int_t run = fAutoRunScheduled;

  StopAutoRunTimer();

  mIntInt_i i = fRun2PidMap.find(run);

  if (i == fRun2PidMap.end())
  {
    Warning("AutoRunTimerTimeout", "run no longer active.");
    return;
  }

  Info("AutoRunTimerTimeout", "Starting display for run %d.", run);

  StartAliEve(i);
  fAutoRunRunning = run;
}

//------------------------------------------------------------------------------
// Handlers of DIM signals.
//------------------------------------------------------------------------------

void AliOnlineReco::StartOfRun(Int_t run)
{
  // Slot called from DIM handler on start of run.

  mIntInt_i i = fRun2PidMap.find(run);
  if (i == fRun2PidMap.end())
  {
    fRun2PidMap[run] = 0;
    fRunList->AddEntrySort(TString::Format("%d", run), run);
    fRunList->Layout();

    if (fAutoRun->IsOn())
    {
      StartAutoRunTimer(run);
    }
  }
  else
  {
    Error("StartOfRun", "Run %d already registered.", run);
  }
}

void AliOnlineReco::EndOfRun(Int_t run)
{
  // Slot called from DIM handler on stop of run.

  mIntInt_i i = fRun2PidMap.find(run);
  if (i != fRun2PidMap.end())
  {
    Int_t pid = i->second;
    fRunList->RemoveEntry(run);
    fRunList->Layout();
    fRun2PidMap.erase(i);
    if (pid)
    {
      KillPid(pid);
    }
    gClient->NeedRedraw(fRunList);

    if (fAutoRunRunning == run)
    {
      fAutoRunRunning = 0;
    }
  }
  else
  {
    Error("EndOfRun", "Run %d not registered.", run);
  }
}

//------------------------------------------------------------------------------
// Handlers of OS signals.
//------------------------------------------------------------------------------

void AliOnlineReco::ChildProcTerm(Int_t pid, Int_t status)
{
  // Slot called on termination of child process.
 
  printf("child process termination pid=%d, status=%d...\n", pid, status);

  mIntInt_i i = FindMapEntryByPid(pid);
  if (i != fRun2PidMap.end())
  {
    Int_t run = i->first;
    fRunList->RemoveEntry(run);
    if (status == 0)
    {
      fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED", run), run);
    }
    else
    {
      fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED [%d]", run, status), run);
    }
    fRunList->Layout();
    fRun2PidMap.erase(i);

    if (fAutoRunRunning == run && fAutoRun->IsOn())
    {
      fAutoRunRunning = 0;
      StartAutoRunTimer(run);
    }
    else
    {
      fAutoRunRunning = 0;
    }
  }
  else
  {
    Warning("ChildProcTerm", "Process with pid=%d not registered.", pid);
  }
 
  Emit("ChildProcTerm(Int_t, Int_t)");
}

void AliOnlineReco::ExitLoopChildProcTerm()
{
  if(fDoExit)
    DoExit();
}

//------------------------------------------------------------------------------
// Handlers of button signals.
//------------------------------------------------------------------------------

void AliOnlineReco::DoAutoRun()
{
  // Slot called from auto-run check-box.

  Bool_t autoRun = fAutoRun->IsOn();

  if (autoRun)
    fStartButt->SetEnabled(kFALSE);
  else
    fStartButt->SetEnabled(kTRUE);    
}

void AliOnlineReco::DoStart()
{
  // Slot called from Start button.

  Int_t run = fRunList->GetSelected();
  mIntInt_i i = fRun2PidMap.find(run);

  if (i == fRun2PidMap.end())
  {
    Error("DoStart", "no selection");
    return;
  }

  StartAliEve(i);
}

void AliOnlineReco::DoStop()
{
  // Slot called from Stop button.

  Int_t run = fRunList->GetSelected();
  mIntInt_i i = fRun2PidMap.find(run);

  if (i == fRun2PidMap.end())
  {
    Error("DoStop", "no selection");
    return;
  }

  Int_t pid = i->second;
  if (pid)
  {
    KillPid(pid);
  }
  else
  {
    Error("DoStop", "Process not running.");
  }
}

void AliOnlineReco::DoExit()
{
  // Slot called from Exit button or CloseWindow.
  
  // kill all started processes
  Int_t pid;
  
  // disable all widgets & AutoRunTimer
  // so that user does not initiate other GUI signals
  if(!fDoExit){
    fAutoRun->SetEnabled(kFALSE);
    fStartButt->SetEnabled(kFALSE);
    fStopButt->SetEnabled(kFALSE);
    fExitButt->SetEnabled(kFALSE);
      
    StopAutoRunTimer();
    fDoExit = kTRUE;
    gROOT->SetInterrupt(kTRUE);
  }
  
  gSystem->ProcessEvents();
  
  // clear runs std::map
  for(mIntInt_i i = fRun2PidMap.begin(); i != fRun2PidMap.end(); i++)
  {
    pid = i->second;
    
    if(pid==0)
    {
      fRun2PidMap.erase(i); // if process is not started just remove it from map
    }
    else
    {
      // send kill signal to started process
      KillPid(pid);
      
      // we need to exit loop to let ROOT process events list
      // after kill signal above, process pid starts signal AliChildProcTerminator::ChildProcTerm(int, int)
      // and arrives in AliOnlineReco::ChildProcTerm(int, int)
      // after this we return in DoExit() to process next run
      break;
    }
    
  }
  
  // we can exit after we killed all processes
  if(fRun2PidMap.empty() ) gSystem->ExitLoop();
}

void AliOnlineReco::CloseWindow()
{
  // Virtual method called when window-manager close-window button is pressed.
  
  DoExit();
  
}

Int_t AliOnlineReco::RetrieveGRP(UInt_t run, TString &gdc)
{
  // Retrieve GRP entry for given run from aldaqdb.

  TString dbHost = gSystem->Getenv("ONLINERECO_DB_HOST");
  if (dbHost.IsNull())
  {
    dbHost = "aldaqdb";
  }

  TString dbPort = gSystem->Getenv("ONLINERECO_DB_PORT");
  if (dbPort.IsNull())
  {
    dbPort = "0";
  }

  TString dbName = gSystem->Getenv("ONLINERECO_DB_NAME");
  if (dbName.IsNull())
  {
    dbName = "LOGBOOK";
  }

  TString user = gSystem->Getenv("ONLINERECO_DB_USER");
  if (user.IsNull())
  {
    user = "logbook";
  }

  TString password = gSystem->Getenv("ONLINERECO_DB_PASSWORD");
  if (password.IsNull())
  {
    password = "alice";
  }

  Int_t ret=AliGRPPreprocessor::ReceivePromptRecoParameters(run, dbHost.Data(),
            dbPort.Atoi(), dbName.Data(),
            user.Data(), password.Data(),
            Form("local://%s",gSystem->pwd()),
            gdc);

  if(ret>0) Info("RetrieveGRP","Last run of the same type is: %d",ret);
  else if(ret==0) Warning("RetrieveGRP","No previous run of the same type found");
  else if(ret<0) Error("Retrieve","Error code while retrieving GRP parameters returned: %d",ret);
  return(ret);
}
Commit	Line	Data
c6d78c69	1	// @(#)root/eve:$Id$
	2	// Author: Matevz Tadel 2007
	3
	4	/**************************************************************************
dc836d53	5	* Copyright(c) 1998-2008, ALICE Experiment at CERN, all rights reserved. *)
c6d78c69	6	* See http://aliceinfo.cern.ch/Offline/AliRoot/License.html for *
	7	* full copyright notice. *
	8	**************************************************************************/
	9
	10	#include "AliOnlineReco.h"
93624d6b	11	#include "AliChildProcTerminator.h"
c6d78c69	12	#include "AliDimIntNotifier.h"
a15a9f84	13	#include "AliCDBManager.h"
a15a9f84	14	#include "AliGRPPreprocessor.h"
c6d78c69	15
dc836d53	16	#include <TTimer.h>
dc836d53	17
c6d78c69	18	#include <TGListBox.h>
	19	#include <TGButton.h>
	20
d2a137c1	21	#include <TInterpreter.h>
e849dadd	22	#include <TROOT.h>
d2a137c1	23
c6d78c69	24	#include <unistd.h>
	25	#include <signal.h>
	26
	27	//______________________________________________________________________________
	28	// Full description of AliOnlineReco
	29	//
	30
	31	ClassImp(AliOnlineReco)
	32
	33	AliOnlineReco::AliOnlineReco() :
	34	TGMainFrame(gClient->GetRoot(), 400, 400),
	35
dc836d53	36	fRunList(0), fAutoRun(0), fStartButt(0), fStopButt(0), fExitButt(0),
dc836d53	37	fAutoRunTimer(0), fAutoRunScheduled(0), fAutoRunRunning(0),
4f6eef9f	38	fRun2PidMap(),
e849dadd	39	fTestMode(kFALSE),
e849dadd	40	fDoExit(kFALSE)
c6d78c69	41	{
48c0589a	42	// Constructor.
48c0589a	43
c6d78c69	44	// GUI components.
	45	fRunList = new TGListBox(this);
	46	AddFrame(fRunList, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	47
	48	TGHorizontalFrame *hf = new TGHorizontalFrame(this, 1, 20);
	49
dc836d53	50	fAutoRun = new TGCheckButton(hf, "AutoRun");
	51	hf->AddFrame(fAutoRun, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	52	fAutoRun->Connect("Clicked()", "AliOnlineReco", this, "DoAutoRun()");
	53
c6d78c69	54	fStartButt = new TGTextButton(hf, "Start");
	55	hf->AddFrame(fStartButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	56	fStartButt->Connect("Clicked()", "AliOnlineReco", this, "DoStart()");
	57
	58	fStopButt = new TGTextButton(hf, "Stop");
	59	hf->AddFrame(fStopButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	60	fStopButt->Connect("Clicked()", "AliOnlineReco", this, "DoStop()");
	61
dc836d53	62	fExitButt = new TGTextButton(hf, "Exit");
	63	hf->AddFrame(fExitButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	64	fExitButt->Connect("Clicked()", "AliOnlineReco", this, "DoExit()");
c6d78c69	65
	66	AddFrame(hf, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX));
	67
	68	MapSubwindows();
	69	Layout();
	70	SetWindowName("Alice Online Reconstruction");
	71
e35c7687	72	// DIM interface.
	73	for (Int_t i = 0; i < 5; ++i)
	74	{
	75	if (i == 0)
	76	{
	77	fSOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS");
	78	fEOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS");
	79	}
	80	else
	81	{
	82	fSOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS_%d", i));
	83	fEOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS_%d", i));
	84	}
dc836d53	85
e35c7687	86	fSOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "StartOfRun(Int_t)");
	87	fEOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "EndOfRun(Int_t)");
	88	}
c6d78c69	89
dc836d53	90	const Int_t autoRunDelay = 10; // should go to config
	91	fAutoRunTimer = new TTimer(autoRunDelay * 1000l);
	92	fAutoRunTimer->Connect("Timeout()", "AliOnlineReco", this, "AutoRunTimerTimeout()");
	93
b6269fdc	94	// OS Signal handlers
c6d78c69	95	// ROOT's TSignalHAndler works not SIGCHLD ...
93624d6b	96	AliChildProcTerminator::Instance()->Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ChildProcTerm(Int_t,Int_t)");
e849dadd	97
	98	// we need this by OnExit() to kill next process child after another
	99	Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ExitLoopChildProcTerm()");
c6d78c69	100	}
c6d78c69	101
dc836d53	102	AliOnlineReco::~AliOnlineReco()
dc836d53	103	{
48c0589a	104	// Destructor.
48c0589a	105
dc836d53	106	delete fAutoRunTimer;
	107	}
	108
	109	Int_t AliOnlineReco::GetLastRun() const
	110	{
48c0589a	111	// Returns the last started run.
48c0589a	112
dc836d53	113	return fRun2PidMap.empty() ? 0 : fRun2PidMap.rbegin()->first;
	114	}
	115
	116	Bool_t AliOnlineReco::GetAutoRunMode() const
	117	{
48c0589a	118	// Return state of auto-run flag.
48c0589a	119
dc836d53	120	return fAutoRun->IsOn();
	121	}
	122
	123	void AliOnlineReco::SetAutoRunMode(Bool_t ar)
	124	{
48c0589a	125	// Set auto-run flag.
48c0589a	126
dc836d53	127	if (ar == fAutoRun->IsOn())
	128	return;
	129
	130	fAutoRun->SetState(ar ? kButtonDown : kButtonUp, kTRUE);
	131	}
	132
	133	//------------------------------------------------------------------------------
	134	// Private methods
	135	//------------------------------------------------------------------------------
	136
c6d78c69	137	AliOnlineReco::mIntInt_i AliOnlineReco::FindMapEntryByPid(Int_t pid)
c6d78c69	138	{
48c0589a	139	// Find run-to-pid map iterator by pid.
	140	// Requires iteration over map.
	141
c6d78c69	142	for (mIntInt_i i = fRun2PidMap.begin(); i != fRun2PidMap.end(); ++i)
	143	{
	144	if (i->second == pid)
	145	return i;
	146	}
	147
	148	return fRun2PidMap.end();
	149	}
	150
dc836d53	151	void AliOnlineReco::StartAliEve(mIntInt_i& mi)
dc836d53	152	{
48c0589a	153	// Start alieve to process run given my the run-pid entry.
48c0589a	154
dc836d53	155	Int_t run = mi->first;
	156
	157	if (mi->second == 0)
	158	{
	159	pid_t pid = fork();
	160	if (pid == -1)
	161	{
	162	perror("DoStart -- Fork failed");
	163	return;
	164	}
	165
	166	if (pid)
	167	{
	168	mi->second = pid;
	169	fRunList->RemoveEntry(run);
	170	fRunList->AddEntrySort(TString::Format("%-20d -- RUNNING", run), run);
	171	fRunList->Layout();
	172	}
	173	else
	174	{
d2a137c1	175	gCINTMutex = 0;
	176
	177	struct sigaction sac;
b6269fdc	178	memset(&sac, 0, sizeof(sac));
b6269fdc	179	sac.sa_handler = NULL;
d2a137c1	180	sigemptyset(&sac.sa_mask);
d2a137c1	181	sac.sa_flags = 0;
07f75a04	182
	183	// The sa_restorer field is Not POSIX and obsolete.
	184	// This is for compilation on other systems
	185	#if defined(__linux) && \
	186	(defined(__i386__) \|\| defined(__x86_64__)) && \
	187	defined(__GNUC__)
	188	sac.sa_restorer= NULL;
	189	#endif
b6269fdc	190	sigaction(SIGCHLD, &sac, NULL);
b6269fdc	191
dc836d53	192	int s;
	193	if (fTestMode)
	194	{
b6269fdc	195	s = execlp("alitestproc", "alitestproc", TString::Format("%d", run).Data(), (char*) 0);
dc836d53	196	}
	197	else
	198	{
b6269fdc	199	Int_t procPID = gSystem->GetPid();
	200	TString logFile = Form("%s/reco/log/run%d_%d.log",
	201	gSystem->Getenv("ONLINERECO_BASE_DIR"),
	202	run,
	203	(Int_t)procPID);
	204	Info("DoStart","Reconstruction log will be written to %s",logFile.Data());
	205	gSystem->RedirectOutput(logFile.Data());
	206
	207	gSystem->cd(Form("%s/reco",gSystem->Getenv("ONLINERECO_BASE_DIR")));
	208
	209	TString gdcs;
	210	if (RetrieveGRP(run,gdcs) <= 0 \|\| gdcs.IsNull())
	211	gSystem->Exit(1);
	212
	213	gSystem->Setenv("DATE_RUN_NUMBER", Form("%d", run));
	214	// Setting CDB
	215	// AliCDBManager * man = AliCDBManager::Instance();
	216	// man->SetDefaultStorage("local:///local/cdb");
	217	// man->SetSpecificStorage("GRP/GRP/Data",
	218	// Form("local://%s",gSystem->pwd()));
	219	// man->SetSpecificStorage("GRP/CTP/Config",
	220	// Form("local://%s",gSystem->pwd()));
	221	// man->SetSpecificStorage("ACORDE/Align/Data",
	222	// "local://$ALICE_ROOT/OCDB");
	223
	224	gSystem->mkdir(Form("run%d_%d", run, (Int_t)procPID));
	225	gSystem->cd(Form("run%d_%d", run, (Int_t)procPID));
	226
	227	TString recMacroPath(gSystem->Getenv("ONLINERECO_MACRO"));
	228	if (recMacroPath.IsNull()) {
	229	recMacroPath = "$ALICE_ROOT/MONITOR/rec.C";
	230	}
e849dadd	231
b6269fdc	232	s = execlp("alieve",
	233	"alieve",
	234	"-q",
	235	Form("%s(\"mem://@*:\")", gSystem->ExpandPathName(recMacroPath.Data())),
	236	(char*) 0);
	237
	238	gSystem->Exec(Form("rm -rf %s/reco/run%d_%d",gSystem->Getenv("ONLINERECO_BASE_DIR"),run,(Int_t)procPID));
dc836d53	239	}
	240
	241	if (s == -1)
	242	{
b6269fdc	243	perror("execlp failed - this will not end well");
b6269fdc	244	gSystem->Exit(1);
dc836d53	245	}
	246	}
	247	}
	248	else
	249	{
	250	Error("DoStart", "Process already running.");
	251	}
	252	}
	253
	254	void AliOnlineReco::KillPid(Int_t pid)
	255	{
48c0589a	256	// Terminate process given by pid.
48c0589a	257
dc836d53	258	// Send terminate signal to process ...
	259
	260	if (fTestMode)
	261	{
	262	kill(pid, SIGTERM);
	263	}
	264	else
	265	{
	266	// alieve will auto-destruct on SIGUSR1
	267	kill(pid, SIGUSR1);
	268	}
	269	}
	270
	271	void AliOnlineReco::StartAutoRunTimer(Int_t run)
	272	{
	273	// Start timer for given run.
	274	// If an auto-started run is already active, this call is ignored.
	275	// If timer is already active, it is restarted.
	276
	277	if (fAutoRunRunning)
	278	return;
	279
	280	fAutoRunTimer->Reset();
	281	fAutoRunTimer->TurnOn();
	282	fAutoRunScheduled = run;
	283
	284	Info("StartAutoRunTimer", "Scheduling run %d for auto-display.", run);
	285	}
	286
	287	void AliOnlineReco::StopAutoRunTimer()
	288	{
48c0589a	289	// Stop auto-run timer.
48c0589a	290
dc836d53	291	fAutoRunTimer->TurnOff();
	292	fAutoRunScheduled = 0;
	293	}
	294
	295	void AliOnlineReco::AutoRunTimerTimeout()
	296	{
48c0589a	297	// Slot called on auto-timer timeout.
48c0589a	298
dc836d53	299	Int_t run = fAutoRunScheduled;
	300
	301	StopAutoRunTimer();
	302
	303	mIntInt_i i = fRun2PidMap.find(run);
	304
	305	if (i == fRun2PidMap.end())
	306	{
	307	Warning("AutoRunTimerTimeout", "run no longer active.");
	308	return;
	309	}
	310
	311	Info("AutoRunTimerTimeout", "Starting display for run %d.", run);
	312
	313	StartAliEve(i);
	314	fAutoRunRunning = run;
	315	}
	316
c6d78c69	317	//------------------------------------------------------------------------------
	318	// Handlers of DIM signals.
	319	//------------------------------------------------------------------------------
	320
	321	void AliOnlineReco::StartOfRun(Int_t run)
	322	{
48c0589a	323	// Slot called from DIM handler on start of run.
48c0589a	324
c6d78c69	325	mIntInt_i i = fRun2PidMap.find(run);
	326	if (i == fRun2PidMap.end())
	327	{
	328	fRun2PidMap[run] = 0;
	329	fRunList->AddEntrySort(TString::Format("%d", run), run);
	330	fRunList->Layout();
dc836d53	331
	332	if (fAutoRun->IsOn())
	333	{
	334	StartAutoRunTimer(run);
	335	}
c6d78c69	336	}
	337	else
	338	{
	339	Error("StartOfRun", "Run %d already registered.", run);
	340	}
	341	}
	342
	343	void AliOnlineReco::EndOfRun(Int_t run)
	344	{
48c0589a	345	// Slot called from DIM handler on stop of run.
48c0589a	346
c6d78c69	347	mIntInt_i i = fRun2PidMap.find(run);
	348	if (i != fRun2PidMap.end())
	349	{
	350	Int_t pid = i->second;
	351	fRunList->RemoveEntry(run);
	352	fRunList->Layout();
	353	fRun2PidMap.erase(i);
	354	if (pid)
	355	{
dc836d53	356	KillPid(pid);
c6d78c69	357	}
c6d78c69	358	gClient->NeedRedraw(fRunList);
dc836d53	359
	360	if (fAutoRunRunning == run)
	361	{
	362	fAutoRunRunning = 0;
	363	}
c6d78c69	364	}
	365	else
	366	{
	367	Error("EndOfRun", "Run %d not registered.", run);
	368	}
	369	}
	370
	371	//------------------------------------------------------------------------------
	372	// Handlers of OS signals.
	373	//------------------------------------------------------------------------------
	374
93624d6b	375	void AliOnlineReco::ChildProcTerm(Int_t pid, Int_t status)
c6d78c69	376	{
48c0589a	377	// Slot called on termination of child process.
b6269fdc	378
93624d6b	379	printf("child process termination pid=%d, status=%d...\n", pid, status);
c6d78c69	380
	381	mIntInt_i i = FindMapEntryByPid(pid);
	382	if (i != fRun2PidMap.end())
	383	{
	384	Int_t run = i->first;
	385	fRunList->RemoveEntry(run);
	386	if (status == 0)
	387	{
82142d1d	388	fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED", run), run);
c6d78c69	389	}
	390	else
	391	{
82142d1d	392	fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED [%d]", run, status), run);
c6d78c69	393	}
c6d78c69	394	fRunList->Layout();
e849dadd	395	fRun2PidMap.erase(i);
dc836d53	396
	397	if (fAutoRunRunning == run && fAutoRun->IsOn())
	398	{
	399	fAutoRunRunning = 0;
	400	StartAutoRunTimer(run);
	401	}
	402	else
	403	{
	404	fAutoRunRunning = 0;
	405	}
c6d78c69	406	}
	407	else
	408	{
93624d6b	409	Warning("ChildProcTerm", "Process with pid=%d not registered.", pid);
c6d78c69	410	}
e849dadd	411
	412	Emit("ChildProcTerm(Int_t, Int_t)");
	413	}
	414
	415	void AliOnlineReco::ExitLoopChildProcTerm()
	416	{
	417	if(fDoExit)
	418	DoExit();
c6d78c69	419	}
	420
	421	//------------------------------------------------------------------------------
	422	// Handlers of button signals.
	423	//------------------------------------------------------------------------------
	424
dc836d53	425	void AliOnlineReco::DoAutoRun()
dc836d53	426	{
48c0589a	427	// Slot called from auto-run check-box.
48c0589a	428
dc836d53	429	Bool_t autoRun = fAutoRun->IsOn();
	430
	431	if (autoRun)
	432	fStartButt->SetEnabled(kFALSE);
	433	else
	434	fStartButt->SetEnabled(kTRUE);
	435	}
	436
c6d78c69	437	void AliOnlineReco::DoStart()
c6d78c69	438	{
48c0589a	439	// Slot called from Start button.
48c0589a	440
c6d78c69	441	Int_t run = fRunList->GetSelected();
	442	mIntInt_i i = fRun2PidMap.find(run);
	443
	444	if (i == fRun2PidMap.end())
	445	{
	446	Error("DoStart", "no selection");
	447	return;
	448	}
	449
dc836d53	450	StartAliEve(i);
c6d78c69	451	}
	452
	453	void AliOnlineReco::DoStop()
	454	{
48c0589a	455	// Slot called from Stop button.
48c0589a	456
c6d78c69	457	Int_t run = fRunList->GetSelected();
	458	mIntInt_i i = fRun2PidMap.find(run);
	459
	460	if (i == fRun2PidMap.end())
	461	{
	462	Error("DoStop", "no selection");
	463	return;
	464	}
	465
	466	Int_t pid = i->second;
	467	if (pid)
	468	{
dc836d53	469	KillPid(pid);
c6d78c69	470	}
	471	else
	472	{
	473	Error("DoStop", "Process not running.");
	474	}
	475	}
	476
dc836d53	477	void AliOnlineReco::DoExit()
c6d78c69	478	{
e849dadd	479	// Slot called from Exit button or CloseWindow.
	480
	481	// kill all started processes
	482	Int_t pid;
	483
	484	// disable all widgets & AutoRunTimer
	485	// so that user does not initiate other GUI signals
	486	if(!fDoExit){
	487	fAutoRun->SetEnabled(kFALSE);
	488	fStartButt->SetEnabled(kFALSE);
	489	fStopButt->SetEnabled(kFALSE);
	490	fExitButt->SetEnabled(kFALSE);
	491
	492	StopAutoRunTimer();
	493	fDoExit = kTRUE;
	494	gROOT->SetInterrupt(kTRUE);
	495	}
	496
	497	gSystem->ProcessEvents();
	498
	499	// clear runs std::map
	500	for(mIntInt_i i = fRun2PidMap.begin(); i != fRun2PidMap.end(); i++)
	501	{
	502	pid = i->second;
	503
	504	if(pid==0)
	505	{
	506	fRun2PidMap.erase(i); // if process is not started just remove it from map
	507	}
	508	else
	509	{
	510	// send kill signal to started process
	511	KillPid(pid);
	512
	513	// we need to exit loop to let ROOT process events list
	514	// after kill signal above, process pid starts signal AliChildProcTerminator::ChildProcTerm(int, int)
	515	// and arrives in AliOnlineReco::ChildProcTerm(int, int)
	516	// after this we return in DoExit() to process next run
	517	break;
	518	}
	519
	520	}
	521
	522	// we can exit after we killed all processes
	523	if(fRun2PidMap.empty() ) gSystem->ExitLoop();
c6d78c69	524	}
	525
	526	void AliOnlineReco::CloseWindow()
	527	{
48c0589a	528	// Virtual method called when window-manager close-window button is pressed.
e849dadd	529
	530	DoExit();
	531
c6d78c69	532	}
a15a9f84	533
48c0589a	534	Int_t AliOnlineReco::RetrieveGRP(UInt_t run, TString &gdc)
	535	{
	536	// Retrieve GRP entry for given run from aldaqdb.
a15a9f84	537
8f772d4b	538	TString dbHost = gSystem->Getenv("ONLINERECO_DB_HOST");
	539	if (dbHost.IsNull())
	540	{
	541	dbHost = "aldaqdb";
	542	}
	543
	544	TString dbPort = gSystem->Getenv("ONLINERECO_DB_PORT");
	545	if (dbPort.IsNull())
	546	{
	547	dbPort = "0";
	548	}
	549
	550	TString dbName = gSystem->Getenv("ONLINERECO_DB_NAME");
	551	if (dbName.IsNull())
	552	{
	553	dbName = "LOGBOOK";
	554	}
	555
	556	TString user = gSystem->Getenv("ONLINERECO_DB_USER");
	557	if (user.IsNull())
	558	{
	559	user = "logbook";
	560	}
	561
	562	TString password = gSystem->Getenv("ONLINERECO_DB_PASSWORD");
	563	if (password.IsNull())
	564	{
	565	password = "alice";
	566	}
	567
	568	Int_t ret=AliGRPPreprocessor::ReceivePromptRecoParameters(run, dbHost.Data(),
b6269fdc	569	dbPort.Atoi(), dbName.Data(),
	570	user.Data(), password.Data(),
	571	Form("local://%s",gSystem->pwd()),
	572	gdc);
8f772d4b	573
a15a9f84	574	if(ret>0) Info("RetrieveGRP","Last run of the same type is: %d",ret);
	575	else if(ret==0) Warning("RetrieveGRP","No previous run of the same type found");
	576	else if(ret<0) Error("Retrieve","Error code while retrieving GRP parameters returned: %d",ret);
	577	return(ret);
	578	}