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

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

//______________________________________________________________________________
// Full description of AliOnlineReco
//

ClassImp(AliOnlineReco)

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

  fRunList(0), fStartButt(0), fStopButt(0), fXyzzButt(0),

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

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

  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()");

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

  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)");
  }

  // Signal handlers
  // ROOT's TSignalHAndler works not SIGCHLD ...
  AliChildProcTerminator::Instance()->Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ChildProcTerm(Int_t,Int_t)");
}

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

  return fRun2PidMap.end();
}

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

void AliOnlineReco::StartOfRun(Int_t run)
{
  mIntInt_i i = fRun2PidMap.find(run);
  if (i == fRun2PidMap.end())
  {
    fRun2PidMap[run] = 0;
    fRunList->AddEntrySort(TString::Format("%d", run), run);
    fRunList->Layout();
  }
  else
  {
    Error("StartOfRun", "Run %d already registered.", run);
  }
}

void AliOnlineReco::EndOfRun(Int_t 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)
    {
      // Send terminate signal to process ...
      if (fTestMode)
      {
	kill(pid, SIGTERM);
      }
      else
      {
	// alieve will auto-destruct on SIGUSR1
	kill(pid, SIGUSR1);
      }
    }
    gClient->NeedRedraw(fRunList);
  }
  else
  {
    Error("EndOfRun", "Run %d not registered.", run);
  }
}

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

void AliOnlineReco::ChildProcTerm(Int_t pid, Int_t status)
{
  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();
    i->second = 0;
  }
  else
  {
    Warning("ChildProcTerm", "Process with pid=%d not registered.", pid);
  }
}

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

void AliOnlineReco::DoStart()
{
  Int_t run = fRunList->GetSelected();
  mIntInt_i i = fRun2PidMap.find(run);

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

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

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

	const char *recMacroPath = "$ALICE_ROOT/test/cosmic/rec.C";

	s = execlp("alieve",
		   "alieve",
		   "-q",
		   Form("%s(\"mem://@*:\")",gSystem->ExpandPathName(recMacroPath)),
		   (char*) 0);
      }

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

void AliOnlineReco::DoStop()
{
  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)
  {
    if (fTestMode)
    {
      kill(pid, SIGTERM);
    }
    else
    {
      // alieve will auto-destruct on SIGUSR1
      kill(pid, SIGUSR1);
    }
  }
  else
  {
    Error("DoStop", "Process not running.");
  }
}

void AliOnlineReco::DoXyzz()
{
  gSystem->ExitLoop();
}

void AliOnlineReco::CloseWindow()
{
  gSystem->ExitLoop();
}

Int_t AliOnlineReco::RetrieveGRP(UInt_t run, TString &gdc) {

  Int_t ret=AliGRPPreprocessor::ReceivePromptRecoParameters(run, "aldaqdb", 0, "LOGBOOK", "logbook", "alice",
							    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	/**************************************************************************
	5	* Copyright(c) 1998-2008, ALICE Experiment at CERN, all rights reserved. *
	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
	16	#include <TGListBox.h>
	17	#include <TGButton.h>
	18
	19	#include <unistd.h>
	20	#include <signal.h>
	21
	22	//______________________________________________________________________________
	23	// Full description of AliOnlineReco
	24	//
	25
	26	ClassImp(AliOnlineReco)
	27
	28	AliOnlineReco::AliOnlineReco() :
	29	TGMainFrame(gClient->GetRoot(), 400, 400),
	30
	31	fRunList(0), fStartButt(0), fStopButt(0), fXyzzButt(0),
	32
c6d78c69	33	fTestMode(kFALSE)
	34	{
	35	// GUI components.
	36	fRunList = new TGListBox(this);
	37	AddFrame(fRunList, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	38
	39	TGHorizontalFrame *hf = new TGHorizontalFrame(this, 1, 20);
	40
	41	fStartButt = new TGTextButton(hf, "Start");
	42	hf->AddFrame(fStartButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	43	fStartButt->Connect("Clicked()", "AliOnlineReco", this, "DoStart()");
	44
	45	fStopButt = new TGTextButton(hf, "Stop");
	46	hf->AddFrame(fStopButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	47	fStopButt->Connect("Clicked()", "AliOnlineReco", this, "DoStop()");
	48
	49	fXyzzButt = new TGTextButton(hf, "Exit");
	50	hf->AddFrame(fXyzzButt, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX \| kLHintsExpandY));
	51	fXyzzButt->Connect("Clicked()", "AliOnlineReco", this, "DoXyzz()");
	52
	53	AddFrame(hf, new TGLayoutHints(kLHintsNormal \| kLHintsExpandX));
	54
	55	MapSubwindows();
	56	Layout();
	57	SetWindowName("Alice Online Reconstruction");
	58
e35c7687	59	// DIM interface.
	60	for (Int_t i = 0; i < 5; ++i)
	61	{
	62	if (i == 0)
	63	{
	64	fSOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS");
	65	fEOR[i] = new AliDimIntNotifier("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS");
	66	}
	67	else
	68	{
	69	fSOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_SOR_PHYSICS_%d", i));
	70	fEOR[i] = new AliDimIntNotifier(Form("/LOGBOOK/SUBSCRIBE/DAQ_EOR_PHYSICS_%d", i));
	71	}
	72	fSOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "StartOfRun(Int_t)");
	73	fEOR[i]->Connect("DimMessage(Int_t)", "AliOnlineReco", this, "EndOfRun(Int_t)");
	74	}
c6d78c69	75
	76	// Signal handlers
	77	// ROOT's TSignalHAndler works not SIGCHLD ...
93624d6b	78	AliChildProcTerminator::Instance()->Connect("ChildProcTerm(Int_t,Int_t)", "AliOnlineReco", this, "ChildProcTerm(Int_t,Int_t)");
c6d78c69	79	}
	80
	81	AliOnlineReco::mIntInt_i AliOnlineReco::FindMapEntryByPid(Int_t pid)
	82	{
	83	for (mIntInt_i i = fRun2PidMap.begin(); i != fRun2PidMap.end(); ++i)
	84	{
	85	if (i->second == pid)
	86	return i;
	87	}
	88
	89	return fRun2PidMap.end();
	90	}
	91
	92	//------------------------------------------------------------------------------
	93	// Handlers of DIM signals.
	94	//------------------------------------------------------------------------------
	95
	96	void AliOnlineReco::StartOfRun(Int_t run)
	97	{
	98	mIntInt_i i = fRun2PidMap.find(run);
	99	if (i == fRun2PidMap.end())
	100	{
	101	fRun2PidMap[run] = 0;
	102	fRunList->AddEntrySort(TString::Format("%d", run), run);
	103	fRunList->Layout();
	104	}
	105	else
	106	{
	107	Error("StartOfRun", "Run %d already registered.", run);
	108	}
	109	}
	110
	111	void AliOnlineReco::EndOfRun(Int_t run)
	112	{
	113	mIntInt_i i = fRun2PidMap.find(run);
	114	if (i != fRun2PidMap.end())
	115	{
	116	Int_t pid = i->second;
	117	fRunList->RemoveEntry(run);
	118	fRunList->Layout();
	119	fRun2PidMap.erase(i);
	120	if (pid)
	121	{
	122	// Send terminate signal to process ...
	123	if (fTestMode)
	124	{
	125	kill(pid, SIGTERM);
	126	}
	127	else
	128	{
	129	// alieve will auto-destruct on SIGUSR1
	130	kill(pid, SIGUSR1);
	131	}
	132	}
	133	gClient->NeedRedraw(fRunList);
	134	}
	135	else
	136	{
	137	Error("EndOfRun", "Run %d not registered.", run);
	138	}
	139	}
	140
	141	//------------------------------------------------------------------------------
	142	// Handlers of OS signals.
143	//------------------------------------------------------------------------------
144
93624d6b	145	void AliOnlineReco::ChildProcTerm(Int_t pid, Int_t status)
c6d78c69	146	{
93624d6b	147	printf("child process termination pid=%d, status=%d...\n", pid, status);
c6d78c69	148
	149	mIntInt_i i = FindMapEntryByPid(pid);
	150	if (i != fRun2PidMap.end())
	151	{
	152	Int_t run = i->first;
	153	fRunList->RemoveEntry(run);
	154	if (status == 0)
	155	{
82142d1d	156	fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED", run), run);
c6d78c69	157	}
	158	else
	159	{
82142d1d	160	fRunList->AddEntrySort(TString::Format("%-20d -- PROCESSED [%d]", run, status), run);
c6d78c69	161	}
	162	fRunList->Layout();
	163	i->second = 0;
	164	}
	165	else
	166	{
93624d6b	167	Warning("ChildProcTerm", "Process with pid=%d not registered.", pid);
c6d78c69	168	}
	169	}
	170
	171	//------------------------------------------------------------------------------
	172	// Handlers of button signals.
	173	//------------------------------------------------------------------------------
	174
	175	void AliOnlineReco::DoStart()
	176	{
	177	Int_t run = fRunList->GetSelected();
	178	mIntInt_i i = fRun2PidMap.find(run);
	179
	180	if (i == fRun2PidMap.end())
	181	{
	182	Error("DoStart", "no selection");
	183	return;
	184	}
	185
	186	if (i->second == 0)
	187	{
	188	pid_t pid = fork();
	189	if (pid == -1)
	190	{
	191	perror("DoStart -- Fork failed");
	192	return;
	193	}
	194
	195	if (pid)
	196	{
	197	i->second = pid;
	198	fRunList->RemoveEntry(run);
	199	fRunList->AddEntrySort(TString::Format("%-20d -- RUNNING", run), run);
	200	fRunList->Layout();
	201	}
	202	else
	203	{
	204	int s;
	205	if (fTestMode)
	206	{
	207	s = execlp("alitestproc", "alitestproc", TString::Format("%d", run).Data(), (char*) 0);
	208	}
	209	else
	210	{
a15a9f84	211	Int_t procPID = gSystem->GetPid();
	212	TString logFile = Form("%s/reco/log/run%d_%d.log",
	213	gSystem->Getenv("ONLINERECO_BASE_DIR"),
	214	run,
	215	(Int_t)procPID);
	216	Info("DoStart","Reconstruction log will be written to %s",logFile.Data());
	217	gSystem->RedirectOutput(logFile.Data());
	218
	219	gSystem->cd(Form("%s/reco",gSystem->Getenv("ONLINERECO_BASE_DIR")));
	220
	221	TString gdcs;
	222	if ((RetrieveGRP(run,gdcs) <= 0) \|\|
	223	gdcs.IsNull())
	224	gSystem->Exit(1);
	225
	226	gSystem->Setenv("DATE_RUN_NUMBER",Form("%d",run));
	227	// Setting CDB
	228	// AliCDBManager * man = AliCDBManager::Instance();
	229	// man->SetDefaultStorage("local:///local/cdb");
	230	// man->SetSpecificStorage("GRP/GRP/Data",
	231	// Form("local://%s",gSystem->pwd()));
	232	// man->SetSpecificStorage("GRP/CTP/Config",
	233	// Form("local://%s",gSystem->pwd()));
	234	// man->SetSpecificStorage("ACORDE/Align/Data",
	235	// "local://$ALICE_ROOT/OCDB");
	236
	237	gSystem->mkdir(Form("run%d_%d",run,(Int_t)procPID));
	238	gSystem->cd(Form("run%d_%d",run,(Int_t)procPID));
	239
	240	const char *recMacroPath = "$ALICE_ROOT/test/cosmic/rec.C";
	241
	242	s = execlp("alieve",
	243	"alieve",
	244	"-q",
	245	Form("%s(\"mem://@*:\")",gSystem->ExpandPathName(recMacroPath)),
	246	(char*) 0);
c6d78c69	247	}
	248
	249	if (s == -1)
	250	{
	251	perror("execlp failed - this will not end well");
	252	gSystem->Exit(1);
	253	}
	254	}
	255	}
	256	else
	257	{
	258	Error("DoStart", "Process already running.");
	259	}
	260	}
	261
	262	void AliOnlineReco::DoStop()
	263	{
	264	Int_t run = fRunList->GetSelected();
	265	mIntInt_i i = fRun2PidMap.find(run);
	266
	267	if (i == fRun2PidMap.end())
	268	{
	269	Error("DoStop", "no selection");
	270	return;
	271	}
	272
	273	Int_t pid = i->second;
	274	if (pid)
	275	{
	276	if (fTestMode)
	277	{
	278	kill(pid, SIGTERM);
	279	}
	280	else
	281	{
	282	// alieve will auto-destruct on SIGUSR1
	283	kill(pid, SIGUSR1);
	284	}
	285	}
	286	else
	287	{
	288	Error("DoStop", "Process not running.");
	289	}
	290	}
	291
	292	void AliOnlineReco::DoXyzz()
	293	{
	294	gSystem->ExitLoop();
	295	}
	296
	297	void AliOnlineReco::CloseWindow()
	298	{
	299	gSystem->ExitLoop();
	300	}
a15a9f84	301
	302	Int_t AliOnlineReco::RetrieveGRP(UInt_t run, TString &gdc) {
	303
	304	Int_t ret=AliGRPPreprocessor::ReceivePromptRecoParameters(run, "aldaqdb", 0, "LOGBOOK", "logbook", "alice",
	305	Form("local://%s",gSystem->pwd()),
	306	gdc);
	307	if(ret>0) Info("RetrieveGRP","Last run of the same type is: %d",ret);
	308	else if(ret==0) Warning("RetrieveGRP","No previous run of the same type found");
	309	else if(ret<0) Error("Retrieve","Error code while retrieving GRP parameters returned: %d",ret);
	310	return(ret);
	311	}