You are here

AliEn plugin for the analysis framework

Content:


Introduction.

The purpose of the plugin is to allow running transparently in AliEn the same user analysis that runs on the local PC or in a PROOF cluster. The practical goal is to do this without having to exit the ROOT prompt. This is achieved currently only to a certain extent: the analysis job is automatically submitted to AliEn and the file merging together with analysis task Terminate() phase executed by the plugin; user intervention is required only for inspecting the job status or resubmitting sub-jobs on demand. The plugin provides the following functionality to hide the complexity of the underlying GRID for users:

  • Allow using all existing types of input data to be processed (root files, raw AliEn collections or XML collections) detecting their type automatically;
  • Generate XML collections corresponding to requested runs and/or data patterns;
  • Automatically detect presence of tags in input data and allow using tag-based cuts in a very simple way;
  • Automatically connect to AliEn (generate a token if needed) - still requires sourcing the environment file produced by alien.token-init;
  • Generate automatically: JDL, analysis macro to be run in grid, execution and validation scripts according a simple to understand user-driven configuration;
  • Copy all needed files in user's AliEn space and submit the job automatically;
  • Start an alien shell to allow inspecting the job status;
  • Do automatic output merging and terminate the analysis tasks.

The plugin is implemented as a base class AliAnalysisGrid inside libANALYSIS providing the API to configure some custom parameters. This can be plugged to the analysis manager class in the same way as the event handlers. Those not yet familiar with ALICE analysis framework shoud read first this. The implementation of the plugin for AliEn is done by AliAnalysisAlien class inside libANALYSISalice library.
One needs  to first configure the plugin before using it. The howto for this and an example are provided below.


Create and configure the plugin.

The simple example below illustrates how the configuration of the plugin can be done via a macro. You can find the example in the file CreateAlienHandler.C. This is the only part of the plugin that requires to be customized.

Example of AliEn plugin configuration

 AliAnalysisGrid* CreateAlienHandler()
{
// Check if user has a valid token, otherwise make one. This has limitations.
// One can always follow the standard procedure of calling alien-token-init then
//   source /tmp/gclient_env_$UID in the current shell.

//   if (!AliAnalysisGrid::CreateToken()) return NULL;
   AliAnalysisAlien *plugin = new AliAnalysisAlien();

// Overwrite all generated files, datasets and output results from a previous session
   plugin->SetOverwriteMode();
// Set the run mode (can be "full", "test", "offline", "submit" or "terminate")
   plugin->SetRunMode("full");  // VERY IMPORTANT - DECRIBED BELOW
// Set versions of used packages
   plugin->SetAPIVersion("V1.1x");
   plugin->SetROOTVersion("v5-26-00b-6");
   plugin->SetAliROOTVersion("v4-19-21-AN");
// Declare input data to be processed.

// Method 1: Create automatically XML collections using alien 'find' command.
// Define production directory LFN

   plugin->SetGridDataDir("/alice/sim/LHC10a6");
// Set data search pattern
//   plugin->SetDataPattern("*ESDs.root");  // simulated, tags not used
//   plugin->SetDataPattern("*ESDs/pass4/*ESDs.root"); // real data check reco pass and data base directory
//   plugin->SetRunPrefix("000");   // real data
   plugin->SetDataPattern("*tag.root");  // Use ESD tags (same applies for AOD's)
// ...then add run numbers to be considered
   plugin->AddRunNumber(125020);    // simulated
//   plugin->AddRunNumber(104065);  // real data

// Method 2: Declare existing data files (raw collections, xml collections, root file)
// If no path mentioned data is supposed to be in the work directory (see SetGridWorkingDir())
// XML collections added via this method can be combined with the first method if
// the content is compatible (using or not tags)

//   plugin->AddDataFile("tag.xml");
//   plugin->AddDataFile("/alice/data/2008/LHC08c/000057657/raw/Run57657.Merged.RAW.tag.root");

// Define alien work directory where all files will be copied. Relative to alien $HOME.
   plugin->SetGridWorkingDir("work");
// Declare alien output directory. Relative to working directory.
   plugin->SetGridOutputDir("output"); // In this case will be $HOME/work/output
// Declare the analysis source files names separated by blancs. To be compiled runtime
// using ACLiC on the worker nodes.

   plugin->SetAnalysisSource("AliAnalysisTaskPt.cxx");
// Declare all libraries (other than the default ones for the framework. These will be
// loaded by the generated analysis macro. Add all extra files (task .cxx/.h) here.

   plugin->SetAdditionalLibs("AliAnalysisTaskPt.h AliAnalysisTaskPt.cxx");
// No need for output file names. Procedure is automatic.
//   plugin->SetOutputFiles("Pt.ESD.1.root");

//   plugin->SetDefaultOutputs();
// No need define the files to be archived. Note that this is handled automatically by the plugin.
//   plugin->SetOutputArchive("log_archive.zip:stdout,stderr");
// Set a name for the generated analysis macro (default MyAnalysis.C) Make this unique !
   plugin->SetAnalysisMacro("AnalysisPt.C");
// Optionally set maximum number of input files/subjob (default 100, put 0 to ignore). The optimum for an analysis

// is correlated with the run time - count few hours TTL per job, not minutes !
   plugin->SetSplitMaxInputFileNumber(100);
// Optionally set number of failed jobs that will trigger killing waiting sub-jobs.
   plugin->SetMaxInitFailed(5);
// Optionally resubmit threshold.
   plugin->SetMasterResubmitThreshold(90);
// Optionally set time to live (default 30000 sec)
   plugin->SetTTL(20000);
// Optionally set input format (default xml-single)
   plugin->SetInputFormat("xml-single");
// Optionally modify the name of the generated JDL (default analysis.jdl)
   plugin->SetJDLName("TaskPt.jdl");
// Optionally modify job price (default 1)
   plugin->SetPrice(1); 
// Optionally modify split mode (default 'se')    
   plugin->SetSplitMode("se");
   return plugin;
}


Use the plugin in your preferred analysis macro.

The only modifications to your macro needed for including the plugin are described below and marked in red. You can download the file runGrid.C together with the task files AliAnalysisTaskPt.h and AliAnalysisTaskPt.cxx

How the plugin can be included in the analysis macro.
 void runGrid()
{
// Load common libraries
  gSystem->Load("libTree.so");
  gSystem->Load("libGeom.so");
  gSystem->Load("libVMC.so");
  gSystem->Load("libPhysics.so");
  gSystem->Load("libSTEERBase");
  gSystem->Load("libESD");
  gSystem->Load("libAOD");
  gSystem->Load("libANALYSIS");
  gSystem->Load("libANALYSISalice");    // The plugin is here

// Use AliRoot includes to compile our task
  gROOT->ProcessLine(".include $ALICE_ROOT/include");

// Create and configure the alien handler plugin
  gROOT->LoadMacro("CreateAlienHandler.C");
  AliAnalysisGrid *alienHandler = CreateAlienHandler();  
  if (!alienHandler) return;

// Create the analysis manager
  AliAnalysisManager *mgr = new AliAnalysisManager("testAnalysis");

// Connect plugin to the analysis manager
  mgr->SetGridHandler(alienHandler);

  gROOT->LoadMacro("AliAnalysisTaskPt.cxx++g");   
  AliAnalysisTask *task = new AliAnalysisTaskPt("TaskPt");
  mgr->AddTask(task);

  AliESDInputHandler* esdH = new AliESDInputHandler();
  esdH->SetInactiveBranches("Calo FMD");
  mgr->SetInputEventHandler(esdH);

// No need to create a chain - this is handled by the plugin
//  gROOT->LoadMacro("$ALICE_ROOT/PWG0/CreateESDChain.C");
//  TChain* chain = CreateESDChain("ESD82XX_30K.txt", 10);

// Create containers for input/output
  AliAnalysisDataContainer *cinput = mgr->CreateContainer("cchain", TChain::Class(), AliAnalysisManager::kInputContainer);
  AliAnalysisDataContainer *coutput = mgr->CreateContainer("chist", TH1::Class(),    AliAnalysisManager::kOutputContainer, "Pt.ESD.1.root");

// Connect input/output
  mgr->ConnectInput(task, 0, cinput);
  mgr->ConnectOutput(task, 0, coutput);

// Enable debug printouts
  mgr->SetDebugLevel(2);

  if (!mgr->InitAnalysis())
    return;

  mgr->PrintStatus();
// Start analysis in grid.
  mgr->StartAnalysis("grid");
};


The execution flow. (plugin->SetRunMode("full"))

After calling mgr->StartAnalysis("grid") the analysis manager will execute task initialization AliAnalysisTask::LocalInit() on the client. Immediately after, the following actions are performed by the AliEn plugin. The description corresponds to the "full" run mode.  

  1. Connect to AliEn using the existing token. Exit if environment not sourced.
  2. Check validity of declared input data directories and files. Check compatibility in case of usage of multiple data sources. Automatically determine if the input data is tag-based or not.
  3. Create XML collections for the requested runs. These will be inserted in AliEn file catalog in the user working directory.
  4. Stream the analysis manager with all configured tasks to the file analysis.root . Copy this file to the AliEn working directory.
  5. Generate the analysis macro to be run in grid and copy this file in the AliEn working directory. This is a standard analysis macro that will execute analysis on worker nodes. It retrieves the analysis manager from the file analysis.root , makes the chain from the collection wn.xml , apply tag-based cuts (if input data contains tags and a macro doing cuts was provided) and initiates the analysis event loop without calling LocalInit() methods for the tasks. Unless special requirements are needed (like usage of .par files), the generated analysis macro does not need to be modified. A way to modify the macro is provided by combining the "offline" and "submit" run modes (see below). A typical example looks like below:
    Automatically generated analysis macro AnalysisPt.C
    const char *anatype = "ESD";

    void AnalysisPt()
    {
    // Analysis using ESD data
    // Automatically generated analysis steering macro executed in grid subjobs

    // load base root libraries

       gSystem->Load("libTree");
       gSystem->Load("libGeom");
       gSystem->Load("libVMC");
       gSystem->Load("libPhysics");

    // load analysis framework libraries
       gSystem->Load("libSTEERBase");
       gSystem->Load("libESD");
       gSystem->Load("libAOD");
       gSystem->Load("libANALYSIS");
       gSystem->Load("libANALYSISalice");

    // add aditional AliRoot libraries below

    // include path (remove if using par files)
       gROOT->ProcessLine(".include $ALICE_ROOT/include");

    // analysis source to be compiled at runtime (if any)
       gROOT->ProcessLine(".L AliAnalysisTaskPt.cxx+g");

    // connect to AliEn and make the chain
       if (!TGrid::Connect("alien://")) return;
       TChain *chain = CreateChainFromTags("wn.xml", anatype);

    // read the analysis manager from file
       TFile *file = TFile::Open("analysis.root");
       if (!file) return;
       TIter nextkey(file->GetListOfKeys());
       AliAnalysisManager *mgr = 0;
       TKey *key;
       while ((key=(TKey*)nextkey())) {
          if (!strcmp(key->GetClassName(), "AliAnalysisManager"))
             mgr = (AliAnalysisManager*)file->Get(key->GetName());
       };
       if (!mgr) {
          ::Error("AnalysisPt", "No analysis manager found in file analysis.root");
          return;
       }

       mgr->PrintStatus();
       mgr->StartAnalysis("localfile", chain);
    }

    TChain* CreateChainFromTags(const char *xmlfile, const char *type="ESD")
    {
    // Create a chain using tags from the xml file.
       TAlienCollection* coll = TAlienCollection::Open(xmlfile);
       if (!coll) {
          ::Error("CreateChainFromTags", "Cannot create an AliEn collection from %s", xmlfile);
          return NULL;
       }
       TGridResult* tagResult = coll->GetGridResult("",kFALSE,kFALSE);
       AliTagAnalysis *tagAna = new AliTagAnalysis(type);
       tagAna->ChainGridTags(tagResult);

       AliRunTagCuts      *runCuts = new AliRunTagCuts();
       AliLHCTagCuts      *lhcCuts = new AliLHCTagCuts();
       AliDetectorTagCuts *detCuts = new AliDetectorTagCuts();
       AliEventTagCuts    *evCuts  = new AliEventTagCuts();
        // Check if the cuts configuration file was provided
       if (!gSystem->AccessPathName("ConfigureCuts.C")) {
          gROOT->LoadMacro("ConfigureCuts.C");
          ConfigureCuts(runCuts, lhcCuts, detCuts, evCuts);
       }
       TChain *chain = tagAna->QueryTags(runCuts, lhcCuts, detCuts, evCuts);
       if (!chain || !chain->GetNtrees()) return NULL;
       chain->ls();
       return chain;
    }

  6. Generate the executable script that launches the analysis macro in batch jobs and copy this in $HOME/bin in AliEn space. Generate the validation script and copy it in the AliEn working directory. The validation script checks if all declared outputs were produced for a given sub-job.
  7. Generate the JDL for the batch analysis and copy it to AliEn. This again does not need to be modified by user unless a need for features not yet supported by the plugin. An example of the JDL produced is below:
    Automatically generated JDL
    JobTag = "Automatically generated analysis JDL";
    # Input xml collections
    InputDataCollection = {
       "LF:/alice/cern.ch/user/m/mgheata/work/300000.xml,nodownload",
       "LF:/alice/cern.ch/user/m/mgheata/work/300001.xml,nodownload"
    };

    # Output directory
    OutputDir = "/alice/cern.ch/user/m/mgheata/work/output/#alien_counter_03i#";

    # List of output files to be registered
    OutputFile = {
       "Pt.ESD.1.root"
    };

    # Packages to be used
    Packages = {
       "VO_ALICE@AliRoot::v4-15-Rev-05",
       "VO_ALICE@ROOT::v5-21-01-alice",
       "VO_ALICE@APISCONFIG::V2.4"
    };

    # List of input files to be uploaded to wn's
    InputFile = {
       "LF:/alice/cern.ch/user/m/mgheata/work/AnalysisPt.C",
       "LF:/alice/cern.ch/user/m/mgheata/work/analysis.root",
       "LF:/alice/cern.ch/user/m/mgheata/work/ConfigureCuts.C",
       "LF:/alice/cern.ch/user/m/mgheata/work/AliAnalysisTaskPt.h",
       "LF:/alice/cern.ch/user/m/mgheata/work/AliAnalysisTaskPt.cxx"
    };

    # This is the startup script
    Executable = "analysis.sh";

    # We split per storage element
    Split = "se";

    # Time to live for the job
    TTL = "30000";

    # Resubmit failed jobs until DONE rate reaches this percentage
    MasterResubmitThreshold = "90%";

    # We want each subjob to get maximum this number of input files
    SplitMaxInputFileNumber = "100";

    # Format of input data
    InputDataListFormat = "xml-single";

    # Collection to be processed on wn
    InputDataList = "wn.xml";

    # Files to be archived
    OutputArchive = {
       "log_archive.zip:stdout,stderr"
    };

    # Maximum number of first failing jobs to abort the master job
    MaxInitFailed = "5";

    # AliEn price for this job
    Price = "1";

    # Validation script to be run for each subjob

    Validationcommand = "/alice/cern.ch/user/m/mgheata/work/validate.sh";

    # AliEn user
    User = "mgheata";

    # JDL variables
    JDLVariables =
    {
       "Packages",
       "OutputDir"
    };

  8. Automatically submit the AliEn job and start an alien shell. The masterjob ID is printed out. The shell is provided for inspecting the job execution in AliEn, but can be interupted any time. There is a simple way to resume analysis termination after all sub-jobs have finished.
  9. After exiting the automatic AliEn shell, the analysis manager tries to merge all output files that were produced and registered in the AliEn output directory. Currently no other check is performed (like how many jobs failed, automatic resubmission or time when the masterjob finished). One can re-run the merging phase by calling SetAnalysisType("terminate") in the plugin and re-running the local analysis macro. Merging is currently done on the local client.
  10. The Terminate() method will be called in case merging succeeded.

The TEST run mode. ( plugin->SetRunMode("test"))

This mode enables testing the generated analysis macro and validation script as they will be executed in batch mode. For this to work, one has to use the input data declared as in Method1 above (base data directory + run number). A sub-sample of 10 files will be generated in xml format and copied to the local directory as wn.xml. Steps 1 to 6 will be performed and the generated script will be invoked in batch mode in a subshell, then the validation script will be run. The Terminate() method of the analysis manager will be however called in graphics mode so that eventual histograms can be visualized.
This mode is recommended to be run systematically before the FULL mode.


The OFFLINE run mode. (plugin->SetRunMode("offline"))

The offline mode will not actually run the generated scripts/macros. These will only be generated so that the user will be able to customize them. A typical use case of this mode is foreseen when running analysis based on .par packages is needed instead of AliRoot-based libraries (default). In this way the user can add manually methods or change the default behavior in the generated analysis macro. If .par files need to be copied in AliEn space this does not need however to be done manually, but better using plugin->SetAdditionalLibs() method. The OFFLINE mode will make steps 1 to 7 from the FULL run mode (except file copying) and is intended to be used together with the SUBMIT mode.


The SUBMIT run mode. (plugin->SetRunMode("submit"))

In this mode the files to be used in the AliEn batch job (generated by the plugin and customized by user in OFFLINE mode) will be copied in AliEn and the job will be automatically submitted in the grid ( steps 8 to 10 ).


The TERMINATE run mode. (plugin->SetRunMode("terminate"))

This mode can be used at any time to re-run steps 9 and 10 described in the FULL mode. Typically used after getting all subjobs in DONE status, it can be used at ANY time after some of the sub-jobs have already registered the output to inspect partial results. Very useful if the analysis task(s) draw some QA histograms in Terminate() phase.


EXAMPLE: Pt analysis

 All files needed to run the Pt example analysis using AliEn plugin can be found in alienplugin.tgz. Just unpack in a folder, modify the plugin configuration and run mode in the file CreateAlienHandler.C then run the macro runGrid.C


Aplying cuts in tag base analysis via the plugin

The plugin detects usage of tags automatically in the input data. The automatic analysis macro that is generated will look for a macro ConfigureCuts.C in the current directory and will execute the method:

Automatic cuts configuration in tag analysis ConfigureCuts.C
 void ConfigureCuts(AliRunTagCuts *runCuts,
                   AliLHCTagCuts *lhcCuts,
                   AliDetectorTagCuts *detCuts,
                   AliEventTagCuts *evCuts)
{
// Configure cuts.
   evCuts->SetMultiplicityRange(10,20);
}