Introduction

This document describes C++ analysis framework for LCLS and how users can make use of its features. Psana design borrows ideas from multitude of other framworks such as pyana, myana, BaBar framework, etc. It's main principles are summarized here:

• support processing of both XTC and HDF5 data format
• user code should be independent of specific data format
• should be easy to use and extend for end users
• support re-use of the existing analysis code
• common simple configuration of user analysis code

This manual is accompanied by the Psana reference Manual which describes interfaces of the classes available in Psana.

Framework Architecture

The central part of the framework is a regular pre-built application (psana) which can dynamically load one or more user analysis modules which are written in C++. The core application is responsible for the following tasks:

• loading and initializing all user modules
• loading one of the input modules to read data from XTC or HDF5
• calling appropriate methods of user modules based on the data being processed
• providing access to data as set of C++ classes
• providing other services such as histogramming to user modules

Other important components of the Psana architecture:

• user module – instance of the C++ class which inherits pre-defined Module class and defines few special methods which are called by framework
• event – special object which transparently stores all event data
• environment – special object which stores non-event data such as configuration objects or EPICS data

Analysis Job Life Cycle

Psana analysis job goes through cycles of state changes such as initialization, configuration, event processing, etc. calling methods of the user modules at every such change. This model follows closely the production activities in LCLS on-line system. DAQ system defines many types of transitions in its data-taking activity, most interesting are here:

• Configure - provides configuration data for complete setup
• BeginRun - start of data taking for one run
• BeginCalibCycle - start of the new scan, some configuration data may change at his point
• L1Accept - this is regular event containing event data from all detectors
• EndCalibCycle - end of single scan
• EndRun - end of data taking for one run
• Unconfigure - stop of all activity

Typically there will be more than one run taken with the same configuration, so there may be more than one BeginRun/EndRun transition for one Configure/Unconfigure, but a data file from single run should contain only one BeginRun/EndRun. Depending on a setup there could be one or more BeginCalibCycle/EndCalibCycle transitions in single run.

For each of the above transitions psana will call corresponding method in user modules notifying them of the possible change in the configuration or just providing event data. Following method names are defined in the user modules:

• beginJob() – this method is called once per analysis job when first Configure transition happens. If there is more than one Configure in single job (when processing multiple runs) this method is not called, use beginRun() to observe configuration changes in this case. This method can access all configuration data through environment object.
• beginRun() – this method is called for every new BeginRun, so it will be called multiple times when processing multiple runs in the same job. This method can access all configuration data through environment object.
• beginCalibCycle() – this method is called for every new BeginCalibCycle, so it will be called multiple times when processing multiple runs in the same job or when single run contains multiple scans. This method can access all configuration data through environment object.
• event() – this method is called for every new L1Accept, it has access to event data through event object as well as configuration data through environment object.
• endCalibCycle() – this method is called for every new EndCalibCycle, it has access to configuration data through environment object.
• endRun() – this method is called for every new EndRun, it has access to configuration data through environment object.
• endJob() – this method is called once at the end of analysis job, it has access to configuration data through environment object.

Typically psana will iterate through all transitions/events from the input files. User modules have a limited control over this event loop, module can request to skip particular event, stop iteration early or abort job using one of the methods described below.

User Modules

User module in psana is an instance of the C++ class which inherits from the Module class (defined in file pasana/Module.h) and implements several methods. These methods are already mentioned above, here is more formal description of each method:

• void beginJob(Env& env)
  Method called once at the beginning of the job. Environment object contains configuration data from the first Configure transition. Default implementation of this method does not do anything.
• void beginRun(Env& env)
  Method called at the beginning of every new run. Default implementation of this method does not do anything.
• void beginCalibCycle(Env& env)
  Method called at the beginning of every new scan. Default implementation of this method does not do anything.
• void event(Event& evt, Env& env)
  Method called for every regular event. Even data is accessible through =evt= argument. There is no default implementation for this method and user module must provide at least this method.
• void endCalibCycle(Env& env)
  Method called at the end of every new scan, can be used to process scan-level statistics collected in event(). Default implementation of this method does not do anything.
• void endRun(Env& env)
  Method called at the end of every run, can be used to process run-level statistics collected in event(). Default implementation of this method does not do anything.
• void endJob(Env& env)
  Method called once at the end of analysis job, can be used to process job-level statistics collected in event(). Default implementation of this method does not do anything.

In addition to event() method every module class must provide a constructor which takes string argument giving the name of the module. Additionally it has to provide a special factory function use to instantiate the modules from the shared libraries, there is special macro defined for definition of this factory function.

Here is the minimal example of the module class declaration with only the event() method implemented and many non-essential details are skipped:

#include "psana/Module.h"

namespace Package {
class ExampleModule: public Module {
public:

  // Constructor takes module name as a parameter
  ExampleModule(const std::string& name);

  // Implementation of event() from base class
  virtual void event(Event& evt, Env& env);

};
} // namespace Package

Definition of the factory function and methods:

#include "Package/ExampleModule.h"
#include "MsgLogger/MsgLogger.h"
#include "PSEvt/EventId.h"

// define factory function
using namespace Package;
PSANA_MODULE_FACTORY(ExampleModule)

// Constructor
ExampleModule::ExampleModule(const std::string& name)
  : Module(name)
{
}

void 
ExampleModule::event(Event& evt, Env& env)
{
  // get event ID
  shared_ptr<EventId> eventId = evt.get();
  if (not eventId.get()) {
    MsgLog(name(), info, "event ID not found");
  } else {
    MsgLog(name(), info, "event ID: " << *eventId);
  }
}

This simple example already does something useful, it retrieves and prints event ID (copied from standard PrintEventId module). Actual modules will do more complex things but this is a simple example of obtaining something from event data.

The easiest way to write new user modules is to use codegen script to generate class from predefined template. This command will create new module ExampleModule in package TestPackage and will copy generated files to the directories in TestPackage:

codegen -l psana-module TestPackage ExampleModule

Data Access in User Modules

As already mentioned above all event data is accessible to user module via Event object, and all non-event data is accessible through Env object. Previous example shows simple use case of extracting data from the event. This section give more detailed description of the Event and Env types and their methods.

When extracting data from event or environment it is necessary to specify at least the type of the data (EventId in the above example). If there are multiple object of the same type in the event then an additional identifying information must be provided – source address and/or additional string key.

Data Source Address

Many pieces of data in the event originate from devices or processes which are parts of the LCLS DAQ. Devices in DAQ system are identified their addresses, which are special C++ data types. There are three types of addresses defined by DAQ:

• DetInfo (class name Pds::DetInfo) – this is the most frequently used type and it defines all regular devices used in DAQ such as cameras, Acqiris, etc. Complete address specification includes 4 items:
  • Detector type, one of the Pds::DetInfo::Detector enum values.
  • Detector ID, a number, in case there is more than one detector of the same type in a system they will have different IDs.
  • Device type, one of the Pds::DetInfo::Device enum values.
  • Device ID, a number, in case there is more than one device of the same type in a system they will have different IDs.
• BldInfo (class name Pds::BldInfo) – this address type is used for Beam Line Data sources, particular source is identified by the Pds::BldInfo::Type enum value.
• ProcInfo (class name Pds::ProcInfo) – this address type is used rarely, and only for information produced by applications constituting DAQ. Sources of this type are identified by IP address of the host where application is running.

(If you look at the C++ code you'll notice that all above classes also include process ID, but it is not used by psana and can be set to 0 if needed.)

User modules should not need to use above C++ classes directly, instead psana provides facility that simplifies specification of the addresses and does not require exact addresses to be known. Class which provides support for these features is called Source (full name is PSEvt::Source). It can be constructed from one of the three above classes, but the most interesting use case is the constructor which accepts string specification of an address. The string specification accept following string formats:

• "DetInfo(Detector.DetID:Device.DevID)"
  Corresponds to DetInfo address type. Detector is the detector name (one of the names of the constants in Pds::DetInfo::Detector enum. DetID is a detector ID number. Device is the device name (one of the names of the constants in Pds::DetInfo::Device enum. DevID is a device ID number. Any or all parts of the specification may be missing. If detector ID or device ID is missing then separating dot is optional. If both device and device ID are missing the separating colon is optional. Missing parts could also be replaces with wildcard '*' symbol.
• "Detector.DetID:Device.DevID"
  Same as the above specification, DetInfo and parentheses can be omitted.
• "Detector-DetID|Device-DevID"
  Same as above, this format is supported for compatibility with pyana but is deprecated.
• "BldInfo(BldType)"
  Corresponds to BldInfo address type. BldType is one of the names of the constants in Pds::BldInfo::Type enum (currently defined types are EBeam, PhaseCavity, FEEGasDetEnergy, Nh2Sb1Ipm01). BldType can be omitted.
• "BldType"
  Same as above, but you cannot omit BldType here.
• "ProcInfo(ipAddr)"
  Corresponds to ProcInfo address type. ipAddr is an IPv4 address in decimal dot notation (123.123.123.123). ipAddr can be omitted.

If the specification includes all pieces then specification is exact and can only match a single data source. If there are missing parts in specification then specification is a match. When requesting data from event with match specification there may be more than one source of data matching it. In this case the first matching source (in unspecified order) will be used. Inexact specification can simplify data access when exact addresses are not known in advance, but one has to be careful if there are multiple devices matching the same address.

Here are few examples of the exact address specifications:

• "DetInfo(AmoITof.0:Acqiris.0)"
• "AmoITof.0:Acqiris.0" – same as above
• "DetInfo(SxrEndstation.0:Opal1000.0)"
• "BldInfo(FEEGasDetEnergy)"
• "FEEGasDetEnergy" – same as above
• "BldInfo(FEEGasDetEnergy)"
• "ProcInfo(0.0.0.0)"

Here are the examples of the address matches:

• "DetInfo(AmoITof.*:Acqiris.*)"
• "DetInfo(AmoITof:Acqiris)" – same as above
• "AmoITof:Acqiris" – same as above
• "DetInfo(AmoITof:*)"
• "DetInfo(AmoITof)" – same as above
• "AmoITof" – same as above
• "DetInfo(*:Acqiris)"
• "DetInfo(:Acqiris)" – same as above
• "*:Acqiris" – same as above
• "DetInfo(*.*:*.*)"
• "DetInfo()" – same as above
• "BldInfo()"
• "" – will match any address type

String Key

Additional key that may be provided when storing or retrieving the data from event is used to distinguish between data objects of the same type and address. As an example the raw data that come from XTC file are stored with the default empty key. User algorithm can apply some algorithm to the data and store new version of the same data using non-default key (such as "fixed" or "calibrated").

Event Data

Event data are accessible through the Event object which is the parameter to event() method of the user module. To access the data on needs to use overloaded get() method which can take different number of arguments. There are three different method signatures:

• get(const std::string& key="", Pds::Src* foundSrc=0)
  This method does not accept data source address argument. It will try to find the data object which was stored without address (such as EventId data which has no corresponding device), otherwise it will return data with any source address.
• get(const Pds::Src& source, const std::string& key="", Pds::Src* foundSrc=0)
  This method takes an exact data source address in the form of Pds::Src class. This method is occasionally useful and its use is explained above.
• get(const Source& source, const std::string& key="", Pds::Src* foundSrc=0)
  This method takes an data source address in the form of Source class which is explained above.

All three above methods take an optional string key which is empty by default. Additionally one can provide a pointer to Pds::Src object as the last argument and the pointed object will be filled with the exact source address of the found object.

All three methods return a special object type that is convertible to a pointer to a specific data type. Thanks to this intermediate special object type the user does not need to provide data type as an argument to get() method which simplifies user code. In fact all important work is done during the conversion of this intermediate object to final pointer, and if this conversion does not happen then get() method does not actually do anything. This implies that the code:

   Pds::Src src;
   evt.get("AmoITof:Acqiris", "", &src);

does not do anything at all and does not update src object. To make it useful one needs to assign the result of get() to a smart pointer:

   Pds::Src src;
   shared_ptr<Psana::Acqiris::DataDescV1> acqData = evt.get("AmoITof:Acqiris", "", &src);

The result of the conversion is a special smart pointer class (boost::shared_ptr) which controls the lifetime of the pointed object. The control is actually shared between Event object and user code, the pointed object will not be destroyed until there is at least one smart pointer for this object. User code can store shared pointer and use the object later, even across multiple events if necessary.

Configuration Data

Access to configuration data happens similarly to event data, except that configuration objects are stored inside environment object. Special configuration storage inside environment can be accessed with the env.configStore() method. Configuration storage object has only two overloaded get() methods:

• get(const Pds::Src& source)
  This method takes an exact data source address in the form of Pds::Src class.
• get(const PSEvt::Source& source, Pds::Src* foundSrc=0)
  This method takes an data source address in the form of Source class which can be exact address or match. Optional second argument can point to an object which will be updated with the exact address of the data if data object is found.

Here is an example of accessing configuration data:

  shared_ptr<Psana::Acqiris::ConfigV1> acqConfig = env.configStore().get("AmoITof:Acqiris");

Matching Event Data to Configuration

In many cases user code is written to use source address match (or approximate address like "AmoITof.:Acqiris."). When it is necessary to find matching configuration object for an event object the approximate addresses (matches) cannot be used because approximate source match can find different devices in event and configuration store. In this case one has to use exact source address (Pds::Src or fully-specified Source) for either both or one of the objects. If exact source address is used for both types of objects one has to pass this address to both get() methods:

// ==== ExampleModule.h ====
class ExampleModule: public Module {
public:
  .....
private:
  Source m_src;
};

// ==== ExampleModule.cpp ====
ExampleModule::ExampleModule(const std::string& name)
  : Module(name)
  , m_src("AmoITof.0:Acqiris.0") // fully-specified source
{
}

void ExampleModule::beginJob(Env& env) {
  shared_ptr<Psana::Acqiris::ConfigV1> acqConfig = env.configStore().get(m_src);
  ......
}

void ExampleModule::event(Event& evt, Env& env) {
  shared_ptr<Psana::Acqiris::DataDescV1> acqData = evt.get(m_src);
  ......
}

If exact source address is not known then one can still use matches for one get() but obtain exact address from the first get() and use it for the second one. There are two possible options here, first is to get exact address from configuration store:

// ==== ExampleModule.h ====
class ExampleModule: public Module {
public:
  .....
private:
  Source m_srcMatch;
  Pds::Src m_src;
};

// ==== ExampleModule.cpp ====
ExampleModule::ExampleModule(const std::string& name)
  : Module(name)
  , m_srcMatch("AmoITof.*:Acqiris.*")  // matching address
  , m_src()
{
}

void ExampleModule::beginJob(Env& env) {
  // use match but obtain exact address
  shared_ptr<Psana::Acqiris::ConfigV1> acqConfig = env.configStore().get(m_srcMatch, &m_src);
  ......
}

void ExampleModule::event(Event& evt, Env& env) {
  // use exact address here
  shared_ptr<Psana::Acqiris::DataDescV1> acqData = evt.get(m_src);
  ......
}

Second option is to use match for getting event object and obtain exact address at the same time, then use exact address to get configuration object:

// ==== ExampleModule.h ====
class ExampleModule: public Module {
public:
  .....
private:
  Source m_srcMatch;
};

// ==== ExampleModule.cpp ====
ExampleModule::ExampleModule(const std::string& name)
  : Module(name)
  , m_srcMatch("AmoITof.*:Acqiris.*")  // matching address
{
}

void ExampleModule::event(Event& evt, Env& env) {
  // use match but obtain exact address
  Pds::Src src;
  shared_ptr<Psana::Acqiris::DataDescV1> acqData = evt.get(m_srcMatch, "", &src);
  if (acqData.get()) {
    // use exact address here
    shared_ptr<Psana::Acqiris::ConfigV1> acqConfig = env.configStore().get(src);
  }
  ......
}

Latter code is less efficient because it searches for configuration object on every event which can be avoided if one uses first option.

Accessing EPICS Data

Access to EPICS data is provided through one more special object in the environments. This object can be accessed trough the cal to env.epicsStore() method which returns reference to the object of PSEnv::EpicsStore class.

It is possible to obtain the full list of PV names using corresponding methods of the EpicsStore object:

  const std::vector<std::string>& pvNames = env.epicsStore().pvNames();

To obtain current value of particular PV the value() method can be used, for example:

  double value = env.epicsStore().value("BEAM:LCLS:ELEC:Q");

or for the array EPICS data:

  double value = env.epicsStore().value("BEAM:LCLS:ELEC:Q", index);

The result returned from value() method can be converted to any numeric type or std::string. The method will throw an exception (which will terminate application if not handled) if the PV name does not exist or if conversion fails.

Status information for particular PV can be obtained with status() method:

  int status, severity;
  PSTime::Time time;
  env.epicsStore().status("BEAM:LCLS:ELEC:Q", status, severity, time);

which returns standard EPICS codes for status and severity plus time of the most recent change of the PV status or value. Time will be set to 0 (UNIX epoch time) when its value is unknown, typically at the beginning of job and may be few first events.

Updating Event Data

User modules can not only read data from event object but also add more data to it. This can be used to exchange information between modules when one module produces some data and another modules use it to calculate their results.

To add data to event one can use Event::put method which accepts smart pointer to the data object and optional source address and string key. There are two overloaded methods in this case:

• void put(const shared_ptr<T>& data, const std::string& key=std::string())
  Adds object to the event without source address, can be used for generic non-device-specific data such as EventId.
• void put(const shared_ptr<T>& data, const Pds::Src& source, const std::string& key=std::string())
  Add object and specify its source address, should be used for detector/device-specific data.

Both methods take optional string key which should be used to distinguish different "versions" of the same data such as data after calibration.

Here is an example code which adds one new object:

    shared_ptr<Image> img(new Image(...));
    evt.put(img, src, "filtered");

Job and Module Configuration

Psana framework has multiple configuration options that can be change via command line or special configuration file. Configuration file can also specify options for user modules so that modules' behavior can be changed at run time without the need to recompile the code.

If no options are specified on the command line then psana tries to read configuration file named psana.cfg from the current directory if the file exists. The location of the configuration file can be changed with the -c <path> option which should provide path of the configuration file.

Configuration File Format

Configuration file has a simple format which is similar to well-known INI file format. The file consists of the sections, each section begins with the section header in the form:

[<section-name>]

Section names can be arbitrary strings, but in psana case section names are the names of the modules which cannot be arbitrary and should not contain speces.

Following the section header there may be zero or more option lines in the form

<option-name> = <option-value>

Option name is anything between beginning of line and '=' character with leading and trailing spaces and tabs stripped. Option value is anything after '=' character with leading and trailing spaces and tabs stripped, option value can be empty. Long option value can be split over multiple lines if the line ends with the backslash character, e.g.:

files = /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s00-c00.xtc \
        /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s01-c00.xtc \
        /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s02-c00.xtc

Lines starting with '#' character are considered comments and ignored.

Option Types

Configuration file does not specify option types, all values in the file are strings. Psana framework provides conversion of these strings to several basic C++ types or sequences. Following types and conversion rules are supported by framework:

• bool
  value strings "yes", "true", "on" become true, "no", "false", "off" become false. Strings which represent non-zero numbers become true, string "0" becomes false.
• char
  value string must be single-character string and it will be assigned to a result.
• C++ numeric types
  option value must represent valid number.
• std::string
  option value will be assigned to result string without change.
• C++ sequence types (e.g. std::list<T>)
  option value will be split into single words at space/tab characters, individual words will be converted to resulting type T.

When the conversion fails because of the incorrectly formatted input framework will throw an exception with the type ExceptionCvtFail.

Psana Options

The options that are needed for the framework are defined in psana section. Here is the list of options which can appear in that section:

• modules
  list of module names to include in the analysis job. Each module name is built of a package name and class name separated by dot (e.g. TestPackage.ExampleModule) optionally followed by colon and modifier. Modifier is not needed if there is only one instance of the module in the job. If there is more than on instance then modules need to include unique modifier to distinguish instances. If the module comes from psana package then package name can be omitted.
• files
  list of file names to process. file names can also be specified on the command line which will override anything specified in configuration file.
• events
  maximum number of events to process in a job.

Here is an example of the framework configuration section:

[psana]
# list of file names
files = /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s00-c00.xtc \
        /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s01-c00.xtc \
        /reg/d/psdm/AMO/amo00000/xtc/e00-r0000-s02-c00.xtc
# list of modules, PrintSeparator and PrintEventId are from psana package
# and do not need package name
modules = PrintSeparator PrintEventId psana_examples.DumpAcqiris

User Modules Options

Options for user modules appear in the separate sections named after the modules. For example the module with name "TestPackage.ExampleModule" will read its options from the section [TestPackage.ExampleModule]. If the module name includes modifier after colon then it will try to find option value in the corresponding section first and if it does not exist there it will try to read option form section which does not have qualifier. In this way the modules can share common options. For example the module "TestPackage.ExampleModule:test" will try to read an option from [TestPackage.ExampleModule:test] section first and section [TestPackage.ExampleModule] after that.

Here is an example of configuration for some fictional analysis job:

[psana]
modules = TestPackage.Analysis:mode1 TestPackage.Analysis:mode2

[TestPackage.Analysis]
# these are common options for all TestPackage.Analysis modules,
# but instances can override then in their own sections
calib-mode = fancy
subpixel = off
threshold = 0.001

[TestPackage.Analysis:mode1]
# options specific to mode1 module
range-min = 0
range-max = 1000000

[TestPackage.Analysis:mode2]
# options specific to mode2 module
range-min = 1000
range-min = 10000
subpixel = on