pvDatabaseCPP

EPICS v4 Working Group, Working Draft, 19-Feb-2014

Latest version:
pvDatabaseCPP.html
This version:
pvDatabaseCPP20140219.html
Previous version:
pvDatabaseCPP20140207.html
Editors:
Marty Kraimer, BNL

Abstract

This document describes pvDatabaseCPP, which is a framework for implementing a network accessable database of smart memory resident records. Network access is via pvAccess. The data in each record is a top level PVStructure as defined by pvData. The framework includes a complete implementation of ChannelProvider as defined by pvAccess. The framework can be extended in order to create record instances that implements services. The minimum that an extenson must provide is a top level PVStructure and a process method.

EPICS version 4 is a set of related products in the EPICS V4 control system programming environment:
relatedDocumentsV4.html

Status of this Document

This is the 19-Feb-2014 version of of pvDatabaseCPP.

Since the last version of the documentation:

examples
The examples have been moved to separate top level build areas.
test
The regression tests have been moved to a separate top level build area. It is built from the top but nothing from the tests appears in the top level bin directory.
exampleServer
This example now also includes pvaSrv, i. e. the pvAccess server for interfacing to iocCore V3 records.

This project is ready for alpha users.

I have not had time to look at two unresolved problems reported in the previous version of this document:

memory leak
arrayPerformanceMain shows a slight memory leak at termination.
channel destroy and recreate
longArrayGet and longArrayPut fail if the channel is destroyed and immediately recreated. If epicsThreadSleep(1.0) is called between destroy and recreate then they work. The current version of each does wait.

Future enhancements in priority order:

channelArray
The arguments that have type int should be changed to size_t. This requires changes to pvAccessCPP.
Monitor Algorithms
Monitor algorithms have not been implemented. Thus all monitors are onPut.
Create more regression tests
Currently only some simple tests exist. Most of the testing has been via the examples.

Table of Contents

Introduction

Overview

The main purpose of this project to make it easier to implement services that are accessed via pvAccess. This project supplies is a complete implementation of the server side of pvAccess. All that a service has to provide is a top level PVStructure and a process method. A service can be run as a main process or can be part of a V3 IOC. Thus services can be developed that interact with V3 records, asynDriver, areaDetector, etc.

A brief description of a pvDatabase is that it is a set of network accessible, smart, memory resident records. Each record has data composed of a top level PVStructure. Each record has a name which is the channelName for pvAccess. A local Channel Provider implements the complete ChannelProvider and Channel interfaces as defined by pvAccess. The local provider provides access to the records in the pvDatabase. This local provider is accessed by the remote pvAccess server. A record is smart because code can be attached to a record, which is accessed via a method named process.

This document describes components that provide the following features:

database
This encapsulates the concept of a database of memory resident smart records. The two main components are:
pvRecord
This encapsulates the concept of a smart record. It can be processed. Changes to field values can be trapped. A record can be locked.
pvDatabase
This is a database of pvRecords. Records can be added and removed from a database.
pvAccess
This is a complete implementation of ChannelProvider and Channel as defined by pvAccess. It is used by the server side of pvAccess to attach to pvRecords. This component also provides a C++ implementation of the monitor and pvCopy components from pvIOCJava
Main and V3IOC
The pvDatabase can be provided via a Main program or can be part of a V3IOC. In the later case the IOC has both a database of V3 Records and a pvDatabase.

Base classes make it easy to create record instances. The code attached to each record must create the top level PVStructure and the following three methods:

init
This is a method for initializing the support. It returns true if successful and false otherwise.
process
This is what makes a record smart.
destroy
This releases and resources used by the implementation.

Getting started

The first step is to build pvDatabaseCPP as described in the next section.

One of the examples is exampleServer. It can be started either via a main program or as part of a V3 IOC.

To start it as a main program do the following:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleServer
mrk> bin/linux-x86_64/exampleServerMain 

You should see something like the following:

result of addRecord exampleServer 1
VERSION : pvAccess Server v3.0.5-SNAPSHOT
PROVIDER_NAMES : local
BEACON_ADDR_LIST : 
AUTO_BEACON_ADDR_LIST : 1
BEACON_PERIOD : 15
BROADCAST_PORT : 5076
SERVER_PORT : 5075
RCV_BUFFER_SIZE : 16384
IGNORE_ADDR_LIST: 
STATE : INITIALIZED
exampleServer
Type exit to stop: 

Then in another window execute a pvput and pvget as follows:

mrk> pvput -r "field(argument.value)" exampleServer World
...
mrk> pvget -r "record[process=true]field(result.value)" exampleServer
exampleServer
structure 
    string value Hello World
mrk> 

To run the example as part of a V3 IOC do the following:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleServer/iocBoot/exampleServer
mrk> ../../bin/linux-x86_64/exampleServer st.cmd 

You will see the following:

> envPaths
epicsEnvSet("ARCH","linux-x86_64")
epicsEnvSet("IOC","exampleServer")
epicsEnvSet("TOP","/home/hg/pvDatabaseCPP/exampleServer")
epicsEnvSet("EPICS_BASE","/home/install/epics/base")
epicsEnvSet("EPICSV4HOME","/home/hg")
cd /home/hg/pvDatabaseCPP/exampleServer
## Register all support components
dbLoadDatabase("dbd/exampleServer.dbd")
exampleServer_registerRecordDeviceDriver(pdbbase)
## Load record instances
dbLoadRecords("db/dbScalar.db","name=pvdouble,type=ao")
dbLoadRecords("db/dbArray.db","name=pvdoubleArray,type=DOUBLE")
dbLoadRecords("db/dbStringArray.db","name=pvstringArray")
dbLoadRecords("db/dbEnum.db","name=pvenum")
dbLoadRecords("db/dbCounter.db","name=pvcounter");
cd /home/hg/pvDatabaseCPP/exampleServer/iocBoot/exampleServer
iocInit()
Starting iocInit
############################################################################
## EPICS R3.14.12.3 $Date: Mon 2012-12-17 14:11:47 -0600$
## EPICS Base built Dec 21 2013
############################################################################
iocRun: All initialization complete
dbl
pvdouble
pvcounter
pvenum
pvdoubleArray
pvstringArray
epicsThreadSleep(1.0)
exampleServerCreateRecord pvaServer
startPVAServer
VERSION : pvAccess Server v3.0.5-SNAPSHOT
PROVIDER_NAMES : dbPv local
BEACON_ADDR_LIST : 
AUTO_BEACON_ADDR_LIST : 1
BEACON_PERIOD : 15
BROADCAST_PORT : 5076
SERVER_PORT : 5075
RCV_BUFFER_SIZE : 16384
IGNORE_ADDR_LIST: 
STATE : INITIALIZED
pvdbl
pvaServer
epics> 

Just like previously you can then execute a pvput and pvget and see Hello World.

The examples, i. e. exampleServer, exampleLink, examplePowerSupply, and exampleDatabase, are described in separate sections below. In addition arrayPerformance can be used to measure that performance of big arrays. It is also described in a later section.

Reading section exampleServer and looking at it's code is a good way to learn how to implement a service.

Relationship with pvIOCJava.

This document descibes a C++ implementation of some of the components in pvIOCJava, which also implements a pvDatabase. PVDatabaseCPP implements the core components required to create a network accessible database of smart memory resident records. pvDatabaseCPP does not implement any of the specialized support that pvIOCJava provides. It is expected that many services will be created that do not require the full features provided by pvIOCJava. In the future pvIOCJava should be split into multiple projects with one of them named pvDatabaseJava.

Similar to epics base, pvIOCJava implements the concept of synchronous and asynchronous record processing. For pvDatabaseCPP the process method is allowed to block. Until a need is demonstrated this will remain true. The main user of a pvDatabase is pvAccess, and in particular, remote pvAccess. The server side of remote pvAccess creates two threads for each client and always accesses a record via these threads. It is expected that these threads will be sufficient to efficently handle all channel methods except channelRPC. For channelRPC pvAccess provides (or will provide) a thread pool for channelRPC requests. If, in the future, a scanning facility is provided by pvDatabaseCPP or some other facility, then the scanning facility will have to provide some way of handling process requests that block.

Phased Development

This documentation describes the first phase of a phased implementation of pvDatabaseCPP:

pvRecord
Wrapper on PVStructure that implements methods required by Local Channel Provider.
pvDatabase
Database of PVRecords. Has methods find, add, and remove.
Local Channel Provider
Complete implementation of ChannelProvider and Channel. This means that pvCopy and monitor are also implemented.

Future phases of pvDatabaseCPP might include:

Install
This provides complete support for on-line add and delete of sets of records. With the first phase each "service" is responsible for it's own implementation. All that is provided is addRecord and removeRecord.
Field support
Add the ability to optionally add support to fields. In addition some of the basic support defined in pvIOCJava could also be implemented.
XML parser
This provides the ability to create record instances without writing any code.

The completion of each phase provides useful features that can be used without waiting for the completion of later phases. The rest of this document discusses only the first phase.

Features Required for localChannelProvider

pvCopy
Creates a PVStructure that contains a copy of an arbitary subset of the fields of another top level PVStructure. It can copy data between the two and maintains a bitSet that show which fields are changed.
monitor
This provides the ability to monitor changes to fields of a record.
PVRecord and PVDatabase
Defined below.
The localChannelProvider itself
This is the C++ implementation of package pvAccess in pvIOCJava. The localChannelProvider accesses data from PVRecords. It implements all channel methods except channelRPC, which is implemented by pvAccessCPP.

Minumum Features Required for pvRecord

The first phase will only implement record processing, i. e. the process method has to do everything itself without any generic field support. This will be sufficient for implementing many services. The following are the minimium features required

PVDatabase
This holds a set of PVRecords. It has methods to find, add, and remove records.
PVRecord
This, and a set of related interfaces, provides the following:
Access to top level PVStructure
PVRecord is a wrapper on a top level pvStructure.
Record locking
A record can be locked and unlocked. A record must be locked whenever data in the pvStructure is accessed.
Trapping data changes
A client can request to be notified when data in the pvStructure is modified. It can do this on a field by field basis.

The following sections describes the classes required for the first phase.

Building pvDatabaseCPP

To build pvDatabaseCPP You must provide a file RELEASE.local in directory configure. Thus do the following:

mrk> pwd
/home/hg/pvDatabaseCPP/configure
mrk> cp ExampleRELEASE.local RELEASE.local

Then edit RELEASE.local so that it has the correct location of each product pvDatabaseCPP requires. Than at the top level just execute make:

mrk> cd ..
mrk> pwd
/home/hg/pvDatabaseCPP
mrk> make

This builds pvDatabaseCPP and also the tests and all examples.

Each example and arrayPerformance is a completely separate top, but is also built when make is run in pvDatabaseCPP itself.

Each is a separate top for the following reasons:

  1. It is easier to understand each example including how it is built so that it can be run as a main or as part of a V3 IOC.
  2. Each example can be copied somewhere else and used as the basis for creating a new service.

If it is desired to build an example all by itself, just follow the same instructions as for building pvDatabaseCPP itself. For example:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleServer/configure
mrk> cp ExampleRELEASE.local RELEASE.local

Then edit RELEASE.local so that it has the correct location of each product the example requires. Than at the top level of the example just execute make:

mrk> cd ..
mrk> pwd
/home/hg/pvDatabaseCPP/exampleServer
mrk> make

This builds the example.

iocshell commands

The following iocsh commands are provided for a V3IOC:

startPVAClient
Starts the client side of pvAccess.s It makes channel provider pvAccess available. After startPVAServer is called the channel provider local will also be available.
stopPVAClient
Stops pvAccess.
startPVAServer
Starts the local channel provider

stopPVAServer
Stop the local channel provider
pvdbl
Provides a list of all the pvRecords in database master

The client commands are provided via PVAClientRegister.dbd and the other commands via PVAServerRegister.dbd.

In addition any code that implements a PVRecord must implement an ioc command. The directory example has examples of how to implement the registration code. See example/V3IOC/exampleCounter/src/ for a simple example.

database

src/database

This Directory has the following files:

pvDatabase.h
This is what is described in this section.
pvDatabase.cpp
The implementation of PVDatabase.
pvRecord.cpp
The implementation of the base class for PVRecord. It can also implement record instances with a process method does nothing. This can be used to create a "dumb" record where all changes are done by clients.

src/special

This directory has the following files:

recordList.h
This implements a PVRecord that provides a list of the names of the records in the PVDatabase. It also serves as an example of how to implement a service. The exampleDatabase creates an instance via the following code:
recordName = "laptoprecordListPGRPC";
pvRecord = RecordListRecord::create(recordName);
result = master->addRecord(pvRecord);
      
traceRecord.h
This implements a PVRecord that can set the trace level for another record. See below for a discussion of trace level.

pvDatabase.h

The classes in pvDatabase.h describe a database of memory resident smart records. It describes the following classes:

PVRecord
This provides the methods required by localChannelProvider to implement Channel.
PVRecordField
PVRecordStructure
These wrap PVField and PVStructure so that pvCopy and monitor can be implemented.
PVRecordClient
This is called by anything that acceses PVRecord.
PVListener
This is implemented by anything that wants to trap calls to PVRecord::message.
PVDatabase
This is a database of PVRecords.

Each class is described in a separate subsection.

C++ namespace and typedefs

namespace epics { namespace pvDatabase {

class PVRecord;
typedef std::tr1::shared_ptr<PVRecord> PVRecordPtr;
typedef std::map<epics::pvData::String,PVRecordPtr> PVRecordMap;

class PVRecordField;
typedef std::tr1::shared_ptr<PVRecordField> PVRecordFieldPtr;
typedef std::vector<PVRecordFieldPtr> PVRecordFieldPtrArray;
typedef std::tr1::shared_ptr<PVRecordFieldPtrArray> PVRecordFieldPtrArrayPtr;

class PVRecordStructure;
typedef std::tr1::shared_ptr<PVRecordStructure> PVRecordStructurePtr;

class PVRecordClient;
typedef std::tr1::shared_ptr<PVRecordClient> PVRecordClientPtr;

class PVListener;
typedef std::tr1::shared_ptr<PVListener> PVListenerPtr;

class RecordPutRequester;
typedef std::tr1::shared_ptr<RecordPutRequester> RecordPutRequesterPtr;

class PVDatabase;
typedef std::tr1::shared_ptr<PVDatabase> PVDatabasePtr;

class PVRecord

NOTES:


PVRecord Methods
class PVRecord
     public epics::pvData::Requester,
     public std::tr1::enable_shared_from_this<PVRecord>
{
public:
    POINTER_DEFINITIONS(PVRecord);

    virtual bool init() {initPVRecord(); return true;}
    virtual void process() {}
    virtual void destroy();

    static PVRecordPtr create(
        std::string const & recordName,
        epics::pvData::PVStructurePtr const & pvStructure);
    virtual ~PVRecord();
    std::string getRecordName();
    PVRecordStructurePtr getPVRecordStructure();
    PVRecordFieldPtr findPVRecordField(
        epics::pvData::PVFieldPtr const & pvField);
    bool addRequester(epics::pvData::RequesterPtr const & requester);
    bool removeRequester(epics::pvData::RequesterPtr const & requester);
    inline void lock_guard() { epics::pvData::Lock theLock(mutex); }
    void lock();
    void unlock();
    bool tryLock();
    void lockOtherRecord(PVRecordPtr const & otherRecord);
    bool addPVRecordClient(PVRecordClientPtr const & pvRecordClient);
    bool removePVRecordClient(PVRecordClientPtr const & pvRecordClient);
    void detachClients();
    bool addListener(PVListenerPtr const & pvListener);
    bool removeListener(PVListenerPtr const & pvListener);
    void beginGroupPut();
    void endGroupPut();
    std::string getRequesterName() {return getRecordName();}
    virtual void message(
        std::string const & message,
        epics::pvData::MessageType messageType);
    void message(
        PVRecordFieldPtr const & pvRecordField,
        std::string const & message,
        epics::pvData::MessageType messageType);
    void toString(epics::pvData::StringBuilder buf);
    void toString(epics::pvData::StringBuilder buf,int indentLevel);
    int getTraceLevel();
    void setTraceLevel(int level);
protected:
    PVRecord(
        std::string const & recordName,
        epics::pvData::PVStructurePtr const & pvStructure);
    void initPVRecord();
    epics::pvData::PVStructurePtr getPVStructure();
    PVRecordPtr getPtrSelf()
    {
        return shared_from_this();
    }
private:
...
}

The methods are:

init
Virtual method. Derived classes must implement this method. This method Must call initPVRecord.
process
Virtual method. Derived classes must implement this method. The base implementation does nothing.
destroy
This is a virtual method. A derived class must call the base class destroy method after it has released any resources it uses.
create
Static method to create dumb records, i.e. records with a process method that does nothing. A derived class should have it';s own static create method.
~PVRecord
The destructor which must be virtual. A derived class must also have a virtual destructor.
getRecordName
Return the recordName.
getPVRecordStructure
Get the top level PVStructure.
findPVRecordField
Given a PVFieldPtr return the PVRecordFieldPtr for the field.
addRequester
Add a requester to receive messages.
removeRequester
Remove a message requester.
lock_guard
This is an inline method that locks the record. The record will automatically be unlocked when control leaves the block that has the call.
lock
unlock
Lock and Unlock the record. Any code accessing the data in the record or calling other PVRecord methods must have the record locked.
tryLock
If true then just like lock. If falseclient can not access record. A client can try to simultaneously hold the lock for more than two records by calling this method. But must be willing to accept failure.
lockOtherRecord
A client that holds the lock for one record can lock one other record. A client must not call this if the client already has the lock for more then one record.
addPVRecordClient
Every client that accesses the record must call this so that the client can be notified when the record is deleted.
removePVRecordClient
Client is no longer accessing the record.
detachClients
Ask all clients to detach from the record
addListener
Add a PVListener. This must be called before calling pvRecordField.addListener.
removeListener
Removes a listener. The listener will also be removed from all fields to which it is attached.
beginGroupPut
Begin a group of puts. This results in all registered PVListeners being called
endGroupPut
End a group of puts. This results in all registered PVListeners being called.
getRequesterName
virtual method of Requester
message
Can be called by implementation code. The message will be sent to every requester.
toString
Just calls the top level PVStructure toString method.
getTraceLevel
This can be used for debugging. There are currently three levels that are used by existing code.
0
Produce no trace messages.
1
Issue a message to std::cout whenever anything is created or destroyed.
2
In addition to lifetime messages also issue a message whenever the record is accessed by pvAccess client.
setTraceLevel
Set the trace level. Note that special, described below. provides a record support that allows a pvAccess client to set the trace level of a record.

The protected methods are:

PVRecord
The constructor. It requires a recordName and a top level PVStructure.
initPVRecord
This method must be called by derived class.
getPVStructure
Called by derived class.

class PVRecordField

class PVRecordField {
     public virtual epics::pvData::PostHandler,
     public std::tr1::enable_shared_from_this<PVRecordField>
public:
    POINTER_DEFINITIONS(PVRecordField);
    PVRecordField(
        epics::pvData::PVFieldPtr const & pvField,
        PVRecordStructurePtr const &parent,
        PVRecordPtr const & pvRecord);
    virtual ~PVRecordField();
    virtual void destroy();
    PVRecordStructurePtr getParent();
    epics::pvData::PVFieldPtr getPVField();
    std::string getFullFieldName();
    std::string getFullName();
    PVRecordPtr getPVRecord();
    bool addListener(PVListenerPtr const & pvListener);
    virtual void removeListener(PVListenerPtr const & pvListener);
    virtual void postPut();
    virtual void message(
        std::string const & message,
        epics::pvData::MessageType messageType);
protected:
    PVRecordFieldPtr getPtrSelf()
    {
        return shared_from_this();
    }
    virtual void init();
    virtual void postParent(PVRecordFieldPtr const & subField);
    virtual void postSubField();
private:
...
};

When PVRecord is created it creates a PVRecordField for every field in the PVStructure that holds the data. It has the following methods:

PVRecordField
The constructor.
~PVRecordField
The destructor.
destroy
Called by PVRecordStructure when it's destroy method is called.
getParent
Get the parent PVRecordStructure for this field.
getPVField
Get the PVField associated with this PVRecordField.
getFullFieldName
This gets the full name of the field, i.e. field,field,..
getFullName
This gets recordName plus the full name of the field, i.e. recordName.field,field,..
getPVRecord
Returns the PVRecord to which this field belongs.
addListener
Add A PVListener to this field. Whenever this field or any subfield if this field is modified the listener will be notified. PVListener is described below. Before a listener can call addListener it must first call PVRecord.registerListener.
removeListener
Remove a PVListener.
postPut
This is called by the code that implements the data interface. It is called whenever the put method is called.
message
Called by implementation code. It calls PVRecord::message after prepending the full fieldname.

class PVRecordStructure

class PVRecordStructure : public PVRecordField {
public:
    POINTER_DEFINITIONS(PVRecordStructure);
    PVRecordStructure(
        epics::pvData::PVStructurePtr const & pvStructure,
        PVRecordFieldPtrArrayPtr const & pvRecordField);
    virtual ~PVRecordStructure();
    virtual void destroy();
    PVRecordFieldPtrArrayPtr getPVRecordFields();
    epics::pvData::PVStructurePtr getPVStructure();
    virtual void removeListener(PVListenerPtr const & pvListener);
    virtual void postPut();
protected:
    virtual void init();
private:
...
};

When PVRecord is created it creates a PVRecordStructure for every structure field in the PVStructure that holds the data. It has the following methods:

PVRecordStructure
The constructor.
~PVRecordStructure
The destructor.
getPVRecordFields
Get the PVRecordField array for the subfields
getPVStructure
Get the PVStructure for this field.
removeListener
Remove a PVListener.
postPut
This is called by the code that implements the data interface. It is called whenever the put method is called.

class PVRecordClient

class PVRecordClient {
    POINTER_DEFINITIONS(PVRecordClient);
    virtual ~PVRecordClient();
    virtual void detach(PVRecordPtr const & pvRecord);
};

where

~PVRecordClient
The destructor.
detach
The record is being removed from the master database,

class PVListener

class PVListener {
    virtual public PVRecordClient
public:
    POINTER_DEFINITIONS(PVListener);
    virtual ~PVListener();
    virtual void dataPut(PVRecordFieldPtr const & pvRecordField) = 0;
    virtual void dataPut(
        PVRecordStructurePtr const &
        requested,PVRecordFieldPtr const & pvRecordField) = 0;
    virtual void beginGroupPut(PVRecordPtr const & pvRecord) = 0;
    virtual void endGroupPut(PVRecordPtr const & pvRecord) = 0;
    virtual void unlisten(PVRecordPtr const & pvRecord);
};

where

~PVListener
The destructor.
dataPut(PVRecordFieldPtr const & pvRecordField)
pvField has been modified. This is called if the listener has called PVRecordField::addListener for pvRecordField.
dataPut( PVRecordStructurePtr const & requested,PVRecordFieldPtr const & pvRecordField)
pvField has been modified. Requested is the field to which the requester issued a pvField-&addListener. This is called if the listener has called PVRecordField-&addListener for requested.
beginGroupPut
A related set of changes is being started.
endGroupPut
A related set of changes is done.
unlisten
The record is being destroyed. The listener must release all access to the record.

class PVDatabase

class PVDatabase : virtual public epics::pvData::Requester {
public:
    POINTER_DEFINITIONS(PVDatabase);
    static PVDatabasePtr getMaster();
    virtual ~PVDatabase();
    virtual void destroy();
    PVRecordPtr findRecord(std::string const& recordName);
    bool addRecord(PVRecordPtr const & record);
    epics::pvData::PVStringArrayPtr getRecordNames();
    bool removeRecord(PVRecordPtr const & record);
    virtual std::string getRequesterName();
    virtual void message(
        std::string const &message,
        epics::pvData::MessageType messageType);
private:
    PVDatabase();
};

where

getMaster
Get the master database. This is the database that localChannelProvider access.
~PVDatabase
The destructor.
destroy
This is called by remote channelAccess when process exits. This destroys and removes all records in the PVDatabase.
findRecord
Find a record. An empty pointer is returned if the record is not in the database.
addRecord
Add a record to the database. If the record already exists it is not modified and false is returned.
getRecordNames
Returns an array of all the record names.
removeRecord
Remove a record from the database. If the record was not in the database false is returned.
getRequesterName
Virtual method of Requester
message
Virtual message of Requester.

pvAccess

This is code that provides an implementation of channelProvider as defined by pvAccess. It provides access to PVRecords and is access by the server side of remote pvAccess.

channelProviderLocal

This is a complete implementation of channelProvider and , except for channelRPC, provides a complete implementation of Channel as defined by pvAccess. For monitors it calls the code described in the following sections.

pvCopy

This provides code that creates a top level PVStructure that is an arbitrary subset of the fields in the PVStructure from a PVRecord. In addition it provides code that monitors changes to the fields in a PVRecord. A client configures the desired set of subfields and monitoring options via a pvRequest structure. pvAccess provides a class CreatePVRequest that creates a pvRequest. The pvCopy code provides the same functionality as the pvCopy code in pvIOCJava.

monitorAlgorithm

Currently all that is implemented is a header file. The only algorithm currently implemented is onPut

monitorFactory

Overview

epics::pvData::monitor defines the monitor interfaces as seen by a client. See pvDatabaseCPP.html For details.

monitorFactory implements the monitoring interfaces for a PVRecord. It implements queueSize=0 and queueSize>=2.

The implementation uses PVCopy and PVCopyMonitor which are implemented in pvCopy. When PVCopyMonitor tells monitor that changes have occurred, monitor applies the appropriate algorithm to each changed field.

Currently only algorithm onPut is implemented but, like pvIOCJava there are plans to support for the following monitor algorithms:

onPut
A monitor is issued whenever a put is issued to the field. This is the default unless the record defines deadbands for a field. An exception is the top level timeStamp which by default is made onChange and monitor will not be raised.
onChange
This provides two options: 1) A monitor is raised whenever a field changes value, and 2) A monitor will never be raised for the field.
deadband
The field must be a numeric scalar. Whenever the absolute or percentage value of the field changes by more than a deadband a monitor is issued. The record instance can also define deadbands.
periodic
A monitor is issued at a periodic rate if a put was issued to any field being monitored.

MonitorFactory

MonitorFactory provides the following methods:

class MonitorFactory
{
    static MonitorPtr create(
        PVRecordPtr const & pvRecord,
        MonitorRequester::shared_pointer const & monitorRequester,
        PVStructurePtr const & pvRequest);
    static void registerMonitorAlgorithmCreater(
        MonitorAlgorithmCreatePtr const & monitorAlgorithmCreate,
        String const & algorithmName);
}

where

create
Create a monitor. The arguments are:
pvRecord
The record being monitored.
monitorRequester
The monitor requester. This is the code to which monitot events will be delivered.
pvRequest
The request options
registerMonitorAlgorithmCreater
Called by code that implements a monitor algorithm.

special

This section provides two useful record support modules and one that is used for testing.

traceRecord

This implements a PVRecord that allows a client to set the trace level of a record. It follows the pattern of a channelPutGet record:

traceRecord
    structure arguments
        string recordName 
        int level 0
    structure result
        string status 
where:
recordName
The name of the record to set the trace level.
level
The level to set. The meaning is:
0
No trace messages generated
1
Lifecycle messages will be generated. This all channel create and destroy instances will be shown.
2
In addition to lifecycle messages a message will be generted for each get and put request.
>2
Currently no definition
result
The result of a cannelPutGet request

testExampleServerMain.cpp has an example of how to create a traceRecord:

PVDatabasePtr master = PVDatabase::getMaster();
PVRecordPtr pvRecord;
String recordName;
bool result(false);
recordName = "traceRecordPGRPC";
pvRecord = TraceRecord::create(recordName);
result = master->addRecord(pvRecord);
if(!result) cout<< "record " << recordName << " not added" << endl;

recordList

This implements a PVRecord that allows a client to set the trace level of a record. It follows the pattern of a channelPutGet record:

traceRecord
    structure arguments
        string database master
        string regularExpression .*
    structure result
        string status 
        string[] names
where:
database
The name of the datbase. The default is "master"
regularExpression
For now this is ignored and the complete list of names is always returned.
status
The status of a putGet request.
names
The list of record names.

Note that swtshell, which is a Java GUI tool, has a command channelList that requires that a record of this type is present and calls it. Thus user code does not have to use a channelGetPut to get the list of record names.

testExampleServerMain.cpp has an example of how to create a traceRecord:

recordName = "laptoprecordListPGRPC";
pvRecord = RecordListRecord::create(recordName);
result = master->addRecord(pvRecord);
if(!result) cout<< "record " << recordName << " not added" << endl;

exampleServer

Overview

The example implements a simple service that has a top level pvStructure:

structure
    structure argument
        string value
    structure result
        string value
        time_t timeStamp
            long secondsPastEpoch
            int nanoSeconds
            int userTag

It is designed to be accessed via a channelPutGet request. The client sets argument.value When the record processes it sets result.value to "Hello " concatenated with argument.value. Thus if the client sets argument.value equal to "World" result.value will be "Hello World". In addition the timeStamp is set to the time when process is called.

The example can be run on linux as follows:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleService
mrk> bin/linux-x86_64/exampleService

Directory Layout

The directory layout is:

exampleServer
    configure
       ExampleRELEASE.local
       ...
    src
       exampleServer.h
       exampleServer.cpp
       exampleServerInclude.dbd
       exampleServerMain.cpp
       exampleServerRegister.cpp
    ioc
       Db
          ...
       src
          exampleServerInclude.dbd
          exampleServerMain.cpp
   iocBoot
      exampleServer
         st.cmd
         ...
where
ExampleRELEASE.local
This is the file that must be copied to RELEASE.local and edited.
exampleServer.h
The header file for the service.
exampleServer.cpp
The service implementation.
exampleServerMain.cpp
A main program that runs the example so that it can be accessed by a pvAccess client.
exampleServerInclude.dbd
This has a register command so that the service can be started on a V3 IOC via iocsh.
exampleServerRegister.cpp
This has the code to start the service via the following iocsh command.
exampleServerCreateRecord exampleServer
Multiple commands can be issued to create multiple service records.
ioc
This is for building a V3 IOC application.
ioc/Db
This has template files for creating V3 records.
ioc/src
The files for running a V3 IOC.
iocBoot/exampleServer
A place to start exampleServer as part of a V3IOC. It has a st.cmd file that starts the ioc and also starts pvAccess and the example.

If only a main program is desired then the directory layout is:

exampleServer
    configure
       ExampleRELEASE.local
       ...
    src
       exampleServer.h
       exampleServer.cpp
       exampleServerMain.cpp

Thus if only a main program is required the directory layout is simple.

Also many sites will want to build the src directory in an area separate from where the iocs are build.

exampleServer.h

The example resides in src The implementation is in exampleServer.cpp.

The description consists of

class ExampleServer;
typedef std::tr1::shared_ptr<ExampleServer> ExampleServerPtr;

class ExampleServer :
    public PVRecord
{
public:
    POINTER_DEFINITIONS(ExampleServer);
    static ExampleServerPtr create(
        std::string const & recordName);
    virtual ~ExampleServer();
    virtual void destroy();
    virtual bool init();
    virtual void process();
private:
    ExampleServer(std::string const & recordName,
        epics::pvData::PVStructurePtr const & pvStructure);

    epics::pvData::PVStringPtr pvArgumentValue;
    epics::pvData::PVStringPtr pvResultValue;
    epics::pvData::PVTimeStamp pvTimeStamp;
    epics::pvData::TimeStamp timeStamp;
};

where

create
This is example specific but each support could provide a similar static method.
~ExampleServer
The destructor must be declared virtual.
destroy
Called when the record is being destroyed. This must call the base class destroy method.
init
A method to initialize the support. It returns true if initialization is successful and false if not. NOTE that this is a virtual method of PVRecord itself.
process
This again is a virtual method of PVRecord.
ExampleServer
For the example this is private.
pvValue
This is the field of the top level structure that process accesses.

The implementation of create method is:

ExampleServerPtr ExampleServer::create(
    std::string const & recordName)
{
    StandardPVFieldPtr standardPVField = getStandardPVField();
    PVDataCreatePtr pvDataCreate = getPVDataCreate();
    PVStructurePtr pvArgument = standardPVField->scalar(pvString,"");
    PVStructurePtr pvResult = standardPVField->scalar(pvString,"timeStamp");
    StringArray names;
    names.reserve(2);
    PVFieldPtrArray fields;
    fields.reserve(2);
    names.push_back("argument");
    fields.push_back(pvArgument);
    names.push_back("result");
    fields.push_back(pvResult);
    PVStructurePtr pvStructure = pvDataCreate->createPVStructure(names,fields);
    ExampleServerPtr pvRecord(
        new ExampleServer(recordName,pvStructure));
    if(!pvRecord->init()) pvRecord.reset();
    return pvRecord;
}
This:
  • Creates the top level structure.
  • Creates a ExampleServerPtr via the constructor.
  • Calls init and if it fails resets the shared pointer.
  • Returns the shared pointer to the newly created record.

The private constructor method is:

ExampleServer::ExampleServer(
    std::string const & recordName,
    epics::pvData::PVStructurePtr const & pvStructure)
: PVRecord(recordName,pvStructure)
{
}
The example is very simple. Note that it calls the base class constructor.

The destructor and destroy methods are:

ExampleServer::~ExampleServer()
{
}

void ExampleServer::destroy()
{
    PVRecord::destroy();
}
The destructor has nothing to do. The destroy method, which is virtual, just calls the destroy method of the base class. A more complicated example can clean up any resources it used but must call the base class destroy method.

The implementation of init is:

bool ExampleServer::init()
{
    initPVRecord();
    PVFieldPtr pvField;
    pvArgumentValue = getPVStructure()->getStringField("argument.value");
    if(pvArgumentValue.get()==NULL) return false;
    pvResultValue = getPVStructure()->getStringField("result.value");
    if(pvResultValue.get()==NULL) return false;
    pvTimeStamp.attach(getPVStructure()->getSubField("result.timeStamp"));
    return true;
}

The implementation of process is:

void ExampleServer::process()
{
    pvResultValue->put(String("Hello ") + pvArgumentValue->get());
    timeStamp.getCurrent();
    pvTimeStamp.set(timeStamp);
}
It gives a value to result.value and then sets the timeStamp to the current time.

src/exampleServerMain.cpp

NOTE: This is a shorter version of the actual code. It shows the essential code. The actual example shows how create an additional record.

The main program is:

int main(int argc,char *argv[])
{
    PVDatabasePtr master = PVDatabase::getMaster();
    ChannelProviderLocalPtr channelProvider = ChannelProviderLocal::create();
    String recordName("exampleServer");
    PVRecordPtr pvRecord = ExampleServer::create(recordName);
    bool result = master->addRecord(pvRecord);
    cout << "result of addRecord " << recordName << " " << result << endl;
    pvRecord.reset();
    startPVAServer(PVACCESS_ALL_PROVIDERS,0,true,true);
    cout << "exampleServer\n";
    string str;
    while(true) {
        cout << "Type exit to stop: \n";
        getline(cin,str);
        if(str.compare("exit")==0) break;

    }
    return 0;
}
This:
  • Gets a pointer to the master database.
  • Creates the local Channel Provider. This starts the pvAccess server.
  • Creates a ExampleServer record with the name exampleServer
  • Prints exampleServer on standard out.
  • Runs forever until the user types exit on standard in.

V3IOC exampleServer

To start exampleServer as part of a V3IOC:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleServer/iocBoot/exampleServer
mrk> ../../../bin/linux-x86_64/exampleServer st.cmd

You can then issue the commands dbl and pvdbl:

epics> dbl
double01
epics> pvdbl
exampleServer
epics> 
double01 is a v3Record. exampleServer is a pvRecord.

It starts pvASrv so that the V3 records can be accessed via Channel Access or via PVAccess.

exampleDatabase

The exampleServer pvDatabase has many records including the following:

exampleDouble
A record that is an instance of a record with a process method that does nothing. To test it start a channelPut and a channelGet and/or monitor.
exampleDoubleArray
An array record that is an instance of a record with a process method that does nothing. It can be tested like exampleDouble. In addition channelArray can also be used.
laptoprecordListPGRPC
Implements the record expected by swtshell channelList. It can also be used via channelPutGet.
traceRecordPGRPC
This can be used via channelPutGet to set the trace level of another record.

It also has a number of other scalar and array records.

exampleDatabase can be started as a main program or as a V3 IOIC. If started as a V3 IOC it also has a number of V3 records, and starts pvaSrv so that the V3 records can be accessed via Channel Access or via PVAccess.

exampleLink

This example show how a service can access other PVRecords. This section 1) starts with a discussion of accessing data via pvAccess and 2) gives a brief description of an example that gets data for an array of doubles.

Discussion

Access Alternatives

The process routine of a PVRecord can access other PVRecords in two ways:

Directly accessing local pvDatabase
If the other PVRecord is accessed via the master PVDatabase then threading issues are up to the implementation. For now this method will not be discussed.
Access via pvAccess
If access is via pvAccess then locking is handled by pvAccess.

Access via pvAccess can be done either by local or remote channel provider.

Access via channelProviderLocal
If the local pvAccess server is used the implementation must be careful that it does not cause deadlocks. When the process method is called the pvRecord for the process method is locked. When it makes a pvAccess get, put, etc request the other record is locked. Thus if a set of pvAccess links are implemented the possibility of deadlocks exists. A simple example is two records that have links to each other. More complex sets are easily created. Unless the developer has complete control of the set of records then remote pvAccess should be used. But this results in more context switches.
Access via remote pvAccess
If remote pvAccess is used then all locking issues are handled by pvAccess. The linked channel can be a pvRecord in the local pvDatabase or can be implemented by a remote pvAccess server.

Data synchronization

If pvAccess is used then it handles data synchronization. This is done by making a copy of the data that is transfered between the two pvRecords. This is true if either remote or local pvAccess is used. Each get, put, etc request results in data being copied between the two records.

If the linked channel is a local pvRecord then, for scalar and structure arrays, raw data is NOT copied for gets. This is because pvData uses shared_vector to hold the raw data. Instead of copying the raw data the reference count is incremented.

For puts the linked array will force a new allocation of the raw data in the linked record, i. e. copy on write semantics are enforced. This is done automatically by pvData and not by pvDatabase.

Some details

As mentioned before a pvDatabase server can be either a separate process, i. e. a main program, or can be part of a V3IOC.

A main pvDatabase server issues the following calls:

 ClientFactory::start();
 ChannelProviderLocalPtr channelProvider = getChannelProviderLocal();
 ...
 ServerContext::shared_pointer serverContext = startPVAServer(PVACCESS_ALL_PROVIDERS,0,true,true);

The first call is only necessary if some of the pvRecords have pvAccess links. These must be called before any code that uses links is initialized. After these two calls there will be two channel providers: local, and pvAccess.

A pvDatabase that is part of a V3IOC has the following in the st.cmd file.

...
iocInit()
startPVAClient
startPVAServer
## commands to create pvRecords

Once the client and local provider code has started then the following creates a channel access link.

PVDatabasePtr master = PVDatabase::getMaster();
ChannelAccess::shared_pointer channelAccess = getChannelAccess();
ChannelProvider::shared_pointer provider = channelAccess->getProvider(providerName);
Channel::shared_pointer channel = provider->createChannel(channelName,channelRequester);

Directory Layout

exampleLink
    configure
       ExampleRELEASE.local
       ...
    src
       exampleLink.h
       exampleLink.cpp
       exampleLinkInclude.dbd
       exampleLinkRegister.cpp
    ioc
       Db
       src
          exampleLinkInclude.dbd
          exampleLinkMain.cpp
   iocBoot
      exampleLink
         st.local
         st.remote
         ...

This example is only built to be run as part of a V3 IOC. Note that two startup files are available: st.local and st.remote. st.local has two records: doubleArray and exampleLink. doubleArray is a record that can be changed via a call to pvput. exampleLink is a record that, when processed, gets the value from doubleArray and sets its value equal to the value read. st.local has both records. st.remote has only one record named exampleLinkRemote.

To start the example:

mrk> pwd
/home/hg/pvDatabaseCPP/exampleLink/iocBoot/exampleLink
mrk> ../../bin/linux-x86_64/exampleLink st.local 

then in another window:

mrk> pvput doubleArray 4 100 200 300 400
Old : doubleArray 0
New : doubleArray 4 100 200 300 400
mrk> pvget -r "record[process=true]field(value)" exampleLink
exampleLink
structure 
    double[] value [100,200,300,400]
mrk> 

exampleLink Implementation

exampleLink.h contains the following:

...
class ExampleLink :
    public PVRecord,
    public epics::pvAccess::ChannelRequester,
    public epics::pvAccess::ChannelGetRequester
{
public:
    POINTER_DEFINITIONS(ExampleLink);
    static ExampleLinkPtr create(
        std::string const & recordName,
        std::string const & providerName,
        std::string const & channelName
        );
    virtual ~ExampleLink() {}
    virtual void destroy();
    virtual bool init();
    virtual void process();
    virtual void channelCreated(
        const epics::pvData::Status& status,
        epics::pvAccess::Channel::shared_pointer const & channel);
    virtual void channelStateChange(
        epics::pvAccess::Channel::shared_pointer const & channel,
        epics::pvAccess::Channel::ConnectionState connectionState);
    virtual void channelGetConnect(
        const epics::pvData::Status& status,
        epics::pvAccess::ChannelGet::shared_pointer const & channelGet,
        epics::pvData::PVStructure::shared_pointer const & pvStructure,
        epics::pvData::BitSet::shared_pointer const & bitSet);
    virtual void getDone(const epics::pvData::Status& status);
private:
...

All the non-static methods are either PVRecord, PVChannel, or PVChannelGet methods and will not be discussed further. The create method is called to create a new PVRecord instance with code that will issue a ChannelGet::get request every time the process method of the instance is called. Some other pvAccess client can issue a channelGet, to the record instance, with a request to process in order to test the example.

All of the initialization is done by a combination of the create and init methods so lets look at them:

ExampleLinkPtr ExampleLink::create(
    String const & recordName,
    String const & providerName,
    String const & channelName)
{
    PVStructurePtr pvStructure = getStandardPVField()->scalarArray(
        pvDouble,"alarm.timeStamp");
    ExampleLinkPtr pvRecord(
        new ExampleLink(
           recordName,providerName,channelName,pvStructure));
    if(!pvRecord->init()) pvRecord.reset();
    return pvRecord;
}

This first creates a new ExampleLink instance, and then calls the init method and the returns a ExampleLinkPtr. Note that if init returns false it returns a pointer to NULL.

The init method is:

bool ExampleLink::init()
{
    initPVRecord();

    PVStructurePtr pvStructure = getPVRecordStructure()->getPVStructure();
    pvTimeStamp.attach(pvStructure->getSubField("timeStamp"));
    pvAlarm.attach(pvStructure->getSubField("alarm"));
    pvValue = static_pointer_cast<PVDoubleArray>(
        pvStructure->getScalarArrayField("value",pvDouble));
    if(pvValue==NULL) {
        return false;
    }
    ChannelAccess::shared_pointer channelAccess = getChannelAccess();
    ChannelProvider::shared_pointer provider =
        channelAccess->getProvider(providerName);
    if(provider==NULL) {
         cout << getRecordName() << " provider "
              << providerName << " does not exist" << endl;
        return false;
    }
    ChannelRequester::shared_pointer channelRequester =
        dynamic_pointer_cast<ChannelRequester>(getPtrSelf());
    channel = provider->createChannel(channelName,channelRequester);
    event.wait();
    if(!status.isOK()) {
        cout << getRecordName() << " createChannel failed "
             << status.getMessage() << endl;
        return false;
    }
    ChannelGetRequester::shared_pointer channelGetRequester =
        dynamic_pointer_cast<ChannelGetRequester>(getPtrSelf());
    PVStructurePtr pvRequest = getCreateRequest()->createRequest(
        "value,alarm,timeStamp",getPtrSelf());
    channelGet = channel->createChannelGet(channelGetRequester,pvRequest);
    event.wait();
    if(!status.isOK()) {
        cout << getRecordName() << " createChannelGet failed "
             << status.getMessage() << endl;
        return false;
    }
    getPVValue = static_pointer_cast<PVDoubleArray>(
        getPVStructure->getScalarArrayField("value",pvDouble));
    if(getPVValue==NULL) {
        cout << getRecordName() << " get value not  PVDoubleArray" << endl;
        return false;
    }
    return true;
}

This first makes sure the pvStructure has the fields it requires:

timeStamp
A timeStamp structure. This will be set to the current time when process is called.
alarm
An alarm structure. This will be used to pass status information to the client when process is called.
value
This must be a scalarArray of type double. It is where data is copied when the channelGet is issued.

Next it makes sure the channelProvider exists.

Next it creates the channel and waits until it connects.

Next it creates the channelGet and waits until it is created.

Next it makes sure it has connected to a double array field.

If anything goes wrong during initialization it returns false. This a return of true means that it has successfully created a channelGet and is ready to issue gets when process is called.

Look at the code for more details.

examplePowerSupply

This is an example of creating a service that requires a somewhat complicated top level PVStructure. It is similar to the powerSupply example that is provided with pvIOCJava. Look at the code for details.

Array Performance and Memory Example

This section describes main programs that demonstrate performance of large arrays and can also be used to check for memory leaks. Checking for memory leaks can be accomplished by running the programs with valgrind or some other memory check program.

Brief Summary

The programs are:

arrayPerformanceMain
This is server and also a configurable number of longArrayMonitor clients. The clients can use either the local or remote providers. The moitor code is the same code that is used by longArrayMonitorMain.
longArrayMonitorMain
Remote client that monitors the array served by arrayPerformanceMain.
longArrayGetMain
Remote client that uses channelGet to access the array served by arrayPerformanceMain.
longArrayPutMain
Remote client that uses channelPut to access the array served by arrayPerformanceMain.

Each has support for -help.

mrk> pwd
/home/hg/pvDatabaseCPP-md
mrk> bin/linux-x86_64/arrayPerformanceMain -help
arrayPerformanceMain recordName size delay providerName nMonitor queueSize waitTime
default
arrayPerformance arrayPerformance 10000000 0.0001 local 1 2 0.0

mrk> bin/linux-x86_64/longArrayMonitorMain -help
longArrayMonitorMain channelName queueSize waitTime
default
longArrayMonitorMain arrayPerformance 2 0.0

mrk> bin/linux-x86_64/longArrayGetMain -help
longArrayGetMain channelName iterBetweenCreateChannel iterBetweenCreateChannelGet delayTime
default
longArrayGetMain arrayPerformance 0 0 1

mrk> bin/linux-x86_64/longArrayPutMain -help
longArrayPutMain channelName arraySize iterBetweenCreateChannel iterBetweenCreateChannelPut delayTime
default
longArrayPutMain arrayPerformance 10 0 0 1

mrk> 

Example output

Note: These may fail if run on a platform that does not have sufficent memory,

To see an example just execute the following commands in four different terminal windows:

bin/linux/<arch>/arrayPerformanceMain
bin/linux/<arch>/longArrayMonitorMain
bin/linux/<arch>/longArrayGetMain
bin/linux/<arch>/longArrayPutMain

Each program generates a report every second when it has somthing to report. Examples are:

mrk> bin/linux-x86_64/arrayPerformanceMain
arrayPerformance arrayPerformance 10000000 0.0001 local 1 2 0
...
 monitors/sec 66 first 131 last 131 changed {1, 2} overrun {} megaElements/sec 656.999
arrayPerformance value 132 time 1.00486 Iterations/sec 65.681 megaElements/sec 656.81
 monitors/sec 66 first 197 last 197 changed {1, 2} overrun {} megaElements/sec 656.304
arrayPerformance value 198 time 1.00563 Iterations/sec 65.6307 megaElements/sec 656.307
 monitors/sec 66 first 263 last 263 changed {1, 2} overrun {} megaElements/sec 654.824
...
mrk> bin/linux-x86_64/longArrayMonitorMain
longArrayMonitorMain arrayPerformance 2 0
...
 monitors/sec 6 first 2357 last 2357 changed {1, 2} overrun {} megaElements/sec 68.6406
 monitors/sec 13 first 2385 last 2385 changed {1, 2} overrun {} megaElements/sec 118.72
 monitors/sec 9 first 2418 last 2418 changed {1, 2} overrun {1, 2} megaElements/sec 85.0984
...
mrk> bin/linux-x86_64/longArrayPutMain
longArrayPutMain arrayPerformance 10 0 0 1
...
put numChannelPut 0 time 1.00148 Elements/sec 79.8819
put numChannelPut 1 time 1.00176 Elements/sec 79.8598
...
mrk> bin/linux-x86_64/longArrayGetMain
longArrayGetMain arrayPerformance 0 0 1
...
get kiloElements/sec 7384.61
get kiloElements/sec 8726.34
...

arrayPerformance

The arguments for arrayPerforamanceMain are:

recordName
The name for the arrayPerform record.
size
The size for the long array of the value field.
delay
The time in seconds to sleep after each iteration.
providerName
The name of the channel provider for the longArrayMonitors created by the main program. This must be either local or pvAccess.
nMonitor
The number of longArrayMonitors to create.
queueSize
The queueSize for the element queue. A value less than 1 will become 1.
waitTime
The time that longArrayMonitor will sleep after poll returns a monitorElement.

arrayPerformance creates a PVRecord that has the structure:.

recordName
    long[] value
    timeStamp timeStamp
    alarm alarm
Thus it holds an array of 64 bit integers.

The record has support that consists of a separate thread that runs until the record is destroyed executing the following algorithm:

report
Once a second it produces a report. In the above example output each line starting with ArrayPerformance is an arrayPerformance report.
create array
A new shared_vector is created and each element is set equal to the interation count.
lock
The arrayPerformance record is locked.
Begin group put
beginGroupReport is called.
replace
The value field of the record is replaced with the newly created shared_vector.
process
The record is then processed. This causes the timeStamp to be set to the current time.
End group put
endGroupPut is called.
unlock
The arrayPerformance record is unlocked.
delay
If delay is greater than zero epicsThreadSleep is called.

longArrayMonitor

This is a pvAccess client that monitors an arrayPerformance record. It generates a report every second showing how many elements has received. For every monitor it also checks that the number of alements is >0 and the the first element equals the last element. It reports an error if either of these conditions is not true.

The arguments for longArrayMonitorMain are:

channelName
The name for the arrayPerform record.
queueSize
The queueSize. Note that any size <2 is made 2.
waitTime
The time to wait after a poll request returns a monitorElement. This can be used to force an overrun of the client even if there is no overrun on the server.

longArrayGet

This is a pvAccess client that uses channelGet to access an arrayPerformance record. Every second it produces a report.

The arguments for longArrayGetMain are:

channelName
The name for the arrayPerform record.
iterBetweenCreateChannel
The number of iterations between destroying and recreating the channel. A value of 0 means never destroy and recreate.
iterBetweenCreateChannelGet
The number of iterations between destroying and recreating the channelGet. A value of 0 means never destroy and recreate.
delayTime
The time to dalay between gets.

longArrayPut

This is a pvAccess client that uses channelPut to access an arrayPerformance record. Every second it produces a report.

The arguments for longArrayPutMain are:

channelName
The name for the arrayPerform record.
arraySize
The capacity and length of the array to put to the server.
iterBetweenCreateChannel
The number of iterations between destroying and recreating the channel. A value of 0 means never destroy and recreate.
iterBetweenCreateChannelPut
The number of iterations between destroying and recreating the channelPut. A value of 0 means never destroy and recreate.
delayTime
The time to dalay between gets.

Some results

array performance

The results were from my laptop. It has a 2.2Ghz intel core i7 with 4Gbytes of memory. The operating system is linux fedora 16.

When test are performed with large arrays it is a good idea to also run a system monitor facility and check memory and swap history. If a test configuration causes physical memory to be exhausted then performance becomes very poor. You do not want to do this.

arrayPerformance results

The simplest test to run arrayPerformance with the defaults:

mrk> pwd
/home/hg/pvDatabaseCPP-md
mrk> bin/linux-x86_64/arrayPerformanceMain

This means that the array will hold 10 million elements. The delay will be a millisecond. There will be a single monitor and it will connect directly to the local channelProvider, i. e. it will not use any network connection.

The report shows that arrayPerformance can perform about 50 iterations per second and is putting about 500million elements per second. Since each element is an int64 this means about 4gigaBytes per second.

When no monitors are requested and a remote longArrayMonitorMain is run:

mr> pwd
/home/hg/pvDatabaseCPP-md
mrk> bin/linux-x86_64/longArrayMonitorMain

The performance drops to about 25 interations per second and 250 million elements per second. The next section has an example that demonstrates what happens. Note that if the array size is small enough to fit in the local cache then running longArrayMonitor has almost no effect of arrayPerforance.

memory leaks

Running longArrayMonitorMain, longArrayPutMain, and longArrayGetMain under valgrind shows no memory leaks.

arrayPerformanceMain shows the following:

==9125== LEAK SUMMARY:
==9125==    definitely lost: 0 bytes in 0 blocks
==9125==    indirectly lost: 0 bytes in 0 blocks
==9125==      possibly lost: 576 bytes in 2 blocks

The possibly leaked is either 1 or 2 blocks. It seems to be the same if clients are connected.

Vector Performance

This example demonstrates how array size effects performance. The example is run as:

bin/linux-x86_64/vectorPerformanceMain -help
vectorPerformanceMain size delay nThread
default
vectorPerformance 50000000 0.01 1 

Consider the following:

bin/linux-x86_64/vectorPerformanceMain 50000000 0.00 1
...
thread0 value 20 time 1.01897 iterations/sec 19.6277 elements/sec 981.383million
thread0 value 40 time 1.01238 iterations/sec 19.7554 elements/sec 987.772million
thread0 value 60 time 1.00878 iterations/sec 19.826 elements/sec 991.299million
...
bin/linux-x86_64/vectorPerformanceMain 50000000 0.00 2
...
thread0 value 21 time 1.00917 iterations/sec 9.90911 elements/sec 495.455million
thread1 value 31 time 1.05659 iterations/sec 9.46443 elements/sec 473.221million
thread0 value 31 time 1.07683 iterations/sec 9.28648 elements/sec 464.324million
thread1 value 41 time 1.0108 iterations/sec 9.89312 elements/sec 494.656million
...
bin/linux-x86_64/vectorPerformanceMain 50000000 0.00 3
thread0 value 7 time 1.0336 iterations/sec 6.77244 elements/sec 338.622million
thread1 value 7 time 1.03929 iterations/sec 6.73534 elements/sec 336.767million
thread2 value 7 time 1.04345 iterations/sec 6.70852 elements/sec 335.426million
thread0 value 14 time 1.03335 iterations/sec 6.77406 elements/sec 338.703million
thread1 value 14 time 1.03438 iterations/sec 6.76734 elements/sec 338.367million
thread2 value 14 time 1.04197 iterations/sec 6.71805 elements/sec 335.903million
...
bin/linux-x86_64/vectorPerformanceMain 50000000 0.00 4
thread2 value 5 time 1.00746 iterations/sec 4.96298 elements/sec 248.149million
thread1 value 5 time 1.02722 iterations/sec 4.86751 elements/sec 243.376million
thread3 value 5 time 1.032 iterations/sec 4.84496 elements/sec 242.248million
thread0 value 6 time 1.18882 iterations/sec 5.04703 elements/sec 252.351million
thread2 value 10 time 1.00388 iterations/sec 4.98068 elements/sec 249.034million
thread3 value 10 time 1.02755 iterations/sec 4.86592 elements/sec 243.296million
thread1 value 10 time 1.04836 iterations/sec 4.76936 elements/sec 238.468million
thread0 value 11 time 1.01575 iterations/sec 4.92249 elements/sec 246.124million

As more threads are running the slower each thread runs.

But now consider a size that fits in a local cache.

bin/linux-x86_64/vectorPerformanceMain 5000 0.00n/linux-x86_64/vectorPerformanceMain 5000 0.00 1
...
thread0 value 283499 time 1 iterations/sec 283498 elements/sec 1417.49million
thread0 value 569654 time 1 iterations/sec 286154 elements/sec 1430.77million
thread0 value 856046 time 1 iterations/sec 286392 elements/sec 1431.96million
... 
bin/linux-x86_64/vectorPerformanceMain 5000 0.00 2
...
thread0 value 541790 time 1 iterations/sec 271513 elements/sec 1357.56million
thread1 value 541798 time 1 iterations/sec 271418 elements/sec 1357.09million
thread0 value 813833 time 1 iterations/sec 272043 elements/sec 1360.21million
thread1 value 813778 time 1 iterations/sec 271979 elements/sec 1359.89million
thread0 value 541790 time 1 iterations/sec 271513 elements/sec 1357.56million
thread1 value 541798 time 1 iterations/sec 271418 elements/sec 1357.09million
thread0 value 813833 time 1 iterations/sec 272043 elements/sec 1360.21million
thread1 value 813778 time 1 iterations/sec 271979 elements/sec 1359.89million
...
bin/linux-x86_64/vectorPerformanceMain 5000 0.00 3
...
thread0 value 257090 time 1 iterations/sec 257089 elements/sec 1285.45million
thread1 value 256556 time 1 iterations/sec 256556 elements/sec 1282.78million
thread2 value 514269 time 1 iterations/sec 257839 elements/sec 1289.19million
thread0 value 514977 time 1 iterations/sec 257887 elements/sec 1289.43million
thread1 value 514119 time 1 iterations/sec 257563 elements/sec 1287.81million
thread2 value 770802 time 1 iterations/sec 256532 elements/sec 1282.66million

Now the number of threads has a far smaller effect on the performance of each thread.