RMOST User Guide
Contents
1. 2 1 2 Configuration of RM Spy Beside the mentioned basic properties RM Spy has some more properties which configure the standart notification and the delay on early termination The properties NotifyOnStart and RMSpy NotifyOnEnd configure if noti fications are send on job start or job termination If the property is set to a non zero value notifications are sent on start termination of the job If the property is set to zero no notifications are sent If the property is not specified no notifications are sent by default The following example send notifications on job termination but not at job start RMSpy NotifyOnStart RMSpy NotifyOnEnd 1 The second group of properties configure the behaviour of early termina tion of the job It consists of the properties HoldOnFailure HoldOnTime and MaxEvents The property HoldOnFailure switches the delay on early termination on or off If it is set to a non zero value a delay are added on early termination If the property is not defined the default is zero which mean no delay The following example enables the delay RMSpy HoldOnFailure 1 RMOST decides if the termination is early by comparing the current number of executed events with a given number of events that the job should process This number is defined by the property MaxEvents If the job terminates be fore the specified number of events are processed the delay is started and a notification about the early termination is s2. RM_IDeserializeMethod Deserialize bool registerByte const char xname char xvalue bool writeable true bool readable true bool registerInt const char xname int value bool writeable true bool readable true bool registerLong const char name long xvalue bool writeable true bool readable true bool registerFloat const char name float xvalue bool writeable true bool readable true bool registerDouble const char xname double xvalue bool writeable true bool readable true 28 32 33 34 36 37 38 39 40 41 42 43 44 virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual bool bool bool bool bool void void void bool registerString const char xname char xxvalue bool writeable true bool readable true registerFile const char fileName registerROOTFile const char fileName unregister const char name clearRegistration check requestValue char xname getValue char xname synchronize char name TObjArray getTable bool bool bool bool bool proceed terminate stop nextStep restart double getTime 29
3. an interactive connection exists To enable an additional notification mechanism via email set the prop erty NotifyByEmail to a non zero value If it is set to zero or not specified notifications are not sent via email SteeringSvc NotifyByEmail 1 For sending emails the target email address is required The email address is set to the property NotificationEmail For example SteeringSvc NotificationEmail myadress mydomain de Note The emails are send from the connection service Thus the host which runs the used connection service must have configured sendmail else no emails are sent 2 2 2 Introduction to the RM SteeringSvc API The RM SteeringSvc provides an API which can be used to enable steering for internal values of own components It offers many advanced possibilities for expert users others can skip this section Get access to steering system In this section it is shown how the source code can be instrumented to monitor and steer data For making the API known we need to include a header file to our source file include RM SteeringSvc h For the compiler find the header files some additional directories with header files need to be specified in the build instructions install root rmost 2 1 0 ResultMonitoring ResultMonitoring install root rmost 2 1 0 Common include install root rmost 2 1 0 GridConnection include install root rmost 2 1 0 Steering include install root rmost 2 1 0 Processi
4. display only connected only disconnected only jobs which received a notification or only jobs which have an error or notification Therefore select the menu item View Jobs and the appropriate subitem If you want to display all jobs again select the sub item View all Detailed view of a job To get a detailed view of a job select the job in the list and use the menu item View job and the sub item Job details A new window pops up It displays a list of registered data with their type and in case they have a basic type also their value If you select one row or cell and press the Request button the data of this cell will be updated with the value of the remote job If you select a file or stream and press the Request button the file or stream is downloaded and stored in a local file Therefore a file dialog opens and you can chose a filename where to store the file In case you request a ROOT file a new T Browser is opened and you can inspect the ROOT file If you request the same ROOT file several times a T Browser instance is started for each request Once the browser is opened all values displayed in this browser steam from the state of the file when it was requested even if the Athena job has progressed further A second browser may display different values if they have changed Values with a simple data type can be edited If you then press the Syn chronize button the value entered by you i
5. the steering service but do not call check O themselves To use RM Checker simply add it to your list of algorithms It has no properties theApp TopAlg RM Checker Of course the RMOST library must be loaded before theApp Dlls ResultMonitoring 2 4 RM EvaluatorBase The algorithm RM EvaluatorBase serves as base class for user defined evalu ations For every event a condition is checked and depending on the result a user notification can be sent To use the class RM EvaluatorBase derive your own class and overwrite the method checkCondition and getMessage The method checkCondition should perform the evaluation If it returns true a notification is sent The text of the message is defined by the return value of getMessage An example is shown in Listing 2 5 Listing 2 5 A customized autometed evaluation class MyEvaluator public rmost RM EvaluatorBase public virtual bool checkCondition virtual std string getMessage bool MyEvaluator checkCondition The result from the evaluation bool rv Your evaluation code return rv std string MyEvaluator getMessage return My test message 15 Chapter 3 Visualization Chapter 2 described the job side of the steering tool and what the user must do to make his data inside the remote job accessible In this chapter the steering and visualization tools that access the data are describ
6. Int bool rv m_SteeringSvc gt registerInt myIntName amp myInt true true The return value will be true if the value is registered successfully and false elsewise The first parameter is a std string and contains the binding name The name is an arbitrary string but it must be unique The second parameter is a pointer to the variable that contains the value Very important This pointer must be valid as long as the value is registered at the service The third parameter is a boolean value that indicates if the value is write able If it is set to true it means that this value may be set by the steering system The forth value is a boolean that indicates if this value is readable by the steering system If it is set to true it means that a steering tool may read this value Similar commands exist for a byte 64 bit integer IEEE floating point num ber IEEE double precision floating point number strings and further data types see appendix A to provide standard methods to serialize and deserialize the data User defined data types Now it is possible to monitor and steer basic data types Normally a user defines a lot of different complex data types that represent his her results and want to monitor this data types Therefore a flexible system is provided which allows to register data of arbitrary data types If user defined data types must be registered you should provide a serialize and a deserialize method for y
7. RMOST User Guide Daniel Lorenz University of Siegen Deliverable of the HEP CG Project Version 2 1 0 July 1 2008 GEF RDERT VOM D Bundesministerium f r Bildung und Forschung UNIVERSIT T SIEGEN Contents 1 Introduction 2 1 1 Overview over RMOST 2 L2 Steering model su 2222 9 be P CPU UO A a 3 2 Enable Athena jobs for steering 5 2 1 The Algorithm RM Spy e 5 2 1 4 Minimal setup of a job options file 6 2 1 2 Configuration of RM Spy 7 2 2 The Service RM SteeringSve 8 2 2 1 Setup Job Options llle 8 2 2 2 Introduction to the RM _SteeringSve API 9 2 2 3 Other Service Methods 14 2 3 07 Necker sa um ues mcr RE xe ce ih 14 2 4 RM EvaluatorB se 2 22 are a 15 3 Visualization 16 3 1 The Graphical Data Browser 2 2 2 2 nn 16 3 2 The Command Line Tool 19 4 Submit a Steered Job 22 5 The connection service 24 A List of RM ISteeringSvc Methods 25 B Data Type Codes 27 C List of TResultMonitorData Methods 28 Chapter 1 Introduction Although Grid computing offers an enormous potential to researchers by pro viding seamless access to a huge collection of data and computing resources scientific computing still suffers by the remarkable delays between submitting a Grid job and receiving its first feedba
8. a connection between the Grid job and the visualization tool Its use and features are describe in Chapter 5 The name service is necessary to establish the connection service For the name service RMOST uses R GMA RMOST is easy to integrate into existing jobs For the most common func tionality no source code changes are necessary only changes to the job options which compose the Athena Grid job are required However expert users can use a lot more functionalities but then source code instrumentation is necessary 1 2 Steering model RMOST implements a distributed shared memory DSM based model for on line steering The application and the visualization access both the same data concurrently The steered variable at the Athena job as well as the displayed value at the user s side are viewed as cached copies of the same shared data object A data object in this context can be a simple variable or a complex data structure It can be located in memory or on disk The application registers data which it shares and the user interface registers the data it want to use from the remote job To each registered data object belongs a so called binding name Each side can register only one data object under a particular binding name The steering system maps data objects with matching binding names to each other which means that changes of the value in one side will be propagated to the data object at the other side Chapter 2 Enable Athena j
9. alizeMyData myData value m_Value value int SerializeMyData serialize std ostream os std istream param Get write position of the stream int n os tellp Write content of m_Value to ostream os xos lt lt m_Value gt Intl lt lt 7 lt lt m_Value gt Floatl 11 13 14 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 m gt Compute length of the written data n os tellp n return n DeserializeMyData DeserializeMyData myData value void DeserializeMyData deserialize std istream xis m_Value value TO DO Set content of m_Value with the data read from istream is xos gt gt m_Value gt Intl gt gt m_Value gt Floatl Now the user can register his object of type myData Listing 2 3 Register customized data types Create object of type myData myData data register the data in data as myDataObject1 readable and writeable bool rv registerValue RMDT_UserTypes myDataObject1 new SerializeMyData amp data new DeserializeMyData amp data Here data is a newly created object of type myData This object is regis tered with the method registerValue This object must exist as long as it is registered The return value is true if the data will be registered successfully The first parameter contains information about the data typ
10. ck Today after submitting a long running Grid job a researcher has to wait until the job has finished before he can retrieve any output Only then he or she can evaluate the results and in some cases see that they are not useful because some job parameters were set incorrectly or still need to be optimized to get significant results Thus the job must be re submitted again with a long waiting time Online steering i e the monitoring of intermediate results combined with an interactive control over the running job has proved to be an effective method to get around these problems thus accelerating computational scientific research This is even more true in a Grid environment given its long submission delays and the insufficient accessibility of running jobs If there would be any possibility to steer running Grid jobs execution time and research time could be reduced significantly 1 1 Overview over RMOST RMOST Result Monitoring and Online Steering Tool is a steering and moni toring system for Grid jobs of the ATLAS experiment software It can connect to a slightly modified Grid job and make the intermediate results accessible from within the widely used ROOT framework It can be applied by simple changes of the job options and then allows steering of the execution monitoring of the most common intermediate results which are stored in ROOT files and upload of new job options which can be applied after a restart of the job without re
11. e char xxvalue bool writeable true bool readable true registerActionType const std string name rmost RM ActionType value bool writeable true bool readable true registerFile const std string name rmost RM ISerializeFiles Serialize rmost RM_IDeserializeFiles Deserialize registerFile const std string fileName registerROOTFile const std string fileName unregister const charx name clearRegistration check notify std string subject std string message int priority int event terminate stop step proceed restart sendUpdate std string name 26 Appendix B Data Type Codes Data Type Numeric Code Sybolic Name byte 1 RMDT Byte int 2 RMDT Int long 3 RMDT Long float 4 RMDT Float double 5 RMDT_Double C string 6 RMDT String steering value 7 RMDT Action Type stream 8 RMDT File ROOT file 9 RMDT_ROOTFile for internal use 10 RMDT Internal data block of fixed size 11 RMDT _DataBlock procedure 12 RMDT Procedure notification 13 RMDT Notification boolean 14 RMDT Bool user defined type 15 RMDT_UserTypes 27 Appendix C List of TResultMonitorData Methods Listing C 1 Methods of TResultMonitoringData virtual virtual virtual virtual virtual virtual virtual bool connect char job bool registerValue rmost RM DataType dtype const char name rmost RM ISerializeMethod x Serialize rmost
12. e RMDT_UserTypes says that it is a user defined data type The second parameter contains the name of the data The third parameter is a pointer to the serialization class used to read the data A new instance of SerializeMyData is created which serializes a myData object The forth parameter is a pointer to the correspond ing deserialize class The service takes ownership of the registered instances of SerializeMyData and DeserializeMyData and deletes them when the object unregister or the service finalizes If the deserialize method is a NULL pointer the data will not be writeable if the serialize method is a NULL pointer the data will not be readable The serialize contains parameter param which is unused until now The steering API offers to specify certain parameters for reading a value This parameter stream is given to the serialize method If no parameters are needed the param stream can be ignored Now arbitrary data from memory can be registered In general any kind of 12 10 11 12 13 14 16 data in memory in files or somewhere else can be registered this way But if data from files are registered the amount of data can quickly become too large to send it within one message but special support is given to data in files and streams and special methods are provided that can register files or streams bool registerFile const std string FileName This method registers a file given by its filename The file
13. e on local machine int integeri 100 Register integer1 RM_Data registerInt integer1 amp integer1 true true Set the value at remote job RM Data synchronize integeri Here we set the value of integerl to 100 and then call the synchronize method This method sets the variable in the job to the value stored in integer1 when the job executes the check method the next time If you need a value regularly you can subscribe for it Then you get the up dated value each time the job performs the check method This should not be done for large amounts of data or even files A call to unsubscribe lt name gt cancels the subscription Both methods have one parameter which specifies the name of the data To subscribe for a value this value must be already regis tered If you subscribe for a value you must make sure to call the check method regularly on the visualization side Else the buffer overflows and the data is lost Or an error occurs Subscribe for eventCounter RM Data subscribe eventCounter Unsubscribe for eventCounter RM Data unsubscribe EventCounter A list of all registered data from the job is returned by getTableO This method returns a TObjArray of objects of type TRMVarEntry Every instance of TRMVarEntry has two members the name and the data type Below is a short example for the use of getTable and TRMVarEntry get Table TObjArray table RM Data getTable get first en
14. ed Because an estab lished framework called ROOT used for visualization exists inside the supported ATLAS community the visualization is integrated into ROOT Two components are provided a command line tool for use within ROOT and a graphical user interface which can be started from within ROOT and from a shell In section 3 1 the graphical user interface is explained in section 3 2 the command line tool is explained For using the visualization within ROOT you need ROOT version 5 the GUI needs QT 3 3 See the RMOST Installation Guide for the exact version with which the binaries are compiled 3 1 The Graphical Data Browser The graphical user interface GUI manages a list of steerable jobs It tries to connect to all known jobs and displays an overview For each job a detailed view in a seperate window is available which displays a list of all registered data and their data type For simple data types a string representation of their value is given Furthermore registered streams or files can be downloaded to the local machine The execution of the job can be steered if the job supports this which is the case if the algorithm RM_Spy is used Start the GUI Before starting the steering tool you need to create a Grid proxy if it has not already been created The steering tool needs the proxy to authenticate the connection to the remote job Without a proxy the data browser can not be started but will only display an error message that i
15. ent The default value for this prop erty is 100 This property takes only effect if HoldOnFailure is set to nonzero In the following example the property is set to the number of events that the application should process RMSpy MaxEvents theApp EvtMax Finally the delay time can be specified by the property RMSpy HoldOnTime The maximum time it could wait are 30 minutes Default are 15 minutes This property takes only effect if HoldOnFailure is set to nonzero RMSpy HoldOnTime 10 2 2 The Service RM SteeringSvc The service RM_SteeringSvc oflers extended steering possibilities for advanced users and is used by the other steering components In the first section the configuration of the steering service via the job options is described RM SteeringSvc is a flexible tool to monitor and steer arbitrary user defined data It can be used within any other service or algorithm to publish the data This means that for using RM SteeringSvc the user needs to insert instrument calls to the RM SteeringSvc into the source code of the algorithm or service With the RM SteeringSvc it is possible to extend own components with the possibility to monitor and steer internal data during runtime and change pa rameters without restart the job The RM SteeringSvc API is described in the second section 2 2 1 Setup Job Options In this section the job options properties of RM SteeringSvc are described Firstly you need to load the necessary lib
16. er Register the variable RM Data registerInt eventCounter amp eventCounter Now the variable is registered but still no data exchange happened To retrieve the value from the remote job and store it in the registered variable the method getValue lt name gt can be used If you need several different values the asynchronous way is much faster In the asynchronous way we first request all necessary data except the last variable Because all requests are answered in the same order they are requested if the last variable is returned the other requests before are answered as well Blocking retrieval of one value RM Data getValue eventCounter a b c d are already registered Asynchronous requests RM Data requestValue a RM Data requestValue b RM Data requestValue c The last request is blocking RM Data getValue d The requestValue method has as only parameter the name of the requested data It sends a request to the job but it does not wait for the answer In coming messages are reviced from the checkmethod which is also nonblocking getValue calls check until the answer has returned it is recommended to use the getValue method for the last data request The next important question is how to set a value The value of event Counter can not be set in the job so it is assumed we have another integer named integeri that is already registered with the name integer1 20 Set valu
17. ice configured you can set an external one in the job options For this purpose the property ConnectionService ofthe steering service must be set theApp ExtSvc RM SteeringSvc SteeringSvc Service RM SteeringSvc SteeringSvc ConnectionService my service host port Hereby you must replace my service host by the fully qualified domain name of the host on which the connection service runs and port must be replaced by the port number the connection service listens on For example if the connection service was started on gcn54 hep physik uni siegen de and listens on port 20030 the connection service is configured with SteeringSvc ConnectionService gcnb4 hep physik uni siegen de 20030 RM Spy supports the remote access of files with intermediate results It makes sure to flush the file buffers of all files in the Athena framework To make files available the filenames of these files must be given to RM Spy via the job option FileNames It is possible to register any kind of files in this way e g ROOT files An arbitrary number of files can be handed to RM Spy the in following way RMSpy FileNames filei RMSpy FileNames file2 The names file1 and file2 must be substituted with the physical file names of your output files Caution Make sure to use instead of else all entries made before are lost Now the job supports basic steering actions and can be submitted
18. job A dialog box appear where the job identifier can be entered To load job identifier from a file select the mnu item Job identifier and the sub item Open ID file The a file dialog appears where you can choose the file which contains the job identifier After the job id file is opened a list of all job identifier in the file is dsiplayed where you can select those jobs which you want to see in the steering tool By default all jobs are selected If you selected all jobs click on the button Add in the dialog box The job list No matter which method you used the new job names should appear in the job list Jobs which are already in the list are ignored You should see 4 columns in the job list 1 The job identifier is displayed in the first column 17 2 The progress column shows the number of already processed events of the job if the connection is already established else a status of the establish ment of the connection is given 3 The thrid column is a message column It displays short forms of error messages or notifications The color of a cell is changed according to the kind of message If a notification is available the cell is red if an error appeared the cell is yellow 4 The comments column is editable for the user to enter his commands about the job If you have a large number of jobs there exist some methods to filter the whole list and display only jobs with certain characteristics You can
19. name is also used as binding name and must be unique under all registered data in the whole job When needed the files are simply opened The service takes no action to flush the file buffer before the data is read on a request of the data This file is registered as writeable and will be overwritten by a write command from the steering tool The second method for streams is more general bool registerFile const std string name RM ISerializeFiles Serialize RM IDeserializeFiles Deserialize In this function name is again the binding name for the registered data but it does not refer to a real file in any way It simply must be unique The second and the third parameter are pointers to access methods to a stream sim ilar to the registerValue method But here you derive the access classes from RM ISerializeFiles and RM IDeserializeFiles instead of deriving them from RM ISerializeMethod and RM IDeserializeMethod To create own data ac cess classes at least the following methods must be implemented Listing 2 4 Serialization methods for streams include RM SerializeFiles h class SerializeMyStream public RM ISerializeFiles public virtual std istream xserialize std istream param h class DeserializeMyStream public RM IDeserializeFiles public virtual void deserialize std istream xis RM_filesize_t offset RM filesize t size The serialize method does not write the data to a stream itself but re
20. ng include install root rmost 2 1 0 access include If you use CMT to build your Athena components you can add the include directories to your project by adding the following line to the requirements file include dirs install root rmost 2 install root rmost 2 install root rmost 2 install root rmost 2 install root rmost 2 install root rmost 2 0 ResultMonitoring ResultMonitoring 0 GridConnection include V 0 Common include V 0 Steering include V 0 Processing include 0 access include Before you can use the service you need to get a pointer to the service instance from the Athena framework This can be done with the following lines in the source code StatusCode sc RM_ISteeringSvc m_SteeringSvc sc service RM_SteeringSvc m_SteeringSvc true If the return value is StatusCode SUCCESS the steering service is available Make data steerable Now the steering service is active Every data object which shall be steerable must be registered with the steering service For registration an unique name that identifies this data must be provided In this manual this unique name is called the binding name of the particular data object If another data object is already registered with the same binding name the registration will fail As an example a 32 bit integer is registered and stored in the variable myInt with the binding name myIntName This is done with the method register
21. ob e Change or replace the job options file The new job options are applied after a restart of the job without job submission e Optionally notifications can be sent at start and termination of the job e On early termination the final termination can optionally be delayed for some time During the delay period interaction with the job is possible On start of the delay a notification is sent This should cover most of the cases for online steering Thus in most cases online steering is easy to apply to a job First a minimal setup of the job is described to enable steering Afterwards the available configuration parameters are described 2 1 1 Minimal setup of a job options file To apply RM Spy only the job options file must be edited This section describes the necessary modifications Before you can use any steering components you need to load the appropriate library This can be done by adding the following line to your job options file theApp Dlls ResultMonitoring You can add the algorithm RM Spy to you job by adding the following lines to your job options file theApp TopAlg RM Spy RMSpy Algorithm RM Spy The steering system uses a so called connection service see Ch 5 to es tablish connections between the Grid job and the steering tool at the user side If the site where the Grid job runs configured a connection service for their site this one is used But if there is no connection serv
22. obs for steering This chapter describes how an Athena job is enabled for steering via RMOST RMOST provides four additional Athena components RM Spy RM_SteeringSvc RM_Checker and RM_EvaluatorBase which can be used in an Athena Grid job The first RM_Spy is described in Section 2 1 and enables basic functionality for steering The RM_SteeringSvc encapsulates the steering API and is used by the other components but can also be used by customized Athena components to steer internal variables The RM_SteeringSvc is described in Section 2 2 The RM_Checker allows to place additional synchronization points in the Algorithm list see Section 2 3 Finally the RM_EvaluatorBase is a base class for user defined automated evaluation and notification see Section 2 4 2 1 The Algorithm RM_Spy The algorithm RM_Spy was designed to allow basic steering of Athena jobs without editing the source code of any component except the job options file If you add the algorithm RM_Spy to your job you already have the following possibilities e Monitoring of all intermediate data stored in ROOT files which can be downloaded or remotely accessed at run time e Terminate the job e Suspend execution e Execute one single event for example if you want to monitor the changes of one particular event After the event has been executed the further execution is suspended e Continue execution of a suspended job e Restart the job without resubmission of the j
23. ou can execute the script at the beginning of your standard startup script which might then look like Listing 4 2 The wrapper script my_script sh bin bash source setup_rmost sh source VO_ATLAS_SW_DIR software 13 0 30 setup sh source SITEROOT AtlasOffline 13 0 30 AtlasOfflineRunTime cmt setup sh export LD_LIBRARY_PATH PWD LD_LIBRARY_PATH export PATH PWD PATH athena py myAthenaJobOptions py Furthermore the gLite job description file JDL file must be modified For transferring the setup_rmost sh script with the job it must be added to the InputSandbox 22 InputSandbox setup_rmost sh my_script sh myAthenaJobOptions py To submit the job run glite wms job submit a vo atlas lt my_jdl_file gt Where you exchange lt my_jdl_file gt by the name of the job description file for your job 23 Chapter 5 The connection service For the steering tool to connect to the job it needs a connection service except the case that the worker node has inbound connectivity from the internet in a range of ports The connection service can be installed permanently by an site administrator somewhere or started by a user The only requirement is that the connection service needs one open port in a firewall where it can be contacted from the outside world The connection service can run under any account with a user or machine certificate To start a connection service you need either create a p
24. our data type yourself This means you must write two classes each containing one method derived from RM ISerializeMethod and RM IDeserializeMethod respectively Then you can register your data with these two classes for the data access For the basic data types standard serialize and deserialize methods are provided For example a data object of type myData should be registered Then two data access classes must be created which are derived from the class 10 20 21 22 23 24 RM_ISerializeMethod or RM_IDeserializeMethod respectively and overwrite the serialize and deserialize methods The method serialize writes all the data to the stream os and returns the number of bytes written The method deserialize reads the data from the stream and stores the data in the myData structure The code in listing 2 1 shows a simple example Listing 2 1 Creating own serialization methods class myData int Intl float Floatl class SerializeMyData public RM_ISerializeMethod public SerializeMyData myData value virtual int serialize std ostream os std istream param private myData xm Value IE class DeserializeMyData public RM IDeserializeMethod public DeserializeMyData myData value virtual void deserialize std istream is private myData m Value The implementation is shown in listing 2 2 Listing 2 2 Creating own serialization methods SerializeMyData Seri
25. r name service than the default R GMA based name service you can specify the name service s client library with the parameter ns library 24 Appendix A List of RM ISteeringSvc Methods Listing A 1 Methods of RM ISteeringSvc virtual virtual virtual virtual virtual virtual virtual bool bool bool bool bool bool bool connect char xjob registerValue RM DataType dtype const char name RM_ISerializeMethod xSerialize RM_IDeserializeMethod Deserialize registerByte const char xname char xvalue bool writeable true bool readable true registerInt const char xname int value bool writeable true bool readable true registerLong const char name long xvalue bool writeable true bool readable true registerFloat const char name float xvalue bool writeable true bool readable true registerDouble const char xname double xvalue bool writeable true bool readable true 25 32 33 34 36 37 38 39 40 41 42 43 44 66 67 68 69 70 71 72 virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual virtual bool bool bool bool bool bool bool void void void void void void void void registerString const char xnam
26. rary in the job options file This can be done by adding the line theApp Dlls ResultMonitoring Then the service must be initialized theApp ExtSvc RM_SteeringSvc SteeringSvc Service RM_SteeringSvc The property ConnectionService is strongly recommended to set The steering system uses a so called connection service to establish connections be tween the Grid job and the steering tool at the user side If the site where the Grid job runs configured a connection service for their site this one is used But if there is no connection service configured you can set an external one in the job options Because you can not be sure that all sites configured an connection service it is highly recommended to specify a default connection service For more informations about the CS see Ch 5 SteeringSvc ConnectionService my service host port In this command my service host must be replaced by the fully quali fied host name of the host where the connection service runs and port should be replaced by the port number the connection service listens on e g SteeringSvc ConnectionService gcnd4 hep physik uni siegen de 20030 Furthermore RM_SteeringSvc has two properties which configure optional notification mechanisms via email By default notifications are only send via the communication channel which is also used for steering This mechanism has the drawback that notifications are only sent if
27. roxy certificate before or have a access to a host or service certificate If you use your proxy certificate the service can only be contacted as long as the proxy is valid The connection service can be stared with rmost_cservice port where port is the port number the connection service listens on The port number is the only required parameter If you want to start the CS as a demon add the parameter demon rmost cservice lt port gt demon Then you can close your session or log off without stopping the CS By default the CS writes its log output to tmp rmost cservice log An other file can be specified for the logging output with log lt logfilename gt Also an alternative PID file can be specified with pid pidfilename The CS writes its process id into the PID file and is used by scripts the termi nate running connections services The default location of the PID file is tmp rmost_cservice pid To establish an interactive connection the connector needs the address of the target job To get this information the connector requests a name service By default R GMA is used as name service which provide its own communication system If a job uses another name service which has no own communication structure the name service is contacted via the CS Then all requests are sent to the CS which invokes the name service client which is a dynamically loadable library Thus if your connection service should support anothe
28. s stored to the according variable in the job If you press the Synchronize button after selecting a file or stream a file dialog opens In this file dialog you can choose a file that will be uploaded to the job and replace the existing one if there is one The old file that was overwritten is lost On the right side of the data browser there are 5 buttons for steering the job execution Stop Restart Terminate Step Continue For these execution steering buttons to work the job must execute the algorithm RM Spy The RM Spy therefore registered the value nextAction Of course you can edit this value inside the table and press synchronize if you know the numeric codes for the commands as well Now a short description of the execution steering commands follows The Terminate command terminates the job 18 The Restart command restarts the application without resubmitting the job It simply terminates the running application and executes the Athena script anew With the restart the previously changed job options can be applied The restart only works if the job options file is registered at the algorithm RM Spy The Stop command makes the job suspend further job execution until a Continue or Step command is given The Step command makes the job execute one further event and switches then to the waiting state The Continue command lets a waiting job resume with its exec
29. sion It perform an exec on itself e stopO The job waits at the next check call e stepO The job process one further check and changes then to wait e proceed Continue with normal execution Finally with notify the job can send a notification to the user notify std string subject std string message int priority int event For a complete method list of the interface RM SteeringSvc see Appendix A 2 3 RM Checker RM Checker is an Athena Algorithm which executes a synchronization point It can be instantiated multiple times thus it and can be used to insert additional synchronization points between algorithms by modifying the job options file The advantage of multiple synchronization points are e Additional waiting points and smaller steps in step execution mode At maximum it is possible to execute only one algorithm in each step Thus more detailed information about the emergence of results are available 14 w o 0 10 11 12 13 14 15 16 17 e Improved response time on requests Especially if long algorithm lists are used and the processing uses a longer time the response time can be reduced because the average time until the next synchronization point is reached is reduced Furthermore RM Checker might be used if RM Spy is not used but one need at least one synchronization point in the algorithms For example if components are used which register some data to
30. submission Advanced steering possibilities are given with instrumentation of the source code like steering and monitoring of arbitrary user defined data On the visualization side an interface to ROOT is provided that makes the steering capabilities available from within ROOT In this manual it is assumed that you are familiar with Athena and ROOT RMOST consists of three different modules see Fig 1 1 steering file library access extended job options RM RM_Spy SteeringSvc name service d nma connection service d Grid command GUI line tool Algorithm 1 Algorithm 2 Algorithm 3 Athena job User Interface Remote Grid job file steering access library event loop Figure 1 1 The different components of RMOST The Grid job access module It consists of the Athena components RM SPY RM SteeringSvc RM Checker and RM EvaluatorBase which are executed with the job on the remote worker node These components are described in Chapter 2 The Visualization tools are integrated into the ROOT framework They are located on the user interface and are your gate to your remote Grid jobs Two visualization tools are provided a graphical data browser and a command line tool Both tools are described in Chapter 3 The connection service is necessary to establish
31. sultMonitor so For accessing the remote job you must create an instance of TResultMonitorData Every instance of this class represents one remote job This can by done with TResultMonitorData RM Data After that connect this instance with the remote job Again only one tool can connect to a certain job at a time and only the user that submitted the job can access the job A second call to connect close the existing connection and restarts the connection process The connect call needs as parameter the job id which was returned at job submission 19 RM_Data connect lt job_id gt where you need to replace lt job_id gt with the identifier of the job you want to connect to For example RM Data connect https grid rb physik uni wuppertal de 9000 6cN8yhqbuubkqk3Gsp8f vw If you want to monitor or steer the data of the job you must register an instance of the same data type with the same identifier as in the remote job Since the methods for data registration are the same as in RM SteeringSvc this works analog to registering the data described in Chapter 2 2 E g if you want to get the values of the eventCounter registered by the algorithm RM Spy firstly declare an integer variable in which the tool stores the value and then register this integer with the TResultMonitorData instance The variable name and the identifier name may not be the same of course Declare variable which will contain the value int eventCount
32. t could not get the credentials There are two possibilities to start the data browser You can run the data browser from a shell or you can start the browser from within ROOT If you want to run the browser from a shell run the command ResultMonitor 16 ResultMonitor lt gcn56 gt Job identifier View Jobs Notifications Progress asgf2opdBcvypeidC Searching for the job RN igaTiushdsOhaHHswiwd 3 hdkad8ferrGhajsPqs ents 223 edsw7uaskHiswoQAga sTkLaQaasbcejwodOs lgdaoJospwsLoasPisw ents 166 Figure 3 1 The overview of known jobs in the GUI To start the GUI from ROOT has the advantage that the remote ROOT files can be inspected with the standart ROOT browser If the GUI is started from command line ROOT files are not displayable If you want to start the browser from within ROOT you firstly need to load the library with the visualization tools and then create an instance of the class TResultMonitorBrowser This will open the graphical user interface shown in fig 3 1 This is done with the following commands in ROOT L ResultMonitor so TResultMonitorBrowser b Add jobs to the list At the beginning the job list will be empty You can add jobs to the list by manually typing the job ids or by loading them from a file For example if you submit your job with glite wms job submit using the o option which writes the job id to a file To enter a job id manually select the menu item Job identifier and the sub item Add
33. try TRMVarEntry entry table gt First get name of first entry TString name entry gt getName get data type of first entry int dt entry gt getType A list of available data types and their numeric codes can be found in Appendix B Files and user defined data can be registered in the same way like in they are registered to RM_SteeringSvc see Ch 2 2 Once they are registered they are requested like any other data type A full list of available methods of the command line tool TResultMonitorData can be found in App C 21 o gt A e lt oN Oo Chapter 4 Submit a Steered Job Once the application is prepared for steering the job can be submitted Because the components needed for steering are not in the standard distributions of the middleware Athena and ROOT some shared libraries need to be sent with the job or downloaded by a startup script see also the Installation Guide http www hep physik uni siegen de grid rmost doc InstallationGuide pdf The bash script for downloading the necessary parts of latest RMOST version and setting up RMOST on a WN is Listing 4 1 Script setup_rmost sh which downloads and sets up RMOST on a WN wget http www hep physik uni siegen de grid rmost versions rmost wn latest tar tar xvf rmost wn latest tar cd rmost export LD LIBRARY PATH SPWD lib LD LIBRARY PATH export PATH PWD bin PATH cd Y
34. turns a pointer to an input stream from where the service can read the data The deserialize method has two additional parameters It writes the first size bytes of is to the position offset The serialize method contains again a parameter param which contains optional information for read operations 13 2 2 3 Other Service Methods If you want to unregister a value perhaps due to the fact that a value is not available anymore or invalid you can do it with the unregister method bool unregister const std string Name The only parameter is the binding name of the data to be unregistered If the call was successful it returns true If no value with this name was registered it returns false Any job that uses the steering service must call the check method regularly what means at least one the algorithms must call it once in its execute method It defines points where modifications to parameters are applied steering com mands are executed and intermediate results are retrieved If the job runs the RM_Spy or RM_Checker algorithm this is done by RM_Spy or RM_Checker If the job does not run RM_Spy of RM_Checker but uses RM_SteeringSvc some other algorithm must call check in its execute method void check This method has no parameters and no return value Furthermore RM_SteeringSvc provides method for controlling the execution of the job e terminate Terminates the job e restart Restarts the job without resubmis
35. ution At the bottom of the window is a text box which prints text messages concerning the job Notifications On urgent events the job can send notifications to the user This notifications are displayed by the steering tool if it is connected If a notification arrives at the GUI the notification cell of the job is painted red and optionally a window pops up with the full message Furthermore the notification is printed to the text box at the bottom of the detailed view for a job By default the popup widnows are enabled They can be disbaled by selecting the menu item Notification and the sub item Disable popup widgets The coloring of the notification view should help the user to identify the occurence of important new events in some jobs The coloring can be reset to normal color thus you are informaed if a new event occurs To reset the coloring select the menu item Notifications and teh sub item reset all Then the color for all jobs is reset If you coose the sub item reset selected only the selected jobs are reset 3 2 The Command Line Tool The command line interface gives access to the remote job from the ROOT command line The command line tool needs ROOT version 5 For accessing the remote Grid job a valid proxy certificate is necessary else the creation of the command line tool instance will quit with an error Now start ROOT and load the result monitoring and steering library by typing L Re
Download Pdf Manuals
Related Search