-- AndreiTsaregorodtsev - 17 Nov 2008

DIRAC Production Management

This page describes the procedures and tools to define and follow the LHCb productions within the DIRAC framework

Production Request

Workflow

Production

Production is a logical object which encapsulates all the necessary information necessary to create a workload corresponding to a Production Request. The workload is created as a number of jobs associated with the Production. Each job has a full set of parameters necessary for its execution, including the Application Workflow, input data, etc.

The Production manipulation includes the following operations provided that the corresponding Production Request and Workflow are defined:

  • Production creation.
  • Production manipulation:
    • Starting, stopping production
    • Production job submission
    • Production extension
    • Creation of derived productions
  • Production finalization.
  • Production monitoring.

Production creation

Production is created with createProduction CLI command ( or with the eventual Web interface ). At creation time the following parameters are provided:

  • Workflow XML file path or the name of a workflow from the Workflow Repository;
  • Production Request reference;
  • For "Processing" type Productions - specification of the input data in one of the following forms:
    • Bookkeeping Query in a form that can be passed to the Bookkeeping Service in order to obtain the list of input files
    • Regular expression filter that can be applied to file LFNs
  • For "Processing" type Productions - specification of how many input files are grouped together for a single job

As a result, the new Production is created with the New status. It is created with no input data associated and no jobs generated.

As soon as the new Production is created a special ProductionInputData Agent will attempt to resolve all the available input data files according to the Production definition by either querying the Bookkeeping service or looking for the eligible LFNs in the ProductionDB internal catalog. After the ProductionInputData Agent performed the input data resolution it sets a LastInputUpdate time stamp. This time stamp is used to invoke the ProductionInputData Agent later at regular predefined intervals to update the set of input files. All the files added to the Production by the ProductionInputData Agent are given initially Unused status.

Production Parameters

Production definition can be complemented by a number of parameters. Currently the following parameters can be defined:

  • MaxNumberOfJobs - this parameter is applicable to the Simulation type Productions. It will not allow to extend the number of jobs for the Production beyond this limit
  • AncestorDepth - this parameter is applicable to the Processing type Productions. It will trigger getting all the ancestors for each input file and take the ancestor locations into account when creating jobs.
  • PluginType - this parameter determines which algorithm will be used to create jobs by the Transformation Agent ( see below ).
  • SubmissionType - this parameter can take values Automatic / Manual and controls the way how the job are submitted ( see below ).

Creation of a Derived Production

It is sometimes necessary to update considerably the Production definition and process the remaining files with the new settings. This is achieved by defining a Derived Production. It is defined by the createDerivedProduction CLI command which is similar to the createProduction command with an additional argument specifying the original Production. Once the Derived Production is created the original Production is moved to the Finished status if not yet there. All the jobs in the original Production that are not in any of the final states ( Done, Completed, Failed or Killed ) are either killed ( for Running jobs ) or deleted.

The Derived Production inherits from the original Production all its files with the following rules applied:

  • Files in Processed and Problematic states are inheriting this state as well as the WMS JobID of jobs that processed them;
  • Files in all the other states ( Unused , Assigned ) are moved as Unused.

Production manipulation

Starting,Stopping Production

Production is initially created in the New state. This state does not allow to create jobs associated with the Production. To create jobs the Production should moved to the Active state. This is done by the start command of the CLI or with a corresponding command in the Web interface. The Production in the Active state is allowing to create jobs. The creation of jobs is different for Simulation and Processing type Productions

  • For the Simulation type productions the new jobs are created by extendProduction CLI command. This will create a specified number of new jobs.
  • In the current version of the ProductionDB for the Processing type productions the new jobs are created by a Transformation Agent. The Transformation Agent runs periodically and for each Active production checks the files in Unused state to see if there are enough files for grouping on the same site in order to create jobs. It creates jobs if it finds enough eligible files.

Production can pass to Stopped state using stop CLI command or with a corresponding command in the Web interface. For the Production in the Stopped state no new jobs can be created and submitted.

Job Submission

Job submission for the given Production is performed either manually or automatically depending on the SubmissionType parameter value

  • Manual job submission is performed by submitJobs CLI command which allows to specify
    • Production ID
    • Number of jobs to submit
    • Destination site (optionally)
  • Automatic job submission is performed by ProductionJob Agent which runs periodically and for each active Production attempts to submit a predefined number of jobs (50).

Production Finalization

The Production finalization consists in submission of jobs for the remaining non-processed files. For that the Production is passed to the Flushed state using setProductionStatus CLI command. For the Productions in this state the Transformation Agent creates jobs for all the remaining Unused files irrespective to the specified number of files per job. Once the jobs are created, the Production passes to Finished state. For the Productions in this state no new files are added and no new jobs can be created. The Production in Finished state can not be restarted again.

Edit | Attach | Watch | Print version | History: r4 < r3 < r2 < r1 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r1 - 2008-11-18 - AndreiTsaregorodtsev
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LHCb All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback