Version 20 (modified by piontek, 13 years ago) (diff)

--

QCG Klient dla infrastruktury PL-Grid

Infrastruktura QoSCosGrid wraz z opisem wymagań i procedur dla projektu PL-Grid.

Dostęp do infrastruktury QCG

Dostęp do infrastruktury QCG możliwy jest z dowolnego komputera na którym zainstalowany jest klient QCG-Broker. Dla wygody użytkowników uruchomiona została maszyna dostępowa (qcg.man.poznan.pl) z zainstalowaną wersją klienta dla użytkowników PL-GRID.

Logowanie: ssh <plguser>@qcg.man.poznan.pl

np.

ssh plgpiontek@qcg.man.poznan.pl

Przy logowaniu należy podać hasło ustawione w portalu PL-GRID.

Po zalogowaniu prosze postępować zgodnie z wytycznymi, w szczególnosci opisującymi sposób przeprowadzenia konfiguracji środowiska.

Job Profile

Every experiment submitted to QCG-Broker has to be described by XML-based document called Job Profile. The structure of Job Profile is formalized by  Job Profile schema.

Examples of Job Profiles describing basic use cases are distributed with QCG-Broker and can be found in <CLIENT_DIR>/examples directory.

Usage

The QCG-Groker command-line java based client can operate in two modes:

  • batch mode – that executes single operation with arguments passed directly to the client during its invocation. The batch mode allows to use the client in any kind of scripts mostly in cases when the processing of output is needed to steer the experiment,
  • console mode – that works similar to shell console in which user can type in lines with operations and arguments to be executed by service. The console mode gives additional useful features like aliases, history accessible by arrows-keys, creation and management of user proxy, help functionality.

The usage of the client depends on the mode:

  • for batch mode: "qcg-client OPRATION [ARG1 .. ARGn]"
  • for console mode: "qcg-client -console" and then user is prompted to type in lines in format "OPERATION [ARG1 .. ARGn]" to be processed by client.

IMPORTANT: To secure the communication between client and service and to delegate user's privileges to the service client needs access to user's proxy certificate.

Operations

Regardless from the mode the QCG-Broker java based command-line client supports following list of operations:

Operation Arguments Description
submit_job <desc_file> [QCG or JSDL] submits a job to be executed. The description of job can be expressed either in native QCG-Broker language (the default one) or in JSDL one. In case when the job to be submitted is described in JSDL format the type (JSDL) must be explicitly specified. If the description is valid client returns to the user a globally unique job identifier, which unambiguously identifies the job in the system. QCG defines jobs as a sets of dependent tasks that constitute a logical whole (workflow). Each task is executed by system only if all tasks it depends on are in specified by the user states.
list_jobs [<limit>] [<status>] lists jobs belonging to the user. It is possible either to limit number of jobs or to display only ones in given state. All possible states are listed below the table.
list_user_jobs <user> [<limit>] [<status>] lists jobs belonging to the given user. The functionality is destined for administrative purposes.
test_description <desc_file> [QCG or JSDL] validates job description
translate_description <desc_file> JSDL translates job description to native QCG-Broker one
job_info <jobId> [<showJobDesc>] return complex information about the given job. If the showJobDesc is false the job description is not shown
cancel_job <jobId> cancels execution of the given job
commit_job <jobId> allows to approve the job submitted with two phase commit mechanism to be processed by the system. The two phase commit mechanism can be used to register notifications before the processing of the job will be started by broker.
list_tasks <jobId> [<status>] lists tasks belonging to given job. Optionally it is possible to specify the task's status. Possible task statuses are listed below the table.
tasks_statuses <jobId> [<summary>] lists tasks constituting the given job with their statuses. If the summary argument is true some additionall statistics is displayed.
register_job_notification <jobId> <url> registers notification consumer for the given job
list_job_notifications <jobId> lists notifications registered for the given job
register_tasks_notification <jobId> <url> register notification for all tasks of the given job
monitor_job <jobId> [<interval>] monitors status changes of tasks belonging to given job. The interval argument determines delay in seconds between next status checks.
monitor_task <jobId> <taskId> [<interval>] monitors status changes of allocations belonging to the given tasks. The interval argument determines delay in seconds between next status checks.
task_info <jobId> <taskId> [<showDesc> [<limit>]] displays information about the given task. If the showDesc is false the task description is not shown. If the limit argument is specified the history of the task is limited to given value.
register_task_notification <jobId> <taskId> <url> registers task's notification consumer
list_task_notifications <jobId> <taskId> lists task's notifications
cancel_task <jobId> <taskId> cancels execution of the given task
commit_task <jobId> <taskId> commits the given task to be processed by the system
reserve_resources [<taskId>] <job_desc> (QCG or JSDL) reserve resources that meet either the wole job or given task requirements. The reservation identifier is returned. This functionality is not implemented yet!
reservation_info <reservationId> return complex information concerning the given reservation: list of reserved resources, local identifiers of reservations, reservation time slot. This functionality id not implemented yet!
cancel_reservation <reservationId> releases reserved resources. This functionality is not implemented yet!

List of Job statuses:

  • UNCOMMITTED - the job was submitted with two phase commit option and waits to be committed,
  • SUBMITTED – the job was submitted to the system and is executed by the system,
  • SUSPENDED – the job was suspended,
  • ACTIVE – the job is active, at least one task is processed,
  • FINISHED – the job was completed,
  • FAILED – the job (at least one crucial task belonging to the job) failed
  • CANCELED – the job was canceled by the user,
  • BROKEN - one or more of crucial tasks failed, system waits until active tasks will finish and change the status of the job to FAILED.|

List of Task statuses:

  • UNSUBMITTED – the task cannot be started because of dependencies,
  • UNCOMMITED - the task waits to be committed,
  • QUEUED – the task was put into the queue and waits for execution,
  • PREPROCESSING – system makes some actions needed to start the task (looks for the resource, stages in files),
  • PENDING – the task is pending in the queueing-system,
  • RUNNING – the task is active,
  • STOPPED – the task was finished or was checkpointed, but system did not start staging out files,
  • POSTPROCESSING – system makes some actions needed to complete the task, for example stages out files, cleares working environment, etc.,
  • FINISHED – the task was completed,
  • SUSPENDED – the task was suspended,
  • FAILED – the task failed,
  • CANCELED – the task was canceled by the user.

Usage examples

Przykłady użycia klienta QCG