System Configuration Parameters in Database

There is the PANDA.CONFIG table, so-called gdpconfig table in the database where you can define any configuration parameter shared by all PanDA applications, so that system admins don’t have to tweak the static cfg files every time they optimize the system.

The table has the following columns:

Name	Description
APP	The application name which uses the parameter
COMPONENT	The component name which uses the parameter
KEY	The parameter name
VALUE	The parameter value
TYPE	The parameter type
VO	The organization name which defines the parameter
DESCR	Description of the parameter

Applications get those parameters through the pandaserver.taskbuffer.TaskBuffer module.

from pandaserver.taskbuffer.TaskBuffer import taskBuffer
p = taskBuffer.getConfigValue(COMPONENT, KEY, APP, VO)

The method returns None if the parameter is undefined.

Parameter List

The following is the list of parameters defined in the PANDA.CONFIG table. The list is not exhaustive and may be updated frequently, so please check the table for the latest information.

Key	Description
ANALY_TIMEOUT[_<project>]	Timeout value for analysis rebrokerage in hours. A project-specific override can be set via ANALY_TIMEOUT_<project> (e.g. ANALY_TIMEOUT_proj-evind)
BASE_DEFAULT_QUEUE_LENGTH_PER_PQ_USER	Base minimal number of queued jobs per PQ per user (nQ(PQ, user))
BASE_EXPECTED_WAIT_HOUR_ON_PQ	If the expected wait time to digest all queued jobs on the PQ is shorter than this value (hours), analysis throttle will skip the PQ
BASE_QUEUE_LENGTH_PER_PQ	If the number of queued jobs on the PQ is lower than this value, analysis throttle will skip the PQ
BASE_QUEUE_RATIO_ON_PQ	Base minimal ratio of nQ(PQ, user)/nR(PQ): queued jobs of one user on a PQ divided by all running jobs on the PQ under the same prodsourcelabel
BUILD_JOB_MAX_WALLTIME	Max walltime for build jobs in hours
CAP_RUNNING_GROUP_CORES	Cap on running cores per group
CAP_RUNNING_GROUP_JOBS	Cap on running jobs per group
CAP_RUNNING_USER_CORES	Cap on running cores per user
CAP_RUNNING_USER_JOBS	Cap on running jobs per user
DATA_CAROUSEL_CONFIG	Configuration of Data Carousel in JSON format
DATA_CHECK_TIMEOUT_USER	Time limit in hours after which analysis brokerage allows input transfer
DATA_LOCATION_CHECK_PERIOD	Time limit for the task brokerage to skip data locality check
DISK_THRESHOLD[_<gshare>]	Disk size threshold in TB to skip the nucleus when available space minus reserved space is insufficient. A gshare-specific override can be set via DISK_THRESHOLD_<gshare> (e.g. DISK_THRESHOLD_MC 16 simul)
EVP_DATASET_LIFETIME	Lifetime of temporary datasets for event picking
FAST_REBRO_THRESHOLD_NQNR_RATIO	Threshold on nQ_pq(gshare)/nR_pq(gshare) ratio to trigger fast rebrokerage at the PQ
FAST_REBRO_THRESHOLD_NQUEUE_FRAC	Threshold on fraction of nQ_pq(gshare)/nQ_total(gshare) to trigger fast rebrokerage at the PQ
FAST_REBRO_THRESHOLD_NQUEUE_<gshare>	Min nQ_pq to trigger fast rebrokerage at the PQ for the given gshare (e.g. FAST_REBRO_THRESHOLD_NQUEUE_MC 16 simul)
FREE_DISK_CUTOFF	Maximum free disk space value in TB factored into the weight calculation
GROUPBYATTR_<activity>	Attribute to group tasks when generating jobs for the given activity (e.g. GROUPBYATTR_Analysis, GROUPBYATTR_User Analysis)
HEARTBEAT_TIMEOUT_<workflow>	Heartbeat timeout for the given workflow in hours (e.g. HEARTBEAT_TIMEOUT_analysis, HEARTBEAT_TIMEOUT_production, HEARTBEAT_TIMEOUT_push_noheartbeat)
IGNORE_MEANRSS	Flag to indicate whether to ignore mean RSS throttling in job dispatcher and pilot streaming
INPUT_NUM_FRACTION	Minimum percentage of available files in the input dataset which allows the task to go to the nucleus
INPUT_NUM_THRESHOLD	Minimum number of files in the input dataset to enable data availability check
INPUT_SIZE_FRACTION	Minimum percentage of available file size in the input dataset which allows the task to go to the nucleus
INPUT_SIZE_THRESHOLD	Minimum size of input dataset in GB to enable data availability check
IO_INTENSITY_CUTOFF	Brokerage IO intensity cutoff in kB/sec to prevent input transfers for heavy jobs
IO_INTENSITY_CUTOFF_USER	Analysis brokerage IO intensity cutoff in kB/sec to allow input transfers for light-weight jobs
JOB_MAX_ATTEMPT_user	Max number of attempts for a job. Tasks cannot be retried if some jobs exceed the limit. 0 to disable
JOB_SUBMISSION	Enable/disable job submission in JEDI
JUMBO_MAX_CURR_PRIO	Max priority of tasks which can enable jumbo jobs
JUMBO_MAX_EVENTS	Total number of events in tasks with jumbo jobs
JUMBO_MAX_FILES_TO_BOOST	Max number of remaining files in tasks with jumbo jobs when priorities are boosted
JUMBO_MAX_TASKS	Max number of tasks with jumbo jobs
JUMBO_MIN_EVENTS_DISABLE	Min number of events in a task to disable jumbo jobs
JUMBO_MIN_EVENTS_ENABLE	Min number of events in a task to enable jumbo jobs
JUMBO_PER_SITE	Number of jumbo jobs per task per site
JUMBO_PER_TASK	Number of jumbo jobs per task
JUMBO_PROG_TO_BOOST	Percentage of events processed in tasks with jumbo jobs when priorities are boosted
LIMIT_IOINTENSITY_managed	For prodSourceLabel=managed, only tasks with ioIntensity higher than this value will be considered
LIMIT_PRIORITY_managed	For prodSourceLabel=managed, only tasks with currentPriority lower than this value will be considered
MAX_ACTIVE_TASKS_PER_USER_user	Max number of active user tasks per user
MAX_DISKIO_DEFAULT	Max average DiskIO limit per core in kB/sec
MAX_EXPECTED_WAIT_HOUR	Max allowed expected wait time (hours) of queued jobs on one PQ per user, before excluding the PQ in brokerage
MAX_FAILED_HEP_SCORE_HOURS_user	Max HEP score hours used by failed jobs. Tasks go to exhausted if a retry is requested and the limit has been reached. 0 to disable
MAX_FAILED_HEP_SCORE_RATE_TO_PAUSE_managed	Max acceptable HEP score rate of failed jobs within a managed task, triggering an automatic pause when reached. 0 to disable
MAX_FAILED_HEP_SCORE_RATE_user	Max rate of failed/total HEP score. Tasks go to exhausted if further reattempts are requested and the limit is reached. 0 to disable
MAX_JOB_FAILURE_RATE_TO_PAUSE_managed	Max acceptable job failure rate within a managed task, triggering an automatic pause when reached. 0 to disable
MAX_JOB_FAILURE_RATE_user	Max single job failure rate. Tasks go to exhausted if a retry is requested and the limit has been reached. 0 to disable
MAX_MISSING_INPUT_FILES	The brokerage sends the task to pending if, at all online storage endpoints, the number of missing files exceeds this value and the ratio of available files drops below MIN_INPUT_COMPLETENESS
MAX_PREASSIGNED_TASKS_managed	Max number of production tasks to preassign to an empty queue per resource type
MAX_PRIO_TO_BOOTSTRAP	Max currentpriority of tasks used to bootstrap PQs where no jobs are running or queued
MAX_TASK_PRIO_WITH_LOCAL_DATA	Maximum task priority to assign the task only to a nucleus which has the input data locally
MERGE_JOB_MAX_WALLTIME_user	Max walltime for user merge jobs in hours
MIN_BAD_JOBS_TO_SKIP_PQ	Skip PQs if the total number of failed or closed jobs is larger than this value
MIN_CPU_EFFICIENCY_user	Minimum average CPU efficiency (%) across finished jobs in a task. User tasks go to exhausted upon retry if the average falls below the limit
MIN_FILES_READY_managed	Min number of ready files a task must have to be eligible for preassignment
MIN_FILES_REMAINING_managed	Min number of remaining files a task must have to be eligible for preassignment
MIN_INPUT_COMPLETENESS	The brokerage sends the task to pending if, at all online storage endpoints, the number of missing files exceeds MAX_MISSING_INPUT_FILES and the ratio of available files drops below this percentage
MIN_INPUT_SIZE_WITH_LOCAL_DATA	Minimum input data size in GB to assign the task only to a nucleus which has the input data locally
MIN_IO_INTENSITY_WITH_LOCAL_DATA	Minimum IO intensity in kBPerS to assign the task only to a nucleus which has the input data locally
MIN_JOBS_TO_PAUSE_managed	Minimum total number of jobs within a managed task required to trigger an automatic pause when the failure rate reaches the limit
MIN_REMAINING_JOBS_TO_PAUSE_managed	Minimum number of remaining jobs within a managed task required to trigger an automatic pause when the failure rate reaches the limit
MIN_WEIGHT_<activity>	Minimum brokerage weight for tasks of the given activity (MIN_WEIGHT_managed, MIN_WEIGHT_user)
NFILES_<gshare>	Number of files to generate jobs for the given gshare in one cycle (e.g. NFILES_Data Derivations)
NQUEUECAP_<gshare>	Maximum number of queued jobs for the given gshare (e.g. NQUEUECAP_Frontier)
NQUEUED_NUC_CAP	Cap on the aggregated number of files to the nucleus
NQUEUED_NUC_CAP_FOR_JOBS	Cap in the job brokerage on the aggregated number of files being transferred to the nucleus
NQUEUED_SAT_CAP	Cap on the number of files queued from satellite to nucleus
NQUEUELIMIT_<gshare>	Minimum number of queued jobs for the given gshare or resource label. Use -GSHARE- keys for gshare-based limits (e.g. NQUEUELIMIT_MC 16) and -RESOURCE- keys for resource-based limits (e.g. NQUEUELIMIT_eventservice)
NQUEUELIMITSITE_<activity>	Min number of jobs to assign to a PQ in the brokerage when the number of running jobs is small at the PQ, for the given activity (e.g. NQUEUELIMITSITE_MC merge)
NRUNNINGCAP_<gshare>	Maximum number of running jobs for the given gshare. Optionally scoped to core type via _MCORE* or _SCORE* suffix (e.g. NRUNNINGCAP_Frontier_MCORE*)
NTASKS_<gshare>	Number of tasks to generate jobs for the given gshare in one cycle (e.g. NTASKS_Group production)
NUM_CUTOFF_TO_MOVE_INPUT	Cutoff on the number of missing input files to move for IO intensive tasks
NW_ACTIVE	Activate the network weights in job brokerage
OUTDISKCOUNT_ANALY_KB	Default outdiskcount for analysis jobs in kB
OVERLOAD_MIN_QUEUE	Minimum number of queued jobs to skip overloaded PQs
OVERLOAD_RATIO_OFFSET	Offset for nQ/nR ratio to skip overloaded PQs
PENDING_TIMEOUT_user	Timeout value for pending tasks in hours
PROD_TASKS_PUSH_STATUS_CHANGES_PERCENT	Percentage of production tasks which enable push messages of status changes
RAM_THR_EXAUSTED	Threshold for ramcount/core. Tasks are set to exhausted when their actual ramcount/core is larger than the threshold and preset ramcount/core is less than the threshold
SCOUT_CHANGE_SR_user	Allow scouts to change splitRule for user tasks
SCOUT_CPUTIME_RANK	Percentile rank for cpuTime calculation based on scouts
SCOUT_DISK_IO_CAP	Cap on disk IO in kBPerS measured by scout jobs
SCOUT_LOW_CPU_EFFICIENCY_user	Upper limit on CPU efficiencies of inefficient scout jobs for analysis (in %)
SCOUT_MAX_IO_INTENSITY_FOR_EXHAUSTED_user	Max IO intensity above which scout jobs are exempt from being sent to exhausted for analysis
SCOUT_MEM_LEAK_PER_CORE_<activity>	Maximum acceptable memory leak in kB/s for the given activity (SCOUT_MEM_LEAK_PER_CORE_managed, SCOUT_MEM_LEAK_PER_CORE_user)
SCOUT_MIN_OK_RATE_EXHAUSTED_user	Min scout success rate to send tasks to exhausted. Tasks below the threshold go to broken
SCOUT_NUM_CPU_INEFFICIENT_user	Max number of acceptable CPU-inefficient scout jobs for analysis
SCOUT_NUM_SHORT_user	Max number of acceptable short scout jobs for analysis
SCOUT_RAMCOUNT_MARGIN	Safety margin in % for ramCount calculation based on scouts
SCOUT_RAMCOUNT_MIN	Minimum value of RAM count set by scouts
SCOUT_RAMCOUNT_RANK	Percentile rank for ramCount calculation based on scouts
SCOUT_SHORT_EXECTIME_user	Upper limit on execution time of short scout jobs for analysis (in minutes)
SCOUT_THR_CPU_INEFFICIENT_user	Cutoff on the expected number of jobs to send analysis tasks to exhausted based on CPU-inefficient scout jobs
SCOUT_THR_SHORT_user	Cutoff on the expected number of jobs to send analysis tasks to exhausted based on short scout jobs
SCOUT_WRONG_CPUTIME_THRESHOLD	Tasks go to exhausted when actual cpuTime is larger than this threshold times the original cpuTime. 0 to disable
SIZE_CUTOFF_TO_MOVE_INPUT	Cutoff on the total size (GB) of missing input files to move for IO intensive tasks
SLOPPY_DISPATCH_RATIO	Ratio (%) of jobs to dispatch without considering gshare
STATIC_MAX_QUEUE_RUNNING_RATIO	Static max allowed nQ(PQ, user)/nR(PQ, user) ratio per user per PQ, before excluding the PQ in brokerage
SUPER_HIGH_PRIO_TASK_RATIO	Percentage of super high priority tasks to generate jobs in a single job generation cycle
TASK_MAX_ATTEMPT_user	Max number of attempts for a task. Tasks go to exhausted if a retry is requested and the number of attempts has reached the limit
TEST_JSON_INTERFACE	Not used; present to test the JSON interface
THROTTLE_THRESHOLD	Throttle threshold on Nqueue/Nrunning for job generation
THROTTLE_THRESHOLD_FOR_WORK_SHORTAGE	Throttle threshold on Nqueue/Nrunning for job generation in case of work shortage
TIMEOUT_defined	Timeout value for defined jobs in hours
TIMEOUT_holding	Timeout value for holding jobs in hours
TW_DONE_JOB_STAT	Finished/failed/closed jobs in the last N hours are considered in the analysis job brokerage decision
TYPNFILES_<activity>__<type>	Typical number of input files per job of the given type for the given activity (e.g. TYPNFILES_managed__archive)
USER_JOB_PRIO_BOOST_DICTS	JSON-serialized list of {"name": Name, "type": Type(user or group), "prio": Priority, "expire": ExpirationDate(YYYYMMDD or null)} defining users/groups that get a job priority boost
USER_JOB_TARGET_WALLTIME	Target walltime of user jobs in hours
USER_PRESTAGE_LIMIT	Max amount of data in GB to read from TAPE per user
USER_TASKS_MAX_CORE_COUNT	Maximum number of CPU cores per job that a single user task is allowed to utilize
USER_TASKS_MESSAGE_DRIVEN_PERCENT	Percentage of user tasks which enable message-driven mode
USER_TASKS_MOVE_INPUT	Percentage of user tasks which enable data motion for input
USER_TASKS_PUSH_STATUS_CHANGES_PERCENT	Percentage of user tasks which enable push messages of status changes
USER_TRANSFER_LIMIT	Max amount of transferring input data in GB per user
USER_USAGE_THRESHOLD_A	Threshold of user running slots in hi-sites for their tasks to stay in class A
USER_USAGE_THRESHOLD_B	Threshold of user running slots in hi and mid-sites for their tasks to stay in class B
WORK_SHORTAGE	Set to True when there is insufficient work available and jobs should be assigned exclusively to pledged resources