Changing environmental variables for the Watson OpenScale service

Although the Watson™ OpenScale service works with default values, you can adjust some of the environmental variables to better meet your needs or processing environment constraints.

Changing the environmental variables for the Watson OpenScale service

You can check for the current setting of the environmental variables and make updates by issuing `oc` commands from a command line interface.

  1. Before you update a variable, to find out what the current value is, run the following command:
    oc describe deployment aiopenscale-ibm-aios-microserve_name | grep -i environmental_variable
  2. To update a variable, run the following command:
    oc patch woservice aiopenscale -p '{"spec": {"microserve_name": {"environmental_variable":"value"} }}' --type=merge

    microserve_name is the name of the specific micro service that uses the environmental variable. The possible values are bias or mrm.

    environmental_variable is the name of the environmental variable that you want to update. Only certain variables can be updated by users.

    value is the value to which you are updating the environmental variable. The value is dependent on the data type and can be Boolean, enumerated, string, or numeric.

Environmental variables for the bias micro service

The bias environmental variables control whether to perform perturbation as part of bias checking and how often and at what intervals to perform various query processes.

The following sample command shows an update for the quality_run_repeat_interval variable:
oc patch woservice aiopenscale -p '{"spec": {"bias": {"quality_run_repeat_interval":"10"} }}' --type=merge
Environmental variable Default Range Data type and description
bias_checking_records_max_limit 1000000 1 to any positive integer Integer. The maximum possible value of the minimum sample size for which fairness can be configured.
max_records_upper_limit 500000 1 to any positive integer Integer. The maximum possible value of the max_records variable for which fairness can be configured.
max_de_bias_data_size 10000 1 to any positive integer Integer. The maximum possible number of payload logging records that is used for debiasing.
perform_bias_check_in_paged_manner False True, False (Case insensitive) String internally converted to a boolean flag to enable/disable bias checking in-paged manner during a no-perturbation scenario
payload_logging_api_records_max_page_size 1000 1 to any positive integer Integer. The hard limit of the page size of the data set API.
perform_perturbation False True | False, case insensitive String internally converted to Boolean. Whether perturbation for bias check should be used or not.
quality_run_repeat_interval 10 1 to any positive integer Integer. Define the number of units of repeat type a schedule should wait between each call to check for Quality Run spark job status.
quality_config_repeat_interval 3 1 to any positive integer Integer. Define the number of units of repeat type a schedule should wait between each call to check for Quality Config spark job status.
quality_schedule_max_invocations 50 1 to any positive integer Integer. The maximum number of times a schedule should poll for Quality Config and Quality Run spark jobs to check the job status.
quality_schedule_repeat_type minute minute, hour, day, week, month String. The unit of time used for the repeat interval.
spark_job_schedule_repeat_interval 5 1 to any positive integer Integer. Define the number of units of repeat type a schedule should wait between each call to check for Spark job status.
spark_job_schedule_repeat_type minute minute, hour, day, week, month String. The unit of time used for the repeat interval.
spark_job_schedule_max_invocations 100 1 to any positive integer Integer. The maximum number of times a schedule should poll for Spark job status before giving up.
spark_job_polling_interval 10 1 to any positive integer Integer. The number of seconds to wait between each polling call for Spark jobs that do not use a schedule, like the table validation job.
spark_job_polling_iterations 30 1 to any positive integer Integer. The number of times the polling should be done for Spark job status where a schedule is not used.

Environmental variables for the configuration micro service

The configuration environmental variables control the settings that are applied for the configuration micro service.

The following command shows an update for the enable_quota_support variable:
oc patch woservice aiopenscale -p '{"spec": {"configuration": {"enable_quota_support":"false"} }}' --type=merge
Environmental variable Default Range Data type and description
enable_quota_support true true, false Boolean. Specifies if quota support is enforced or not.
additional_jvm_options N/A N/A String. Specifies parameters to change quota configuration.

Environmental variables for the drift micro service

The drift environmental variables control the default limitations on the size of the data sets for the drift micro service.

The following command shows an update for the data_drift_column_limit variable:
oc patch woservice aiopenscale -p '{"spec": {"drift": {"data_drift_column_limit":"150"} }}' --type=merge
Environmental variable Default Range Data type and description
annotations_payload_limit 10000 1 to any positive integer Integer. Defines the limit on the number of records to patch with drift annotations by request.
data_drift_column_limit 200 1 to any positive integer Integer. Defines the limit for the learning of two-column constraints. If the number of columns is more than this value, two-column constraints are not learned. Only applicable to online training of drift model/learning of data constraints.
payload_provider_limit 1000 1 to any positive integer Integer. Defines the limit on the number of records to retrieve from the payload logging service by request.

Environmental variables for the explainability micro service

The explainability environmental variables control how often and at what interval to schedule batch jobs for the explainability micro service.

The following command shows an update for the max_explanation_tasks variable:
oc patch woservice aiopenscale -p '{"spec": {"explainability": {"max_explanation_tasks":"20"} }}' --type=merge
Environmental variable Default Range Data type and description
batch_job_poll_schedule_repeat_interval 15 1 to any positive integer Integer. The polling interval for checking the explanations computation batch job status.
batch_job_poll_schedule_max_invocations 480 1 to any positive integer Integer. The maximum number of invocations that the polling schedule runs.
batch_job_poll_schedule_repeat_type minute minute, second String. The unit type of the batch polling interval.
max_explanation_tasks 10 1 to any positive integer Integer. The maximum number of explanations(scoring_ids) that can be triggered/specified in the POST explanation tasks API.
single_feature_pn_default true true, false String. Flag to indicate whether to use the single feature method for computing the pertinent negative in a contrastive explanation
sync_batch_job_timeout 900 1 to any positive integer Integer. The maximum number of seconds the explainability service waits for the synchronous batch jobs like Explainability configuration, the Get explanations job and the Adhoc explanations job.

Environmental variables for the feedback micro service

The feedback environmental variables control the default limitations on the size of the data for the feedback micro service.

The following sample command shows an update for the max_records variable:
oc patch woservice aiopenscale -p '{"spec": {"feedback" {"max_records":"12000"} }}' --type=merge
Environmental variable Default Range Data type and description
max_records 10000 1 to any positive integer Integer. The maximum number of records for which quality metrics can be evaluated.
enable_last_run_time_window false true, false String. Flag to indicate whether to process the records since last run time during the quality run evaluation

Environmental variables for the mrm micro service

The mrm environmental variables control how often and at what interval to verify payload data for the model risk management micro service.

The following sample command shows an update for the get_payload_data_size_retries variable:
oc patch woservice aiopenscale -p '{"spec": {"mrm": {"get_payload_data_size_retries":"10"} }}' --type=merge
Environmental variable Default Range Data type and description
get_payload_data_size_retries 18 1 to any positive integer Integer. Define the number of retries to verify whether test data records reached the payload logging table.
get_payload_data_size_interval 10 1 to any positive integer Integer. Define the sleep time (in seconds) between the checks to verify whether test data records reached the payload logging table.
get_online_monitor_run_status_retries 10 1 to any positive integer Integer. Define the number of retries to check the status of the monitor run completion for an online subscription.
get_online_monitor_run_status_interval 30 1 to any positive integer Integer. Define the sleep time (in seconds) between the checks to verify the status of the monitor run completion for an online subscription.
get_batch_monitor_run_status_retries 15 1 to any positive integer Integer. Define the number of retries to check the status of the monitor run completion for a batch subscription.
get_batch_monitor_run_status_interval 60 1 to any positive integer Integer. Define the sleep time (in seconds) between the checks to verify the status of the monitor run completion for a batch subscription.

Environment variables for the payload_logging_api micro service

The payload-logging-api environment variables control the default limitations on data set APIs, such as the maximum page size.

The following sample command shows an update for the payload_logging_api_records_max_page_size variable:
oc patch woservice aiopenscale -p '{"spec": {"payload_logging_api": {"payload_logging_api_records_max_page_size":"2000"} }}' --type=merge
Environment variable Default Range Data type and description
payload_logging_api_records_max_page_size 1000 1 to any positive integer Integer. The hard limit of the page size of the data set API.
akka_server_max_content_length 100m 1 to any positive integer byte(b), Kbyte(k), and Mbyte(m) String. The maximum content length that the server can receive.
akka_server_request_timeout 60s 1 to any positive integer second(s), minute(m), and hour(h) String. The timeout to produce an HttpResponse. It must be less than the akka_server_idle_timeout variable.
akka_server_idle_timeout 70s 1 to any positive integer second(s), minute(m), and hour(h) String. The time after which an idle connection will be automatically closed.
database_client_num_threads 10 1 to any positive integer Integer. The upper limit of threads for the database connection.
database_client_connection_timeout 30s 1 to any positive integer second(s) and minute(m) String. The amount of time to wait before getting a connection timeout.