Advanced Parameters

Juice offers several advanced parameters to fine-tune your GPU sessions and application runs:

Run Command Options

• --disable-cache

  Prevents the system from using previously cached data for the current run. Use when you need to ensure fresh data for each run, e.g., when testing data-dependent applications or when data integrity is crucial.

  Example:

  juice run --disable-cache my_data_processing_app.py 

• --disable-compression

  Turns off data compression between the client and the GPU agent. Useful for applications sensitive to data integrity or when network bandwidth isn't a bottleneck. Can also help in diagnosing data transfer issues.

  Example:

  juice run --disable-compression my_high_fidelity_image_processing.py 

• --ephemeral

  Creates a new session for the run and immediately releases it after completion. Ideal for one-off tasks or when you need a clean state for each run, preventing any potential interference from previous sessions.

  Example:

  juice run --ephemeral my_benchmarking_script.py 

• --match-labels <string>

  Filters available GPUs based on specified labels, ensuring the task runs on a GPU with matching attributes. Use when you need to select GPUs with specific characteristics (e.g., GPU type, installed software).

  Example:

  juice run --match-labels "gpu-type=v100,env=pytorch" my_ml_model.py 

• --no-version-check

  Bypasses the version compatibility check between the client and the GPU agent. Use cautiously when you need to run on potentially incompatible versions, typically in testing environments.

  Example:

  juice run --no-version-check my_experimental_app.py 

• --on-connection-error <string>

  Specifies the behavior when a connection error occurs. Options: [fail, continue] Use in scripts or automated workflows where you want to control how the system responds to connection issues.

  Example:

  juice run --on-connection-error continue my_resilient_app.py 

• --on-queue-timeout <string>

  Defines the action to take when the queue time exceeds the specified limit. Options: [fail, continue] Use in time-sensitive scenarios or when you need to control the behavior of long-waiting jobs.

  Example:

  juice run --on-queue-timeout continue --queue-timeout 300 my_long_running_job.py 

• --pool-ids <string>

  Restricts GPU selection to specific pools identified by their IDs. Use for resource management in multi-team setups or when you need to use GPUs from particular pools.

  Example:

  juice run --pool-ids pool1,pool2 my_app.py 

• --queue-timeout <uint>

  Sets the maximum time (in seconds) to wait for GPU availability. Use to prevent indefinite waits in resource-constrained environments or for time-sensitive applications.

  Example:

  juice run --queue-timeout 600 my_time_sensitive_app.py 

• --tolerates <string>

  Specifies which taints the session can tolerate, allowing use of GPUs with specific constraints. Use when you need to use GPUs that have been marked with certain restrictions or characteristics.

  Example:

  juice run --tolerates "high-priority=true" my_critical_task.py

Global Command Options

• --log-file

  Directs CLI output to a specified file instead of standard output. Useful for debugging, creating audit trails, or when you need to review operation logs later.

  Example: juice --log-file=/path/to/juice.log run my_app.py

• --log-level

  Sets the verbosity level of logging. Options: debug, info, warn, error, fatal. Adjust for more detailed logs when troubleshooting or less verbose output in production.

  Example:

  juice --log-level debug run my_problematic_app.py 

• --quiet

  Suppresses most output from the CLI. Use in scripts or automated environments where you want to minimize output.

  Example:

  juice --quiet run my_background_task.py 

• --no-banner

  Removes the startup banner from the CLI output. Use in scripts or when you need cleaner output without the banner information.

  Example:

  juice --no-banner run my_app.py 

• --controller

  Specifies a non-default controller URL for the Juice system. Use in complex setups with multiple Juice environments or when using a custom controller.

  Example:

  juice --controller=https://custom-controller.example.com run my_app.py 

These advanced parameters provide fine-grained control over Juice's behavior, allowing users to optimize their workflows for specific use cases and environments.