Use of java.util.Random to generate seed causes failure

[MattWellie]

gatk/src/main/java/org/broadinstitute/hellbender/tools/copynumber/GermlineCNVCaller.java

Lines 367 to 384 in 3cbea75

	ProcessOutput pythonProcessOutput = executor.executeScriptAndGetOutput(
	new Resource(script, GermlineCNVCaller.class),
	null,
	composePythonArguments(intervalSubsetReadCountFiles, STARTING_SEED));
	if (pythonProcessOutput.getExitValue() != 0) {
	// We restart once if the inference diverged
	if (pythonProcessOutput.getExitValue() == DIVERGED_INFERENCE_EXIT_CODE) {
	final Random generator = new Random(STARTING_SEED);
	final int nextGCNVSeed = generator.nextInt();
	logger.info("The inference failed to converge and will be restarted once with a different random seed.");
	pythonProcessOutput = executor.executeScriptAndGetOutput(
	new Resource(script, GermlineCNVCaller.class),
	null,
	composePythonArguments(intervalSubsetReadCountFiles, nextGCNVSeed));
	} else {
	throw executor.getScriptException(executor.getExceptionMessageFromScriptError(pythonProcessOutput));
	}
	}

Based on the documentation here the use of Random.nextInt() returns any integer, so has a 50% chance of returning a negative value. Negative values are not acceptable arguments to the python script this process is preparing arguments for:

(from GATK-SV logs)

01:26:00.580 INFO  GermlineCNVCaller - The inference failed to converge and will be restarted once with a different random seed.
...
01:26:00.586 DEBUG ScriptExecutor -   /mnt/disks/cromwell_root/tmp.MOkT0J/case_denoising_calling.6807924686880120959.py
01:26:00.586 DEBUG ScriptExecutor -   --ploidy_calls_path=/mnt/disks/cromwell_root/contig-ploidy-calls-dir
01:26:00.586 DEBUG ScriptExecutor -   --output_calls_path=/mnt/disks/cromwell_root/out/case-calls
01:26:00.586 DEBUG ScriptExecutor -   --output_tracking_path=/mnt/disks/cromwell_root/out/case-tracking
01:26:00.586 DEBUG ScriptExecutor -   --input_model_path=/mnt/disks/cromwell_root/gcnv-model
01:26:00.586 DEBUG ScriptExecutor -   --random_seed=-1623339239
01:26:00.586 DEBUG ScriptExecutor -   --read_count_tsv_files
....

Traceback (most recent call last):
  File "/mnt/disks/cromwell_root/tmp.MOkT0J/case_denoising_calling.6807924686880120959.py", line 199, in <module>
    task = gcnvkernel.CaseDenoisingCallingTask(
  File "/opt/miniconda/envs/gatk/lib/python3.10/site-packages/gcnvkernel/tasks/task_case_denoising_calling.py", line 120, in __init__
    super().__init__(hybrid_inference_params, denoising_model, copy_number_emission_sampler, copy_number_caller,
  File "/opt/miniconda/envs/gatk/lib/python3.10/site-packages/gcnvkernel/tasks/inference_task_base.py", line 279, in __init__
    self.continuous_model_advi = ADVIDeterministicAnnealing(
  File "/opt/miniconda/envs/gatk/lib/python3.10/site-packages/gcnvkernel/inference/deterministic_annealing.py", line 47, in __init__
    approx = MeanField(local_rv=local_rv,
  File "/opt/miniconda/envs/gatk/lib/python3.10/site-packages/pymc/variational/approximations.py", line 338, in __init__
    groups = [self._group_class(None, *args, **kwargs)]
  File "/opt/miniconda/envs/gatk/lib/python3.10/site-packages/pymc/variational/opvi.py", line 746, in __init__
    self.rng = np.random.RandomState(random_seed)
  File "numpy/random/mtrand.pyx", line 185, in numpy.random.mtrand.RandomState.__init__
  File "_mt19937.pyx", line 168, in numpy.random._mt19937.MT19937._legacy_seeding
  File "_mt19937.pyx", line 182, in numpy.random._mt19937.MT19937._legacy_seeding
ValueError: Seed must be between 0 and 2**32 - 1
01:26:20.754 INFO  GermlineCNVCaller - Shutting down engine
[January 29, 2026 at 1:26:20 AM GMT] org.broadinstitute.hellbender.tools.copynumber.GermlineCNVCaller done. Elapsed time: 14.50 minutes.
Runtime.totalMemory()=5968494592
org.broadinstitute.hellbender.utils.python.PythonScriptExecutorException: 
python exited with 1

Swapping this out for Random.nextInt(int Bound) should give a guaranteed positive value. Or running an abs(input) on the input seed before it's passed through to the gcnvkernel module.

[MattWellie]

Given that the initial seed is always 1984, and 1984 is then used to seed the random process (so the second random is always the same?) could you remove the random call and just use 1984 and 1985 as seeds?

Alternatively a bounded random call, without seeding Random would give this a shot at producing a fresh state with each run.