[BUG] Error while inferring cluster when cluster properties file is provided to the CLI

[parthosa]

Describe the bug
When running the Tools CLI on Dataproc event logs with a custom cluster info file, an error occurs in the cluster inference module. This is probably due to some of recent changes to the cluster information on the Scala side, which are incompatible with the Python implementation.

Steps/Code to reproduce bug

spark_rapids qualification --platform dataproc --cluster test-cluster-info.json --eventlogs gs://spark-events --tools_jar <tools_jar> --verbose 

Cluster Info File: test-cluster-info.json

Output

2024-10-23 13:43:38,313 ERROR rapids.tools.cluster_inference: Error while inferring cluster: cannot convert float NaN to integer
2024-10-23 13:43:38,313 INFO rapids.tools.cluster_inference: For App ID: application_1717064107684_0009, Unable to infer CPU cluster. Reason - No matching worker node found for num cores = -4.0
2024-10-23 13:43:38,314 INFO rapids.tools.cluster_inference: For App ID: application_1717064107684_0008, Unable to infer CPU cluster. Reason - No matching worker node found for num cores = -4.0
2024-10-23 13:43:38,314 INFO rapids.tools.cluster_inference: For App ID: application_1717064107684_0007, Unable to infer CPU cluster. Reason - No matching worker node found for num cores = -4.0
2024-10-23 13:43:38,314 INFO rapids.tools.cluster_inference: For App ID: application_1717064107684_0006, Unable to infer CPU cluster. Reason - No matching worker node found for num cores = -4.0

[cindyyuanjiang]

Does the negative num cores = -4.0 come from Scala output? If not, where does this number come from?

[parthosa]

Total Cores is calculated as:

spark-rapids-tools/user_tools/src/spark_rapids_pytools/common/cluster_inference.py

Lines 79 to 81 in 1504968

	cores_per_executor = cluster_info_df.get('Cores Per Executor')
	execs_per_node = cluster_info_df.get('Num Executors Per Node')
	total_cores_per_node = execs_per_node * cores_per_executor

Scala output would generate Num Executors Per Node = -1. This will cause Total Cores = -4

[cindyyuanjiang]

Scala output generates Num Executors Per Node = -1 causing Total Cores to be -4

Thanks for clarifying!