New handling of conversion from CSV/ZIP to Parquet via config file

This commit is contained in:
2025-11-19 17:00:07 +01:00
parent 2a5c4aec17
commit 6ab90b1a56
2 changed files with 133 additions and 64 deletions

View File

@@ -0,0 +1,62 @@
# SP2XR CSV/ZIP to Parquet Conversion Configuration Template
#
# This file contains all parameters for converting raw SP2XR data files
# to time-indexed Parquet format.
#
# USAGE:
# 1. Update the paths and parameters below for your dataset
# 2. Run: python scripts/sp2xr_csv2parquet.py --config my_conversion_config.yaml
#
# NOTES:
# - For local execution, the script auto-detects available CPU cores and memory
# - Output files are organized by date and hour: target_directory/date=YYYY-MM-DD/hour=HH/
# - Processing is parallelized using Dask for efficient handling of large datasets
# - You can monitor progress via the Dask dashboard (URL printed when script starts)
# Directory containing your raw SP2XR files (CSV or ZIP format)
source_directory: data/SP2XR_orig_files
# Output directory for converted Parquet files
target_directory: data/pbp_files_parquet_2
# Path to your data schema config file (generated by sp2xr_generate_config.py)
schema_config: config/config_schema_with_mapping.yaml
# Pattern to filter which files to process
# "PbP" for particle-by-particle data, "hk" for housekeeping data
file_filter: PbP
# Number of files to process in each batch
# Larger values = more memory usage but potentially faster
# Smaller values = less memory usage but more overhead
chunk_size: 100
# Execution mode: "local" or "slurm"
# - local: Use your local machine (laptop/desktop)
# - slurm: Use a SLURM cluster (HPC environment)
execution_mode: local
# --- SLURM-specific parameters (ignored if execution_mode: local) ---
# Number of CPU cores per SLURM job
slurm_cores: 64
# Memory per SLURM job (e.g., "128GB", "256GB")
slurm_memory: 128GB
# SLURM partition to use
# Common values: "hourly", "daily", "general"
slurm_partition: daily
# Walltime for SLURM job (e.g., "01:00:00" for 1 hour, "23:59:00" for 1 day)
# If not specified, defaults based on partition:
# - hourly: 00:59:00
# - daily: 23:59:00
# - general: 7-00:00:00
slurm_walltime: null

View File

@@ -4,6 +4,8 @@ import gc
import dask
import multiprocessing
import psutil
import yaml
from pathlib import Path
from dask.distributed import Client, LocalCluster
from dask_jobqueue import SLURMCluster
from dask import delayed
@@ -13,58 +15,63 @@ from sp2xr.helpers import find_files, chunks
def main():
parser = argparse.ArgumentParser(
description="Batch convert SP2XR zip to parquet using Dask (local or SLURM cluster)"
description="Batch convert SP2XR zip to parquet using Dask (local or SLURM cluster)",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Example:
python sp2xr_csv2parquet.py --config my_conversion_config.yaml
See config/conversion_config_template.yaml for configuration options.
""",
)
parser.add_argument(
"--source", required=True, help="Directory containing input zip/CSV files"
)
parser.add_argument(
"--target", required=True, help="Output directory for parquet files"
)
parser.add_argument(
"--config", required=True, help="Path to YAML schema config file"
)
parser.add_argument(
"--filter", default="PbP", help="Pattern to filter files (default: PbP)"
)
parser.add_argument(
"--chunk",
type=int,
default=100,
help="Number of files per batch (default: 100)",
)
parser.add_argument(
"--local",
action="store_true",
help="Use local Dask client (auto-detects resources) instead of SLURM cluster",
)
parser.add_argument(
"--cores",
type=int,
default=64,
help="Cores per SLURM job (ignored for --local, default: 64)",
)
parser.add_argument(
"--memory",
default="128GB",
help="Memory per SLURM job (ignored for --local, default: 128GB)",
)
parser.add_argument(
"--walltime",
default=None,
help="Walltime for SLURM (ignored for --local, default depends on partition)",
)
parser.add_argument(
"--partition",
default="daily",
help="SLURM partition (ignored for --local, default: daily)",
"--config",
required=True,
help="Path to conversion config YAML file",
)
args = parser.parse_args()
# Load config from file
config_path = Path(args.config)
if not config_path.exists():
print(f"Error: Config file not found: {config_path}")
return 1
with open(config_path, "r") as f:
config = yaml.safe_load(f)
print(f"Loaded conversion config from: {config_path}")
# Extract parameters from config
source = config.get("source_directory")
target = config.get("target_directory")
schema_config = config.get("schema_config")
file_filter = config.get("file_filter", "PbP")
chunk_size = config.get("chunk_size", 100)
execution_mode = config.get("execution_mode", "local")
# SLURM parameters
slurm_cores = config.get("slurm_cores", 64)
slurm_memory = config.get("slurm_memory", "128GB")
slurm_partition = config.get("slurm_partition", "daily")
slurm_walltime = config.get("slurm_walltime")
# Validate required parameters
if not source or not target or not schema_config:
print("Error: Missing required parameters in config file:")
print(" - source_directory")
print(" - target_directory")
print(" - schema_config")
return 1
use_local = execution_mode == "local"
# --- Setup cluster (local or SLURM) ---
start_time = time.time()
if args.local:
if use_local:
# Local execution: auto-detect resources
total_cores = multiprocessing.cpu_count()
total_memory = psutil.virtual_memory().total # in bytes
@@ -83,26 +90,26 @@ def main():
print(f"Dask dashboard: {client.dashboard_link}")
else:
# SLURM execution
if args.walltime is None:
if args.partition == "hourly":
args.walltime = "00:59:00"
elif args.partition == "daily":
args.walltime = "23:59:00"
elif args.partition == "general":
args.walltime = "7-00:00:00"
if slurm_walltime is None:
if slurm_partition == "hourly":
slurm_walltime = "00:59:00"
elif slurm_partition == "daily":
slurm_walltime = "23:59:00"
elif slurm_partition == "general":
slurm_walltime = "7-00:00:00"
else:
args.walltime = "00:59:00"
slurm_walltime = "00:59:00"
print("Running in SLURM mode")
print(f"Resources: {args.cores} cores, {args.memory} memory")
print(f"Partition: {args.partition}, Walltime: {args.walltime}")
print(f"Resources: {slurm_cores} cores, {slurm_memory} memory")
print(f"Partition: {slurm_partition}, Walltime: {slurm_walltime}")
cluster = SLURMCluster(
cores=args.cores,
processes=args.cores,
memory=args.memory,
walltime=args.walltime,
job_extra_directives=[f"--partition={args.partition}"],
cores=slurm_cores,
processes=slurm_cores,
memory=slurm_memory,
walltime=slurm_walltime,
job_extra_directives=[f"--partition={slurm_partition}"],
)
cluster.scale(1)
client = Client(cluster)
@@ -110,19 +117,19 @@ def main():
try:
# --- Find files ---
files = find_files(args.source, args.filter)
print(f"Found {len(files)} files matching pattern '{args.filter}'")
files = find_files(source, file_filter)
print(f"Found {len(files)} files matching pattern '{file_filter}'")
if len(files) == 0:
print("No files found matching the filter pattern")
return
# --- Process in chunks ---
i = 0
total_chunks = (len(files) - 1) // args.chunk + 1
for chunk in chunks(files, args.chunk):
total_chunks = (len(files) - 1) // chunk_size + 1
for chunk in chunks(files, chunk_size):
print(f"Processing chunk {i+1} / {total_chunks}")
tasks = [
delayed(process_sp2xr_file)(f, args.config, args.target) for f in chunk
delayed(process_sp2xr_file)(f, schema_config, target) for f in chunk
]
dask.compute(*tasks)
gc.collect()