change sacct delimiter for safer parsing

slurm-eff-tool.py

@@ -63,6 +63,8 @@ VERSION=0.3
 # Default GB per CPUs for the cluster
 default_mempercpu_gb = 2.2
 
+SACCT_DELIM = "\x1f"   # ASCII Unit Separator
+
 SACCT_FIELDS = [
     "JobIDRaw",
     "JobID",
@@ -667,11 +669,13 @@ def pct(numerator: float | None, denominator: float | None) -> float | None:
         return None
     return 100.0 * numerator / denominator
 
-
+#TODO: Put the parsing on safer ground. The delimiter based parsing may
+#      fail in exotic cases. Better to use json (but slower)
 def run_sacct(args: argparse.Namespace) -> list[dict[str, str]]:
     cmd = [
         "sacct",
         "-P",
+        f"--delimiter={SACCT_DELIM}",
         "-n",
         "--units=K",
         "--format=" + ",".join(SACCT_FIELDS),
@@ -686,6 +690,8 @@ def run_sacct(args: argparse.Namespace) -> list[dict[str, str]]:
     if args.state:
         cmd += ["--state", args.state]
 
+    # print(f"DEBUG: sacct cmd: {cmd}")
+
     # Include job steps because MaxRSS often lives on batch/extern/step rows.
     # We later collapse rows back to the base job ID.
     try:
@@ -701,7 +707,7 @@ def run_sacct(args: argparse.Namespace) -> list[dict[str, str]]:
 
 def parse_pipe_rows(lines: Iterable[str]) -> list[dict[str, str]]:
     rows: list[dict[str, str]] = []
-    reader = csv.reader(lines, delimiter="|")
+    reader = csv.reader(lines, delimiter=SACCT_DELIM)
     for parts in reader:
         if not parts:
             continue
@@ -719,7 +725,7 @@ def write_cache_raw(path: str, rows: list[dict[str, str]]) -> None:
         sys.exit(1)
     
     with open(path, "w", newline="", encoding="utf-8") as f:
-        writer = csv.DictWriter(f, fieldnames=SACCT_FIELDS, delimiter="|", lineterminator="\n")
+        writer = csv.DictWriter(f, fieldnames=SACCT_FIELDS, delimiter=SACCT_DELIM, lineterminator="\n")
         writer.writeheader()
         writer.writerows(rows)
 
@@ -727,7 +733,7 @@ def write_cache_raw(path: str, rows: list[dict[str, str]]) -> None:
 def read_cache_raw(path: str) -> list[dict[str, str]]:
     """Read a raw sacct output file."""
     with open(path, newline="", encoding="utf-8") as f:
-        reader = csv.DictReader(f, delimiter="|")
+        reader = csv.DictReader(f, delimiter=SACCT_DELIM)
         missing = [x for x in SACCT_FIELDS if x not in (reader.fieldnames or [])]
         # Older cache files from versions before CPUTimeRAW can still be read.
         # CPU efficiency will be NA for those rows unless CPUTimeRAW is present.