Merge pull request #104 from NessieCanCode/check-slurmctld.service-for-config-path

Discover slurm config path from service file

Author: NessieCanCode

README.md

@@ -83,12 +83,14 @@ The workflow tags the repository, builds packages, and publishes artifacts to Gi
 
 The `src/slurmdb.py` utility can connect to a running **SlurmDBD** instance and
 export usage metrics as JSON. Connection details are automatically scraped from
-`/etc/slurm/slurmdbd.conf` (or a custom path specified via the environment
-variable `SLURMDB_CONF` or the `--conf` flag). Environment variables
+`slurmdbd.conf` located alongside `slurm.conf` (discovered from
+`slurmctld.service` via the `ConditionPathExists` directive, defaulting to
+`/etc/slurm/slurm.conf`). A custom path can be specified via the environment
+variable `SLURMDB_CONF` or the `--conf` flag. Environment variables
 `SLURMDB_HOST`, `SLURMDB_PORT`, `SLURMDB_USER`, `SLURMDB_PASS` and `SLURMDB_DB`
 override any values found in the configuration file. The cluster prefix used to
-select the job tables is determined from `/etc/slurm/slurm.conf` but can be set
-using `SLURM_CLUSTER`, `--cluster` or `--slurm-conf`.
+select the job tables is determined from the Slurm configuration file. The
+setting can be overridden using `SLURM_CLUSTER`, `--cluster` or `--slurm-conf`.
 
 
 ```bash

src/slurmdb.py

@@ -63,6 +63,25 @@ def _write_last_run(end_date):
         logging.warning("Failed to write state file %s: %s", STATE_FILE, e)
 
 
+def _find_slurm_conf(service_paths=None):
+    """Return path to slurm.conf by inspecting slurmctld.service."""
+    paths = service_paths or [
+        "/usr/lib/systemd/system/slurmctld.service",
+        "/lib/systemd/system/slurmctld.service",
+        "/etc/systemd/system/slurmctld.service",
+    ]
+    for svc in paths:
+        try:
+            with open(svc) as fh:
+                for line in fh:
+                    line = line.strip()
+                    if line.startswith("ConditionPathExists=") and line.endswith("slurm.conf"):
+                        return line.split("=", 1)[1].strip()
+        except OSError:
+            continue
+    return "/etc/slurm/slurm.conf"
+
+
 class SlurmDB:
     """Simple wrapper around the Slurm accounting database."""
 
@@ -77,7 +96,17 @@ def __init__(
         cluster=None,
         slurm_conf=None,
     ):
-        conf_path = config_file or os.environ.get("SLURMDB_CONF", "/etc/slurm/slurmdbd.conf")
+        slurm_conf_path = (
+            slurm_conf
+            or os.environ.get("SLURM_CONF")
+            or _find_slurm_conf()
+        )
+
+        conf_path = (
+            config_file
+            or os.environ.get("SLURMDB_CONF")
+            or os.path.join(os.path.dirname(slurm_conf_path), "slurmdbd.conf")
+        )
         cfg = self._load_config(conf_path)
 
         self.host = host or os.environ.get("SLURMDB_HOST") or cfg.get("host", "localhost")
@@ -88,7 +117,7 @@ def __init__(
         self._conn = None
         self._tres_map = None
         self._config_file = conf_path
-        self._slurm_conf = slurm_conf or os.environ.get("SLURM_CONF", "/etc/slurm/slurm.conf")
+        self._slurm_conf = slurm_conf_path
         self.cluster = (
             cluster
             or os.environ.get("SLURM_CLUSTER")

test/unit/slurmdb_validation.test.py

@@ -1,7 +1,9 @@
 import unittest
 import json
+import os
+import tempfile
 import pymysql
-from slurmdb import SlurmDB
+from slurmdb import SlurmDB, _find_slurm_conf
 from slurm_schema import extract_schema, extract_schema_from_dump
 
 class SlurmDBValidationTests(unittest.TestCase):
@@ -17,6 +19,22 @@ def test_valid_config_allowed(self):
         db = SlurmDB(host="localhost", port=3306, user="slurm", password="", database="slurm_acct_db", cluster="cluster1")
         self.assertEqual(db.cluster, "cluster1")
 
+    def test_slurm_conf_from_service_file(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            svc = os.path.join(tmp, "slurmctld.service")
+            slurm_conf = os.path.join(tmp, "slurm.conf")
+            with open(svc, "w") as fh:
+                fh.write(f"ConditionPathExists={slurm_conf}\n")
+            with open(slurm_conf, "w") as fh:
+                fh.write("ClusterName=test\n")
+            slurmdbd = os.path.join(tmp, "slurmdbd.conf")
+            with open(slurmdbd, "w") as fh:
+                fh.write("StorageHost=localhost\n")
+            path = _find_slurm_conf([svc])
+            self.assertEqual(path, slurm_conf)
+            db = SlurmDB(slurm_conf=path)
+            self.assertEqual(db._config_file, slurmdbd)
+
     def test_invalid_time_format(self):
         db = SlurmDB()
         with self.assertRaises(ValueError):