Allow selection of specific cgroup params for cgroup metrics plugin

config/job_metrics_conf.xml.sample

@@ -38,9 +38,15 @@
   <!-- <env variables="HOSTNAME,SLURM_CPUS_ON_NODE,SLURM_JOBID" /> -->
 
   <!-- If galaxy jobs are run in cgroups, like slurm does if memory limits
-       are enforced, we can try to grep some information from this. -->
+       are enforced, we can try to grep some information from this. By default,
+       only a small set of cgroup parameters will be recorded, the list of which
+       can be found in lib/galaxy/jobs/metrics/instrumenters/cgroup.py.  -->
   <!-- <cgroup /> -->
+  <!-- All params from the `cpuacct,cpu` and `memory` controllers can be
+       recorded.  -->
   <!-- <cgroup verbose="true" /> -->
+  <!-- Or, specific params can be recorded. -->
+  <!-- <cgroup params="cpuacct.usage,memory.max_usage_in_bytes,memory.memsw.max_usage_in_bytes" /> -->
 
   <!-- Uncomment to record hostname - *nix only -->
   <!-- <hostname /> -->

lib/galaxy/jobs/metrics/instrumenters/cgroup.py

@@ -54,6 +54,12 @@ class CgroupPlugin(InstrumentPlugin):
 
     def __init__(self, **kwargs):
         self.verbose = asbool(kwargs.get("verbose", False))
+        params_str = kwargs.get("params", None)
+        if params_str:
+            params = [v.strip() for v in params_str.split(",")]
+        else:
+            params = TITLES.keys()
+        self.params = params
 
     def post_execute_instrument(self, job_directory):
         commands = []
@@ -92,7 +98,7 @@ def __read_metrics(self, path):
         return metrics
 
     def __add_metric(self, metrics, metric):
-        if metric and (metric.subkey in TITLES or self.verbose):
+        if metric and (metric.subkey in self.params or self.verbose):
             metrics[metric.subkey] = metric.value
 
     def __read_key_value(self, line, prev_metric):