Add hack for obtaining cgroup stats

Docker mounts cgroup stats inconsistently with the hierarchy 
in /proc/self/cgroup. This breaks the logic for traversing
from /proc/self/cgroup to the stats for a process running inside a
container. In core Elasticsearch, an undocumented hack was added to
support this, for now. Thus, for cgroup stats to be available from
Elasticsearch running inside a Docker container, the Elasticsearch
process should be started with `-Des.cgroups.hierarchy.override=/`.

Add hack for obtaining cgroup stats and comment it.

Relates #25 
Relates elastic/elasticsearch#22757

build/elasticsearch/bin/es-docker

@@ -22,4 +22,18 @@ do
     fi
 done < <(env)
 
+# The virtual file /proc/self/cgroup should list the current cgroup
+# membership. For each hierarchy, you can follow the cgroup path from
+# this file to the cgroup filesystem (usually /sys/fs/cgroup/) and
+# introspect the statistics for the cgroup for the given
+# hierarchy. Alas, Docker breaks this by mounting the container
+# statistics at the root while leaving the cgroup paths as the actual
+# paths. Therefore, Elasticsearch provides a mechanism to override
+# reading the cgroup path from /proc/self/cgroup and instead uses the
+# cgroup path defined the JVM system property
+# es.cgroups.hierarchy.override. Therefore, we set this value here so
+# that cgroup statistics are available for the container this process
+# will run in.
+export ES_JAVA_OPTS="-Des.cgroups.hierarchy.override=/ $ES_JAVA_OPTS"
+
 exec bin/elasticsearch ${es_opts}