From 9e0e2f82c98b238d574ca6870f0fbece39c3bb55 Mon Sep 17 00:00:00 2001
From: Gibson Chikafa <gibbschikafa@gmail.com>
Date: Mon, 24 Apr 2023 12:03:10 +0200
Subject: [PATCH] [HWORKS-512] Read cgroup parent from database (#1347) (#1329)

* Pyhton resources fixes

* Remove empty line

* Take care of the .

* Rename variable
---
 .../PythonResourcesController.java            | 23 +++++++++++++------
 .../hops/hopsworks/common/util/Settings.java  |  9 ++++++++
 2 files changed, 25 insertions(+), 7 deletions(-)

diff --git a/hopsworks-common/src/main/java/io/hops/hopsworks/common/pythonresources/PythonResourcesController.java b/hopsworks-common/src/main/java/io/hops/hopsworks/common/pythonresources/PythonResourcesController.java
index e4b204eea6..fb07b08143 100644
--- a/hopsworks-common/src/main/java/io/hops/hopsworks/common/pythonresources/PythonResourcesController.java
+++ b/hopsworks-common/src/main/java/io/hops/hopsworks/common/pythonresources/PythonResourcesController.java
@@ -152,25 +152,34 @@ private Map<String, String> getNodeExporterQueriesKube() {
   }
 
   private Map<String, String> getCadvisorQueries() {
+    final String dockerCgroupParent = settings.getDockerCgroupParent();
+    String cGroupParentForRegex = dockerCgroupParent.replaceAll("\\.", "\\\\\\\\.");
+    final String headNodeInstanceQuery = "instance='" + glassfishIp + ":" + nodeExporterPort + "'";
     return new HashMap<String, String>() {
       {
         put(DOCKER_CURRENT_CPU_USAGE_KEY,
-            "sum(avg by (cpu) (rate(container_cpu_usage_seconds_total{id=~'.*/docker/.*'}[60s]) * 100))");
+            "sum(avg by (cpu) (rate(container_cpu_usage_seconds_total{id=~'.*/" + cGroupParentForRegex
+                + "/.*'}[60s]) * 100))");
         put(DOCKER_TOTAL_ALLOCATABLE_CPU_KEY,
-            "(container_spec_cpu_quota{id='/docker'}/" + settings.getDockerCgroupCpuPeriod() + ")*100");
-        put(DOCKER_CURRENT_MEMORY_USAGE_KEY, "sum(container_memory_working_set_bytes{id=~'.*/docker/.*'})");
-        put(DOCKER_TOTAL_ALLOCATABLE_MEMORY_KEY, "container_spec_memory_limit_bytes{id='/docker'}");
+            "(sum(container_spec_cpu_quota{id='/" + dockerCgroupParent + "'})/("
+                + settings.getDockerCgroupCpuPeriod()
+                + " * (count(count(node_cpu_seconds_total{" + headNodeInstanceQuery + "}) without (mode,instance,job))"
+                + "without (cpu))))*100");
+        put(DOCKER_CURRENT_MEMORY_USAGE_KEY, "sum(container_memory_working_set_bytes{id=~'.*/" + cGroupParentForRegex
+            + "/.*'})");
+        put(DOCKER_TOTAL_ALLOCATABLE_MEMORY_KEY, "container_spec_memory_limit_bytes{id='/" + dockerCgroupParent
+            + "'}");
       }
     };
   }
 
   private Map<String, String> getNodeExporterQueriesHeadNode() {
-    String headNodeQuery = "instance='" + glassfishIp + ":" + nodeExporterPort + "'";
+    final String headNodeQuery = "instance='" + glassfishIp + ":" + nodeExporterPort + "'";
     return new HashMap<String, String>() {
       {
         put(CLUSTER_CURRENT_CPU_USAGE,
-            "100 - ((sum((avg by (instance) (rate(node_cpu_seconds_total{mode='idle', " + headNodeQuery + "}[1m])) " +
-                "* 100)))/(count(node_memory_Active_bytes{" + headNodeQuery + "})))");
+            "100 - ((sum((avg by (instance) (rate(node_cpu_seconds_total{mode='idle', " + headNodeQuery + "}[1m])) "
+                + "* 100)))/(count(node_memory_Active_bytes{" + headNodeQuery + "})))");
         put(CLUSTER_CURRENT_MEMORY_USAGE, "sum(node_memory_Active_bytes{" + headNodeQuery + "})");
         put(CLUSTER_TOTAL_MEMORY_CAPACITY, "sum(node_memory_MemTotal_bytes{" + headNodeQuery + "})");
       }
diff --git a/hopsworks-common/src/main/java/io/hops/hopsworks/common/util/Settings.java b/hopsworks-common/src/main/java/io/hops/hopsworks/common/util/Settings.java
index ce879d5d9f..b61005fad1 100644
--- a/hopsworks-common/src/main/java/io/hops/hopsworks/common/util/Settings.java
+++ b/hopsworks-common/src/main/java/io/hops/hopsworks/common/util/Settings.java
@@ -405,6 +405,7 @@ public class Settings implements Serializable {
   private static final String VARIABLE_DOCKER_CGROUP_CPU_QUOTA = "docker_cgroup_cpu_quota_percentage";
   private static final String VARIABLE_DOCKER_CGROUP_CPU_PERIOD = "docker_cgroup_cpu_period";
   private static final String VARIABLE_DOCKER_CGROUP_MONITOR_INTERVAL = "docker_cgroup_monitor_interval";
+  private static final String VARIABLE_DOCKER_CGROUP_PARENT = "docker_cgroup_parent";
 
   private static final String VARIABLE_PROMETHEUS_PORT = "prometheus_port";
 
@@ -902,6 +903,7 @@ private void populateCache() {
       ENABLE_GIT_READ_ONLY_REPOSITORIES = setBoolVar(VARIABLE_ENABLE_GIT_READ_ONLY_REPOSITORIES,
               ENABLE_GIT_READ_ONLY_REPOSITORIES);
 
+      //Docker cgroups
       DOCKER_CGROUP_ENABLED = setBoolVar(VARIABLE_DOCKER_CGROUP_ENABLED, DOCKER_CGROUP_ENABLED);
       DOCKER_CGROUP_MEMORY_LIMIT = setStrVar(VARIABLE_DOCKER_CGROUP_HARD_LIMIT_MEMORY,
           DOCKER_CGROUP_MEMORY_LIMIT);
@@ -911,6 +913,7 @@ private void populateCache() {
       DOCKER_CGROUP_CPU_PERIOD = setIntVar(VARIABLE_DOCKER_CGROUP_CPU_PERIOD, DOCKER_CGROUP_CPU_PERIOD);
       DOCKER_CGROUP_MONITOR_INTERVAL = setStrVar(VARIABLE_DOCKER_CGROUP_MONITOR_INTERVAL,
           DOCKER_CGROUP_MONITOR_INTERVAL);
+      DOCKER_CGROUP_PARENT = setStrVar(VARIABLE_DOCKER_CGROUP_PARENT, DOCKER_CGROUP_PARENT);
 
       PROMETHEUS_PORT = setIntVar(VARIABLE_PROMETHEUS_PORT, PROMETHEUS_PORT);
   
@@ -1785,6 +1788,12 @@ public synchronized String getDockerCgroupIntervalMonitor() {
     return DOCKER_CGROUP_MONITOR_INTERVAL;
   }
 
+  private String DOCKER_CGROUP_PARENT = "docker.slice";
+  public synchronized String getDockerCgroupParent() {
+    checkCache();
+    return DOCKER_CGROUP_PARENT;
+  }
+
 
   // Service key rotation interval
   private static final String JUPYTER_SHUTDOWN_TIMER_INTERVAL = "jupyter_shutdown_timer_interval";