[ganglia] maintenance release (#1289)

* Fixing ganglia.sh Fix the script to fix two scenarios: 1. No source list file needs to be updated 2. Multiple source list files need to be updated * ganglia/BUILD: * using perl to verify http server instead of python ganglia/ganglia.sh: * export DEBIAN_FRONTEND=noninteractive * use latest implementation of repair_old_backports * include some utility functions to help check for running OS, and versions of spark, dataproc and OS versions * refactor ganglia worker setup into function * install worker on master when using single node configuration * reduce noise from apt and wait for lock release ganglia/test_ganglia.py: * increase test instance size to n1-standard-8 * replaced complicated python test with simple perl * run test with conda python * corrected version nit about where ganglia ui is supported ganglia/verify_ganglia_running.py, ganglia/verify_ganglia_running.pl: * replaced complicated python with simple perl --------- Co-authored-by: C.J. Collier <[email protected]>
GoogleCloudDataproc · Jan 9, 2025 · 45a0a24 · 45a0a24
1 parent 4a477e3
commit 45a0a24
Show file tree

Hide file tree

Showing 5 changed files with 134 additions and 130 deletions.
diff --git a/ganglia/BUILD b/ganglia/BUILD
@@ -6,17 +6,12 @@ py_test(
     srcs = ["test_ganglia.py"],
     data = [
         "ganglia.sh",
+	"verify_ganglia_running.pl",
     ],
     local = True,
     shard_count = 4,
     deps = [
-        ":verify_ganglia_running",
         "//integration_tests:dataproc_test_case",
         "@io_abseil_py//absl/testing:parameterized",
     ],
 )
-
-py_library(
-    name = "verify_ganglia_running",
-    srcs = ["verify_ganglia_running.py"],
-)
diff --git a/ganglia/ganglia.sh b/ganglia/ganglia.sh
@@ -17,14 +17,51 @@
 
 set -euxo pipefail
 
+function os_id()       ( set +x ;  grep '^ID=' /etc/os-release | cut -d= -f2 | xargs ; )
+function os_version()  ( set +x ;  grep '^VERSION_ID=' /etc/os-release | cut -d= -f2 | xargs ; )
+function os_codename() ( set +x ;  grep '^VERSION_CODENAME=' /etc/os-release | cut -d= -f2 | xargs ; )
+
+# For version (or real number) comparison
+# if first argument is greater than or equal to, greater than, less than or equal to, or less than the second
+# ( version_ge 2.0 2.1 ) evaluates to false
+# ( version_ge 2.2 2.1 ) evaluates to true
+function version_ge() ( set +x ;  [ "$1" = "$(echo -e "$1\n$2" | sort -V | tail -n1)" ] ; )
+function version_gt() ( set +x ;  [ "$1" = "$2" ] && return 1 || version_ge $1 $2 ; )
+function version_le() ( set +x ;  [ "$1" = "$(echo -e "$1\n$2" | sort -V | head -n1)" ] ; )
+function version_lt() ( set +x ;  [ "$1" = "$2" ] && return 1 || version_le $1 $2 ; )
+
+function define_os_comparison_functions() {
+
+  readonly -A supported_os=(
+    ['debian']="10 11 12"
+    ['rocky']="8 9"
+    ['ubuntu']="18.04 20.04 22.04"
+  )
+
+  # dynamically define OS version test utility functions
+  if [[ "$(os_id)" == "rocky" ]];
+  then _os_version=$(os_version | sed -e 's/[^0-9].*$//g')
+  else _os_version="$(os_version)"; fi
+  for os_id_val in 'rocky' 'ubuntu' 'debian' ; do
+    eval "function is_${os_id_val}() ( set +x ;  [[ \"$(os_id)\" == '${os_id_val}' ]] ; )"
+
+    for osver in $(echo "${supported_os["${os_id_val}"]}") ; do
+      eval "function is_${os_id_val}${osver%%.*}() ( set +x ; is_${os_id_val} && [[ \"${_os_version}\" == \"${osver}\" ]] ; )"
+      eval "function ge_${os_id_val}${osver%%.*}() ( set +x ; is_${os_id_val} && version_ge \"${_os_version}\" \"${osver}\" ; )"
+      eval "function le_${os_id_val}${osver%%.*}() ( set +x ; is_${os_id_val} && version_le \"${_os_version}\" \"${osver}\" ; )"
+    done
+  done
+  eval "function is_debuntu()  ( set +x ;  is_debian || is_ubuntu ; )"
+}
+
 function err() {
   echo "[$(date +'%Y-%m-%dT%H:%M:%S%z')]: $*" >&2
   return 1
 }
 
 function update_apt_get() {
   for ((i = 0; i < 10; i++)); do
-    if apt-get update; then
+    if apt-get update > /dev/null ; then
       return 0
     fi
     sleep 5
@@ -34,61 +71,93 @@ function update_apt_get() {
 
 function setup_ganglia_host() {
   # Install dependencies needed for Ganglia host
-  DEBIAN_FRONTEND=noninteractive apt-get install -y \
+  apt-get install -qq -y -o DPkg::Lock::Timeout=60 \
     rrdtool \
     gmetad \
-    ganglia-webfrontend || err 'Unable to install packages'
+    ganglia-webfrontend >/dev/null || err 'Unable to install packages'
 
-  ln -s /etc/ganglia-webfrontend/apache.conf /etc/apache2/sites-enabled/ganglia.conf
-  sed -i "s/my cluster/${master_hostname}/" /etc/ganglia/gmetad.conf
+  ln -sf /etc/ganglia-webfrontend/apache.conf /etc/apache2/sites-enabled/ganglia.conf
+  perl -pi -e "s:^data_source.*:data_source \"${master_hostname}\" localhost:g" /etc/ganglia/gmetad.conf
   sed -i '26s/ \$context_metrics \= \"\"\;/ \$context_metrics \= array\(\)\;/g' /usr/share/ganglia-webfrontend/cluster_view.php
   systemctl restart ganglia-monitor gmetad apache2
 }
 
-function remove_old_backports {
+function setup_ganglia_worker() {
+  # on single node instances, also configure ganglia-monitor
+  sed -e "/deaf = no /s/no/yes/" -i /etc/ganglia/gmond.conf
+  sed -i '/udp_recv_channel {/,/}/d' /etc/ganglia/gmond.conf
+  systemctl restart ganglia-monitor
+}
+
+function repair_old_backports {
+  if ! is_debuntu ; then return ; fi
   # This script uses 'apt-get update' and is therefore potentially dependent on
   # backports repositories which have been archived.  In order to mitigate this
-  # problem, we will remove any reference to backports repos older than oldstable
+  # problem, we will use archive.debian.org for the oldoldstable repo
 
   # https://github.com/GoogleCloudDataproc/initialization-actions/issues/1157
-  oldstable=$(curl -s https://deb.debian.org/debian/dists/oldstable/Release | awk '/^Codename/ {print $2}');
-  stable=$(curl -s https://deb.debian.org/debian/dists/stable/Release | awk '/^Codename/ {print $2}');
-
-  matched_files="$(grep -rsil '\-backports' /etc/apt/sources.list*)"
-  if [[ -n "$matched_files" ]]; then
-    for filename in "$matched_files"; do
-      grep -e "$oldstable-backports" -e "$stable-backports" "$filename" || \
-        sed -i -e 's/^.*-backports.*$//' "$filename"
-    done
-  fi
+  debdists="https://deb.debian.org/debian/dists"
+  oldoldstable=$(curl -s "${debdists}/oldoldstable/Release" | awk '/^Codename/ {print $2}');
+  oldstable=$(   curl -s "${debdists}/oldstable/Release"    | awk '/^Codename/ {print $2}');
+  stable=$(      curl -s "${debdists}/stable/Release"       | awk '/^Codename/ {print $2}');
+
+  matched_files=( $(test -d /etc/apt && grep -rsil '\-backports' /etc/apt/sources.list*||:) )
+
+  for filename in "${matched_files[@]}"; do
+    # Fetch from archive.debian.org for ${oldoldstable}-backports
+    perl -pi -e "s{^(deb[^\s]*) https?://[^/]+/debian ${oldoldstable}-backports }
+                  {\$1 https://archive.debian.org/debian ${oldoldstable}-backports }g" "${filename}"
+  done
 }
 
 function main() {
   local master_hostname=$(/usr/share/google/get_metadata_value attributes/dataproc-master)
   local cluster_name=$(/usr/share/google/get_metadata_value attributes/dataproc-cluster-name)
 
+  export DEBIAN_FRONTEND=noninteractive
+
   OS=$(. /etc/os-release && echo "${ID}")
+
+  define_os_comparison_functions
+
+  # Detect dataproc image version
+  SPARK_VERSION="$(spark-submit --version 2>&1 | sed -n 's/.*version[[:blank:]]\+\([0-9]\+\.[0-9]\).*/\1/p' | head -n1)"
+  readonly SPARK_VERSION
+
+  if (! test -v DATAPROC_IMAGE_VERSION) ; then
+    if test -v DATAPROC_VERSION ; then
+      DATAPROC_IMAGE_VERSION="${DATAPROC_VERSION}"
+    else
+      if   version_lt "${SPARK_VERSION}" "3.2" ; then DATAPROC_IMAGE_VERSION="2.0"
+      elif version_lt "${SPARK_VERSION}" "3.4" ; then DATAPROC_IMAGE_VERSION="2.1"
+      elif version_lt "${SPARK_VERSION}" "3.6" ; then DATAPROC_IMAGE_VERSION="2.2"
+      else echo "Unknown dataproc image version" ; exit 1 ; fi
+    fi
+  fi
+
   if [[ ${OS} == debian ]] && [[ $(echo "${DATAPROC_IMAGE_VERSION} <= 2.1" | bc -l) == 1 ]]; then
-    remove_old_backports
+    repair_old_backports
   fi
 
-  update_apt_get || err 'Unable to update apt-get'
-  apt-get install -y ganglia-monitor
+  update_apt_get > /dev/null || err 'Unable to update apt-get'
+  apt-get install -qq -y -o DPkg::Lock::Timeout=60 ganglia-monitor > /dev/null
 
   sed -e "/send_metadata_interval = 0 /s/0/5/" -i /etc/ganglia/gmond.conf
   sed -e "/name = \"unspecified\" /s/unspecified/${cluster_name}/" -i /etc/ganglia/gmond.conf
   sed -e '/mcast_join /s/^  /  #/' -i /etc/ganglia/gmond.conf
   sed -e '/bind /s/^  /  #/' -i /etc/ganglia/gmond.conf
   sed -e "/udp_send_channel {/a\  host = ${master_hostname}" -i /etc/ganglia/gmond.conf
 
+  local worker_count=$(/usr/share/google/get_metadata_value dataproc-worker-count)
   if [[ "$(hostname -s)" == "${master_hostname}" ]]; then
     # Setup Ganglia host only on the master node ("0"-master in HA mode)
     setup_ganglia_host || err 'Setting up Ganglia host failed'
+    if [[ "${worker_count}" == "0" ]] ; then
+      # on single node instances, also configure ganglia-monitor
+      setup_ganglia_worker
+    fi
   else
-    # Configure non-host Ganglia nodes
-    sed -e "/deaf = no /s/no/yes/" -i /etc/ganglia/gmond.conf
-    sed -i '/udp_recv_channel {/,/}/d' /etc/ganglia/gmond.conf
-    systemctl restart ganglia-monitor
+    setup_ganglia_worker
   fi
 }
 

diff --git a/ganglia/test_ganglia.py b/ganglia/test_ganglia.py
@@ -6,25 +6,17 @@
 
 from integration_tests.dataproc_test_case import DataprocTestCase
 
-
 class GangliaTestCase(DataprocTestCase):
     COMPONENT = 'ganglia'
     INIT_ACTIONS = ['ganglia/ganglia.sh']
-    TEST_SCRIPT_FILE_NAME = 'verify_ganglia_running.py'
+    TEST_SCRIPT_FILE_NAME = 'verify_ganglia_running.pl'
 
     def verify_instance(self, name):
         test_script_path = os.path.join(
             os.path.dirname(os.path.abspath(__file__)),
             self.TEST_SCRIPT_FILE_NAME)
         self.upload_test_file(test_script_path, name)
-        self.assert_instance_command(name,
-                                     "yes | sudo apt-get install python3-pip libxml2-dev libxslt-dev")
-        self.assert_instance_command(name, "sudo -H pip3 install --upgrade pip")
-        self.assert_instance_command(name, "sudo pip3 install requests-html")
-        self.assert_instance_command(name, "sudo pip install -U urllib3 requests")
-        self.assert_instance_command(name, "pip install lxml[html_clean]")
-        self.assert_instance_command(
-            name, "python3 {}".format(self.TEST_SCRIPT_FILE_NAME))
+        self.assert_instance_command(name,"/usr/bin/perl {}".format(self.TEST_SCRIPT_FILE_NAME))
         self.remove_test_script(self.TEST_SCRIPT_FILE_NAME, name)
 
     @parameterized.parameters(
@@ -38,9 +30,13 @@ def test_ganglia(self, configuration, machine_suffixes):
             self.skipTest("Not supported in Rocky Linux-based images")
 
         if self.getImageVersion() > pkg_resources.parse_version("2.0"):
-            self.skipTest("Ganglia UI is not supported for 2.0+ versions")
+            self.skipTest("Ganglia UI is not supported for 2.1+ versions")
 
-        self.createCluster(configuration, self.INIT_ACTIONS)
+        self.createCluster(
+            configuration,
+            self.INIT_ACTIONS,
+            machine_type="n1-standard-8",
+        )
         for machine_suffix in machine_suffixes:
             self.verify_instance("{}-{}".format(self.getClusterName(),
                                                 machine_suffix))

diff --git a/ganglia/verify_ganglia_running.pl b/ganglia/verify_ganglia_running.pl
@@ -0,0 +1,32 @@
+#!/usr/bin/perl -w
+# -*-CPerl-*-
+
+# verify_ganglia_running.py: Script for ganglia initialization action test.
+
+use strict;
+use LWP::UserAgent;
+
+my $hostname       = qx(hostname -s); chomp $hostname;
+my $role           = qx(/usr/share/google/get_metadata_value attributes/dataproc-role);
+my $primary_master = qx(/usr/share/google/get_metadata_value attributes/dataproc-master);
+my $cluster_name   = qx(/usr/share/google/get_metadata_value attributes/dataproc-cluster-name);
+
+if ( $hostname eq $primary_master ){
+  my $hostname = 'localhost';
+  my $port = '80';
+
+  my $ua = LWP::UserAgent->new;
+
+  my $response = $ua->get("http://${hostname}:${port}/ganglia/");
+
+  die $response->status_line unless $response->is_success;
+  my( $page_title ) = ( $response->decoded_content =~ m:<b id="page_title">([^>]+)</b>: );
+  die 'Ganglia UI is not found on master node' unless( $page_title =~ /^${cluster_name}/ );
+  print("Ganglia UI is running on this node.",$/);
+}else{
+  if ( $hostname =~ /-w-/ ){
+    print("Ganglia UI should not run on worker node",$/);
+  }elsif( $hostname =~ /-m-/ ){
+    print("Ganglia UI should not run on additional master",$/);
+  }
+}
diff --git a/ganglia/verify_ganglia_running.py b/ganglia/verify_ganglia_running.py