Merge branch 'develop' into wip/mgiacomo/3120/fix-arm-ubuntu-1211-1

himani2411 · web-flow · commit 72855860a147 · 2024-12-26T11:58:44.000-05:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,11 @@ aws-parallelcluster-cookbook CHANGELOG
 ======================================
 
 This file is used to list changes made in each version of the AWS ParallelCluster cookbook.
+3.13.0
+------
+
+**CHANGES**
+- On Ubuntu 22.04, install the Nvidia driver with the same compiler version used to compile the kernel.
 
 3.12.0
 ------
@@ -25,7 +30,7 @@ This file is used to list changes made in each version of the AWS ParallelCluste
   - Open MPI: `openmpi40-aws-4.1.7-1` and `openmpi50-aws-5.0.5`
 - Auto-restart slurmctld on failure.
 - Upgrade mysql-community-client to version 8.0.39.
-- On Ubuntu 22.04, install the Nvidia driver with the same compiler version used to compile the kernel.
+
 
 **BUG FIXES**
 - Fix retrieval of regions when managing volumes to correctly handle local zones.
diff --git a/cookbooks/aws-parallelcluster-awsbatch/metadata.rb b/cookbooks/aws-parallelcluster-awsbatch/metadata.rb
@@ -7,7 +7,7 @@
 issues_url 'https://github.com/aws/aws-parallelcluster/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'apt', '~> 7.5.22'
 depends 'iptables', '~> 8.0.0'
@@ -17,4 +17,4 @@
 depends 'pyenv', '~> 4.2.3'
 depends 'yum', '~> 7.4.13'
 depends 'yum-epel', '~> 5.0.2'
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-computefleet/metadata.rb b/cookbooks/aws-parallelcluster-computefleet/metadata.rb
@@ -7,6 +7,6 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-entrypoints/metadata.rb b/cookbooks/aws-parallelcluster-entrypoints/metadata.rb
@@ -7,11 +7,11 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
-depends 'aws-parallelcluster-platform', '~> 3.12.0'
-depends 'aws-parallelcluster-environment', '~> 3.12.0'
-depends 'aws-parallelcluster-computefleet', '~> 3.12.0'
-depends 'aws-parallelcluster-slurm', '~> 3.12.0'
-depends 'aws-parallelcluster-awsbatch', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
+depends 'aws-parallelcluster-platform', '~> 3.13.0'
+depends 'aws-parallelcluster-environment', '~> 3.13.0'
+depends 'aws-parallelcluster-computefleet', '~> 3.13.0'
+depends 'aws-parallelcluster-slurm', '~> 3.13.0'
+depends 'aws-parallelcluster-awsbatch', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-environment/files/amazon-2023/network_interfaces/configure_nw_interface.sh b/cookbooks/aws-parallelcluster-environment/files/amazon-2023/network_interfaces/configure_nw_interface.sh
@@ -4,15 +4,16 @@ set -ex
 
 if
   [ -z "${DEVICE_NAME}" ] ||          # name of the device
-  [ -z "${DEVICE_NUMBER}" ] ||       # number of the device
+  [ -z "${DEVICE_NUMBER}" ] ||       # index of the device
+  [ -z "${NETWORK_CARD_INDEX}" ] ||   # index of the network card
   [ -z "${DEVICE_IP_ADDRESS}" ] ||       # ip of the device
   [ -z "${MAC}" ] ||                 # mac address of the device
   [ -z "${CIDR_BLOCK}" ]                 # CIDR block of the subnet
 then
   echo 'One or more environment variables missing'
   exit 1
 fi
-echo "Configuring NIC, Device name: ${DEVICE_NAME}, Device number: ${DEVICE_NUMBER}"
+echo "Configuring NIC, Device name: ${DEVICE_NAME}, Device number: ${DEVICE_NUMBER}, Network card index:${NETWORK_CARD_INDEX}"
 
 configuration_directory="/etc/systemd/network"
 file_name="70-${DEVICE_NAME}.network"
@@ -23,12 +24,13 @@ fi
 
 cd "$configuration_directory"
 
-ROUTE_TABLE=100${DEVICE_NUMBER}
+SUFFIX=$(printf "%03d" $NETWORK_CARD_INDEX)$(printf "%02d" $DEVICE_NUMBER)
+ROUTE_TABLE="$(( $SUFFIX + 10 ))"
 
 ln -s /usr/lib/systemd/network/80-ec2.network ${file_name} # Use default EC2 configuration. This include MTU, etc.
 
 /bin/cat <<EOF > ${sub_directory}/eni.conf
-# Configuration for ${DEVICE_NUMBER} generated by ParallelCluster
+# Configuration for network card: ${NETWORK_CARD_INDEX}, device number: ${DEVICE_NUMBER} generated by ParallelCluster
 # This is inspired by https://github.com/amazonlinux/amazon-ec2-net-utils/blob/v2.4.1/lib/lib.sh
 [Match]
 MACAddress=${MAC}
diff --git a/cookbooks/aws-parallelcluster-environment/files/default/network_interfaces/configure_nw_interface.sh b/cookbooks/aws-parallelcluster-environment/files/default/network_interfaces/configure_nw_interface.sh
@@ -10,7 +10,8 @@ set -e
 
 if
   [ -z "${DEVICE_NAME}" ] ||          # name of the device
-  [ -z "${DEVICE_NUMBER}" ] ||        # number of the device
+  [ -z "${DEVICE_NUMBER}" ] ||        # index of the device
+  [ -z "${NETWORK_CARD_INDEX}" ] ||   # index of the network card
   [ -z "${GW_IP_ADDRESS}" ] ||        # gateway ip address
   [ -z "${DEVICE_IP_ADDRESS}" ] ||    # ip address to assign to the interface
   [ -z "${CIDR_PREFIX_LENGTH}" ] ||   # the prefix length of the device IP cidr block
@@ -20,9 +21,11 @@ then
   exit 1
 fi
 
-ROUTE_TABLE="100${DEVICE_NUMBER}"
+SUFFIX=$(printf "%03d" $NETWORK_CARD_INDEX)$(printf "%02d" $DEVICE_NUMBER)
 
-echo "Configuring ${DEVICE_NAME} with IP:${DEVICE_IP_ADDRESS} CIDR_PREFIX:${CIDR_PREFIX_LENGTH} NETMASK:${NETMASK} GW:${GW_IP_ADDRESS} ROUTING_TABLE:${ROUTE_TABLE}"
+ROUTE_TABLE="$(( $SUFFIX + 10 ))"
+
+echo "Configuring device name: ${DEVICE_NAME} with IP:${DEVICE_IP_ADDRESS} CIDR_PREFIX:${CIDR_PREFIX_LENGTH} NETMASK:${NETMASK} GW:${GW_IP_ADDRESS} ROUTING_TABLE:${ROUTE_TABLE}"
 
 # config file
 FILE="/etc/sysconfig/network-scripts/ifcfg-${DEVICE_NAME}"
diff --git a/cookbooks/aws-parallelcluster-environment/files/redhat-8.network_interfaces/configure_nw_interface.sh b/cookbooks/aws-parallelcluster-environment/files/redhat-8.network_interfaces/configure_nw_interface.sh
@@ -13,7 +13,8 @@ set -e
 
 if
   [ -z "${DEVICE_NAME}" ] ||          # name of the device
-  [ -z "${DEVICE_NUMBER}" ] ||        # number of the device
+  [ -z "${DEVICE_NUMBER}" ] ||        # index of the device
+  [ -z "${NETWORK_CARD_INDEX}" ] ||   # index of the network card
   [ -z "${GW_IP_ADDRESS}" ] ||        # gateway ip address
   [ -z "${DEVICE_IP_ADDRESS}" ] ||    # ip address to assign to the interface
   [ -z "${CIDR_PREFIX_LENGTH}" ]      # the prefix length of the device IP cidr block
@@ -23,9 +24,10 @@ then
 fi
 
 con_name="System ${DEVICE_NAME}"
-route_table="100${DEVICE_NUMBER}"
-priority="100${DEVICE_NUMBER}"
-metric="100${DEVICE_NUMBER}"
+SUFFIX=$(printf "%03d" $NETWORK_CARD_INDEX)$(printf "%02d" $DEVICE_NUMBER)
+route_table="$(( $SUFFIX + 10 ))"
+priority="${route_table}"
+metric="${route_table}"
 
 # Rename connection
 original_con_name=`nmcli -t -f GENERAL.CONNECTION device show ${DEVICE_NAME} | cut -f2 -d':'`
diff --git a/cookbooks/aws-parallelcluster-environment/files/rocky/network_interfaces/configure_nw_interface.sh b/cookbooks/aws-parallelcluster-environment/files/rocky/network_interfaces/configure_nw_interface.sh
@@ -13,7 +13,8 @@ set -e
 
 if
   [ -z "${DEVICE_NAME}" ] ||          # name of the device
-  [ -z "${DEVICE_NUMBER}" ] ||        # number of the device
+  [ -z "${DEVICE_NUMBER}" ] ||        # index of the device
+  [ -z "${NETWORK_CARD_INDEX}" ] ||   # index of the network card
   [ -z "${GW_IP_ADDRESS}" ] ||        # gateway ip address
   [ -z "${DEVICE_IP_ADDRESS}" ] ||    # ip address to assign to the interface
   [ -z "${CIDR_PREFIX_LENGTH}" ]      # the prefix length of the device IP cidr block
@@ -23,9 +24,10 @@ then
 fi
 
 con_name="System ${DEVICE_NAME}"
-route_table="100${DEVICE_NUMBER}"
-priority="100${DEVICE_NUMBER}"
-metric="100${DEVICE_NUMBER}"
+SUFFIX=$(printf "%03d" $NETWORK_CARD_INDEX)$(printf "%02d" $DEVICE_NUMBER)
+route_table="$(( $SUFFIX + 10 ))"
+priority="${route_table}"
+metric="${route_table}"
 
 # Rename connection
 original_con_name=`nmcli -t -f GENERAL.CONNECTION device show ${DEVICE_NAME} | cut -f2 -d':'`
diff --git a/cookbooks/aws-parallelcluster-environment/files/ubuntu/network_interfaces/configure_nw_interface.sh b/cookbooks/aws-parallelcluster-environment/files/ubuntu/network_interfaces/configure_nw_interface.sh
@@ -10,7 +10,8 @@ set -e
 
 if
   [ -z "${DEVICE_NAME}" ] ||          # name of the device
-  [ -z "${DEVICE_NUMBER}" ] ||        # number of the device
+  [ -z "${DEVICE_NUMBER}" ] ||        # index of the device
+  [ -z "${NETWORK_CARD_INDEX}" ] ||   # index of the network card
   [ -z "${GW_IP_ADDRESS}" ] ||        # gateway ip address
   [ -z "${DEVICE_IP_ADDRESS}" ] ||    # ip address to assign to the interface
   [ -z "${CIDR_PREFIX_LENGTH}" ] ||   # the prefix length of the device IP cidr block
@@ -40,7 +41,8 @@ if [ "${STATIC_IP_CONFIG}" = "" ]
 fi
 
 FILE="/etc/netplan/${DEVICE_NAME}.yaml"
-ROUTE_TABLE="100${DEVICE_NUMBER}"
+SUFFIX=$(printf "%03d" $NETWORK_CARD_INDEX)$(printf "%02d" $DEVICE_NUMBER)
+ROUTE_TABLE="$(( $SUFFIX + 10 ))"
 
 echo "Configuring ${DEVICE_NAME} with IP:${DEVICE_IP_ADDRESS} CIDR_PREFIX:${CIDR_PREFIX_LENGTH} NETMASK:${NETMASK} GW:${GW_IP_ADDRESS} ROUTING_TABLE:${ROUTE_TABLE}"
 
diff --git a/cookbooks/aws-parallelcluster-environment/metadata.rb b/cookbooks/aws-parallelcluster-environment/metadata.rb
@@ -7,9 +7,9 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'line', '~> 4.5.13'
 depends 'nfs', '~> 5.1.2'
 
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-environment/recipes/config/network_interfaces.rb b/cookbooks/aws-parallelcluster-environment/recipes/config/network_interfaces.rb
@@ -16,7 +16,13 @@
 return if on_docker?
 
 def network_card_index(mac, token)
+  # This IMDS call is not available on single NIC instance, therefore fallback to 0
   uri = URI("http://169.254.169.254/latest/meta-data/network/interfaces/macs/#{mac}/network-card")
+  get_metadata_with_token(token, uri) || 0
+end
+
+def device_number(mac, token)
+  uri = URI("http://169.254.169.254/latest/meta-data/network/interfaces/macs/#{mac}/device-number")
   get_metadata_with_token(token, uri)
 end
 
@@ -69,6 +75,7 @@ def cidr_to_netmask(cidr)
   # Configure nw interfaces
   macs.each do |mac|
     device_name = device_name(mac)
+    device_number = device_number(mac, token)
     network_card_index = network_card_index(mac, token)
     gw_ip_address = gateway_address
     device_ip_address = device_ip(mac, token)
@@ -84,7 +91,8 @@ def cidr_to_netmask(cidr)
       environment(
         # TODO: The variables are a superset of what's required by individual scripts. Consider simplification.
         'DEVICE_NAME' => device_name,
-        'DEVICE_NUMBER' => "#{network_card_index}", # in configure_nw_interface DEVICE_NUMBER actually means network card index
+        'DEVICE_NUMBER' => "#{device_number}",
+        'NETWORK_CARD_INDEX' => "#{network_card_index}",
         'GW_IP_ADDRESS' => gw_ip_address,
         'DEVICE_IP_ADDRESS' => device_ip_address,
         'CIDR_PREFIX_LENGTH' => cidr_prefix_length,
diff --git a/cookbooks/aws-parallelcluster-platform/files/ami_cleanup.sh b/cookbooks/aws-parallelcluster-platform/files/ami_cleanup.sh
@@ -9,6 +9,11 @@ rm -f /etc/udev/rules.d/70-persistent-net.rules
 grep -l "Created by cloud-init on instance boot automatically" /etc/sysconfig/network-scripts/ifcfg-* | xargs rm -f
 rm -rf /var/crash/*
 
+if [ -f /opt/parallelcluster/pin_releasesever ]; then
+  rm -f /opt/parallelcluster/pin_releasesever
+  rm -f /etc/yum/vars/releasever
+fi
+
 # https://bugs.centos.org/view.php?id=13836#c33128
 source /etc/os-release
 if [ "${ID}${VERSION_ID}" == "centos7" ]; then
diff --git a/cookbooks/aws-parallelcluster-platform/metadata.rb b/cookbooks/aws-parallelcluster-platform/metadata.rb
@@ -7,8 +7,8 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'line', '~> 4.5.13'
 
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-shared/attributes/versions.rb b/cookbooks/aws-parallelcluster-shared/attributes/versions.rb
@@ -3,7 +3,7 @@
 default['cluster']['python-major-minor-version'] = '3.9'
 
 # ParallelCluster versions
-default['cluster']['parallelcluster-version'] = '3.12.0'
-default['cluster']['parallelcluster-cookbook-version'] = '3.12.0'
-default['cluster']['parallelcluster-node-version'] = '3.12.0'
+default['cluster']['parallelcluster-version'] = '3.13.0'
+default['cluster']['parallelcluster-cookbook-version'] = '3.13.0'
+default['cluster']['parallelcluster-node-version'] = '3.13.0'
 default['cluster']['parallelcluster-awsbatch-cli-version'] = '1.4.0'
diff --git a/cookbooks/aws-parallelcluster-shared/metadata.rb b/cookbooks/aws-parallelcluster-shared/metadata.rb
@@ -7,7 +7,7 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'pyenv', '~> 4.2.3'
 depends 'yum', '~> 7.4.13'
diff --git a/cookbooks/aws-parallelcluster-slurm/metadata.rb b/cookbooks/aws-parallelcluster-slurm/metadata.rb
@@ -7,7 +7,7 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'apt', '~> 7.5.22'
 depends 'iptables', '~> 8.0.0'
@@ -17,6 +17,6 @@
 depends 'pyenv', '~> 4.2.3'
 depends 'yum', '~> 7.4.13'
 depends 'yum-epel', '~> 5.0.2'
-depends 'aws-parallelcluster-computefleet', '~> 3.12.0'
-depends 'aws-parallelcluster-environment', '~> 3.12.0'
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
+depends 'aws-parallelcluster-computefleet', '~> 3.13.0'
+depends 'aws-parallelcluster-environment', '~> 3.13.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
diff --git a/cookbooks/aws-parallelcluster-tests/metadata.rb b/cookbooks/aws-parallelcluster-tests/metadata.rb
@@ -7,10 +7,10 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
-depends 'aws-parallelcluster-platform', '~> 3.12.0'
-depends 'aws-parallelcluster-environment', '~> 3.12.0'
-depends 'aws-parallelcluster-computefleet', '~> 3.12.0'
-depends 'aws-parallelcluster-slurm', '~> 3.12.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
+depends 'aws-parallelcluster-platform', '~> 3.13.0'
+depends 'aws-parallelcluster-environment', '~> 3.13.0'
+depends 'aws-parallelcluster-computefleet', '~> 3.13.0'
+depends 'aws-parallelcluster-slurm', '~> 3.13.0'
diff --git a/kitchen.ec2.yml b/kitchen.ec2.yml
@@ -1,5 +1,5 @@
 <%
-  pcluster_version = ENV['KITCHEN_PCLUSTER_VERSION'] || '3.12.0'
+  pcluster_version = ENV['KITCHEN_PCLUSTER_VERSION'] || '3.13.0'
   pcluster_prefix = "aws-parallelcluster-#{pcluster_version}"
 %>
 ---
diff --git a/metadata.rb b/metadata.rb
@@ -7,7 +7,7 @@
 issues_url 'https://github.com/aws/aws-parallelcluster-cookbook/issues'
 source_url 'https://github.com/aws/aws-parallelcluster-cookbook'
 chef_version '>= 18'
-version '3.12.0'
+version '3.13.0'
 
 depends 'apt', '~> 7.5.22'
 depends 'iptables', '~> 8.0.0'
@@ -17,10 +17,10 @@
 depends 'pyenv', '~> 4.2.3'
 depends 'yum', '~> 7.4.13'
 depends 'yum-epel', '~> 5.0.2'
-depends 'aws-parallelcluster-slurm', '~> 3.12.0'
-depends 'aws-parallelcluster-awsbatch', '~> 3.12.0'
-depends 'aws-parallelcluster-platform', '~> 3.12.0'
-depends 'aws-parallelcluster-environment', '~> 3.12.0'
-depends 'aws-parallelcluster-computefleet', '~> 3.12.0'
-depends 'aws-parallelcluster-shared', '~> 3.12.0'
-depends 'aws-parallelcluster-entrypoints', '~> 3.12.0'
+depends 'aws-parallelcluster-slurm', '~> 3.13.0'
+depends 'aws-parallelcluster-awsbatch', '~> 3.13.0'
+depends 'aws-parallelcluster-platform', '~> 3.13.0'
+depends 'aws-parallelcluster-environment', '~> 3.13.0'
+depends 'aws-parallelcluster-computefleet', '~> 3.13.0'
+depends 'aws-parallelcluster-shared', '~> 3.13.0'
+depends 'aws-parallelcluster-entrypoints', '~> 3.13.0'
