Skip to content

Commit 0f27b2b

Browse files
Enrico Usaihanwen-pcluste
Enrico Usai
authored and
hanwen-pcluste
committed
Remove every reference to Torque and SGE schedulers
* Remove recipes and scripts * Remove nodewatcher, jobwatcher, sqswatcher references * Remove from CloudWatch Logs configuration * Remove from Kitchen tests Cherry pick of: 40c715a with some conflicts: .kitchen.yml | Modified | Modified CHANGELOG.md | Modified | Modified recipes/compute_sge_config.rb | Modified | Deleted attributes/default.rb | Modified | Modified recipes/default.rb | Modified | Modified recipes/sge_install.rb | Modified | Deleted files/amazon-2/sge_preinstall.sh | Modified | Deleted recipes/torque_install.rb | Modified | Deleted Signed-off-by: Enrico Usai <usai@amazon.com>
1 parent d1035b1 commit 0f27b2b

36 files changed

+22
-1635
lines changed

‎.kitchen.yml‎

Lines changed: 0 additions & 142 deletions
Original file line numberDiff line numberDiff line change
@@ -40,16 +40,6 @@ suites:
4040
nvidia:
4141
enabled: 'yes'
4242

43-
- name: sge_install
44-
run_list:
45-
- recipe[aws-parallelcluster::sge_install]
46-
attributes:
47-
48-
- name: torque_install
49-
run_list:
50-
- recipe[aws-parallelcluster::torque_install]
51-
attributes:
52-
5343
- name: slurm_install
5444
run_list:
5545
- recipe[aws-parallelcluster::slurm_install]
@@ -75,72 +65,6 @@ suites:
7565
# os attribute is used in pipeline code. i.e. centos7, centos7-custom
7666
os: <%= ENV['OS'] %>
7767

78-
- name: sge_config_MasterServer
79-
run_list:
80-
- recipe[aws-parallelcluster::prep_env]
81-
- recipe[aws-parallelcluster::sge_config]
82-
- recipe[aws-parallelcluster::finalize]
83-
- recipe[aws-parallelcluster::tests]
84-
attributes:
85-
kitchen: true
86-
cfncluster:
87-
cfn_node_type: 'MasterServer'
88-
cfn_scheduler: 'sge'
89-
stack_name: <%= ENV['AWS_STACK_NAME'] %>
90-
cfn_volume: <%= ENV['CFN_VOLUME'] %>
91-
cfn_region: <%= ENV['AWS_DEFAULT_REGION'] %>
92-
cfn_encrypted_ephemeral: <%= ENV['CFN_ENCRYPTED_VOLUME'] %>
93-
cfn_ephemeral_dir: <%= ENV['CFN_EPHEMERAL_DIR'] %>
94-
cfn_shared_dir: <%= ENV['CFN_SHARED_DIR'] %>
95-
cfn_cluster_user: <%= ENV['CFN_CLUSTER_USER'] %>
96-
cfn_sqs_queue: <%= ENV['CFN_SQS_QUEUE'] %>
97-
cfn_ddb_table: <%= ENV['CFN_DDB_TABLE'] %>
98-
# cfn_base_os is one of the base OSs supported by pcluster
99-
cfn_base_os: <%= ENV['CFN_BASE_OS'] %>
100-
custom_node_package: <%= ENV['PARALLELCLUSTER_NODE_URL'] %>
101-
# os attribute is used in pipeline code. i.e. centos7, centos7-custom
102-
os: <%= ENV['OS'] %>
103-
ganglia_enabled: 'yes'
104-
dcv_enabled: 'master'
105-
dcv_port: '8443'
106-
enable_intel_hpc_platform: 'true'
107-
enable_efa: 'compute'
108-
nvidia:
109-
enabled: <%= ENV['NVIDIA_ENABLED'] %>
110-
111-
- name: torque_config_MasterServer
112-
run_list:
113-
- recipe[aws-parallelcluster::prep_env]
114-
- recipe[aws-parallelcluster::torque_config]
115-
- recipe[aws-parallelcluster::finalize]
116-
- recipe[aws-parallelcluster::tests]
117-
attributes:
118-
kitchen: true
119-
cfncluster:
120-
cfn_node_type: 'MasterServer'
121-
cfn_scheduler: 'torque'
122-
stack_name: <%= ENV['AWS_STACK_NAME'] %>
123-
cfn_volume: <%= ENV['CFN_VOLUME'] %>
124-
cfn_region: <%= ENV['AWS_DEFAULT_REGION'] %>
125-
cfn_encrypted_ephemeral: <%= ENV['CFN_ENCRYPTED_VOLUME'] %>
126-
cfn_ephemeral_dir: <%= ENV['CFN_EPHEMERAL_DIR'] %>
127-
cfn_shared_dir: <%= ENV['CFN_SHARED_DIR'] %>
128-
cfn_cluster_user: <%= ENV['CFN_CLUSTER_USER'] %>
129-
cfn_sqs_queue: <%= ENV['CFN_SQS_QUEUE'] %>
130-
cfn_ddb_table: <%= ENV['CFN_DDB_TABLE'] %>
131-
# cfn_base_os is one of the base OSs supported by pcluster
132-
cfn_base_os: <%= ENV['CFN_BASE_OS'] %>
133-
custom_node_package: <%= ENV['PARALLELCLUSTER_NODE_URL'] %>
134-
# os attribute is used in pipeline code. i.e. centos7, centos7-custom
135-
os: <%= ENV['OS'] %>
136-
ganglia_enabled: 'yes'
137-
dcv_enabled: 'master'
138-
dcv_port: '8443'
139-
enable_intel_hpc_platform: 'true'
140-
enable_efa: 'compute'
141-
nvidia:
142-
enabled: <%= ENV['NVIDIA_ENABLED'] %>
143-
14468
- name: slurm_config_MasterServer
14569
run_list:
14670
- recipe[aws-parallelcluster::prep_env]
@@ -208,72 +132,6 @@ suites:
208132
nvidia:
209133
enabled: <%= ENV['NVIDIA_ENABLED'] %>
210134

211-
- name: sge_config_ComputeFleet
212-
run_list:
213-
- recipe[aws-parallelcluster::prep_env]
214-
- recipe[aws-parallelcluster::sge_config]
215-
- recipe[aws-parallelcluster::finalize]
216-
- recipe[aws-parallelcluster::tests]
217-
attributes:
218-
kitchen: true
219-
cfncluster:
220-
cfn_node_type: 'ComputeFleet'
221-
cfn_scheduler: 'sge'
222-
stack_name: <%= ENV['AWS_STACK_NAME'] %>
223-
cfn_volume: <%= ENV['CFN_VOLUME'] %>
224-
cfn_region: <%= ENV['AWS_DEFAULT_REGION'] %>
225-
cfn_encrypted_ephemeral: <%= ENV['CFN_ENCRYPTED_VOLUME'] %>
226-
cfn_ephemeral_dir: <%= ENV['CFN_EPHEMERAL_DIR'] %>
227-
cfn_shared_dir: <%= ENV['CFN_SHARED_DIR'] %>
228-
cfn_cluster_user: <%= ENV['CFN_CLUSTER_USER'] %>
229-
cfn_sqs_queue: <%= ENV['CFN_SQS_QUEUE'] %>
230-
cfn_master: <%= ENV['CFN_MASTER'] %>
231-
cfn_master_private_ip: <%= ENV['CFN_MASTER_PRIVATE_IP'] %>
232-
# cfn_base_os is one of the base OSs supported by pcluster
233-
cfn_base_os: <%= ENV['CFN_BASE_OS'] %>
234-
custom_node_package: <%= ENV['PARALLELCLUSTER_NODE_URL'] %>
235-
# os attribute is used in pipeline code. i.e. centos7, centos7-custom
236-
os: <%= ENV['OS'] %>
237-
ganglia_enabled: 'yes'
238-
dcv_enabled: 'master'
239-
dcv_port: '8443'
240-
enable_efa: 'compute'
241-
nvidia:
242-
enabled: <%= ENV['NVIDIA_ENABLED'] %>
243-
244-
- name: torque_config_ComputeFleet
245-
run_list:
246-
- recipe[aws-parallelcluster::prep_env]
247-
- recipe[aws-parallelcluster::torque_config]
248-
- recipe[aws-parallelcluster::finalize]
249-
- recipe[aws-parallelcluster::tests]
250-
attributes:
251-
kitchen: true
252-
cfncluster:
253-
cfn_node_type: 'ComputeFleet'
254-
cfn_scheduler: 'torque'
255-
stack_name: <%= ENV['AWS_STACK_NAME'] %>
256-
cfn_volume: <%= ENV['CFN_VOLUME'] %>
257-
cfn_region: <%= ENV['AWS_DEFAULT_REGION'] %>
258-
cfn_encrypted_ephemeral: <%= ENV['CFN_ENCRYPTED_VOLUME'] %>
259-
cfn_ephemeral_dir: <%= ENV['CFN_EPHEMERAL_DIR'] %>
260-
cfn_shared_dir: <%= ENV['CFN_SHARED_DIR'] %>
261-
cfn_cluster_user: <%= ENV['CFN_CLUSTER_USER'] %>
262-
cfn_sqs_queue: <%= ENV['CFN_SQS_QUEUE'] %>
263-
cfn_master: <%= ENV['CFN_MASTER'] %>
264-
cfn_master_private_ip: <%= ENV['CFN_MASTER_PRIVATE_IP'] %>
265-
# cfn_base_os is one of the base OSs supported by pcluster. i.e. centos7
266-
cfn_base_os: <%= ENV['CFN_BASE_OS'] %>
267-
custom_node_package: <%= ENV['PARALLELCLUSTER_NODE_URL'] %>
268-
# os attribute is used in pipeline code. i.e. centos7, centos7-custom
269-
os: <%= ENV['OS'] %>
270-
ganglia_enabled: 'yes'
271-
dcv_enabled: 'master'
272-
dcv_port: '8443'
273-
enable_efa: 'compute'
274-
nvidia:
275-
enabled: <%= ENV['NVIDIA_ENABLED'] %>
276-
277135
- name: slurm_config_ComputeFleet
278136
run_list:
279137
- recipe[aws-parallelcluster::prep_env]

‎CHANGELOG.md‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,8 @@ This file is used to list changes made in each version of the AWS ParallelCluste
77
-----
88

99
**CHANGES**
10+
- Drop support for SGE and Torque schedulers.
11+
- Remove nodewatcher, sqswatcher, jobwatcher related code.
1012
- Disable log4j-cve-2021-44228-hotpatch service on Amazon Linux to avoid incurring in potential performance degradation.
1113
- Upgrade NVIDIA driver to version 470.103.01.
1214
- Upgrade CUDA library to version 11.4.4.

‎attributes/default.rb‎

Lines changed: 1 addition & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -101,12 +101,6 @@
101101
default['cfncluster']['cfncluster-node-version'] = '2.11.5'
102102

103103
# URLs to software packages used during install recipes
104-
# Gridengine software
105-
default['cfncluster']['sge']['version'] = '8.1.9'
106-
default['cfncluster']['sge']['url'] = 'https://deb.debian.org/debian/pool/main/g/gridengine/gridengine_8.1.9+dfsg.orig.tar.gz'
107-
# Torque software
108-
default['cfncluster']['torque']['version'] = '6.1.2'
109-
default['cfncluster']['torque']['url'] = 'https://github.com/adaptivecomputing/torque/archive/6.1.2.tar.gz'
110104
# Slurm software
111105
default['cfncluster']['slurm_plugin_dir'] = '/etc/parallelcluster/slurm_plugin'
112106
default['cfncluster']['slurm']['version'] = '20-11-8-1'
@@ -318,10 +312,6 @@
318312
default['cfncluster']['ganglia']['httpd_service'] = 'httpd'
319313
default['cfncluster']['chrony']['service'] = "chronyd"
320314
default['cfncluster']['chrony']['conf'] = "/etc/chrony.conf"
321-
default['cfncluster']['torque']['trqauthd_source'] = 'file:///opt/torque/contrib/init.d/trqauthd'
322-
default['cfncluster']['torque']['pbs_mom_source'] = 'file:///opt/torque/contrib/init.d/pbs_mom'
323-
default['cfncluster']['torque']['pbs_sched_source'] = 'file:///opt/torque/contrib/init.d/pbs_sched'
324-
default['cfncluster']['torque']['pbs_server_source'] = 'file:///opt/torque/contrib/init.d/pbs_server'
325315

326316
when 'debian'
327317
default['openssh']['server']['subsystem'] = 'sftp internal-sftp'
@@ -332,7 +322,6 @@
332322
r-base libblas-dev libfftw3-dev libffi-dev libxml2-dev mdadm
333323
libgcrypt20-dev libmysqlclient-dev libevent-dev iproute2 python3 python3-pip
334324
libatlas-base-dev libglvnd-dev linux-headers-aws]
335-
default['cfncluster']['sge']['version'] = '8.1.9+dfsg-9'
336325

337326
case node['platform_version']
338327
when '18.04'
@@ -354,10 +343,6 @@
354343
default['cfncluster']['ganglia']['httpd_service'] = 'apache2'
355344
default['cfncluster']['chrony']['service'] = "chrony"
356345
default['cfncluster']['chrony']['conf'] = "/etc/chrony/chrony.conf"
357-
default['cfncluster']['torque']['trqauthd_source'] = 'file:///opt/torque/contrib/init.d/debian.trqauthd'
358-
default['cfncluster']['torque']['pbs_mom_source'] = 'file:///opt/torque/contrib/init.d/debian.pbs_mom'
359-
default['cfncluster']['torque']['pbs_sched_source'] = 'file:///opt/torque/contrib/init.d/debian.pbs_sched'
360-
default['cfncluster']['torque']['pbs_server_source'] = 'file:///opt/torque/contrib/init.d/debian.pbs_server'
361346

362347
if Chef::VersionConstraint.new('>= 15.04').include?(node['platform_version'])
363348
default['nfs']['service_provider']['idmap'] = Chef::Provider::Service::Systemd
@@ -423,7 +408,7 @@
423408
default['cfncluster']['cfn_proxy'] = 'NONE'
424409
default['cfncluster']['cfn_postinstall'] = 'NONE'
425410
default['cfncluster']['cfn_postinstall_args'] = 'NONE'
426-
default['cfncluster']['cfn_scheduler'] = 'sge'
411+
default['cfncluster']['cfn_scheduler'] = 'slurm'
427412
default['cfncluster']['cfn_scheduler_slots'] = 'vcpus'
428413
default['cfncluster']['cfn_disable_hyperthreading_manually'] = 'false'
429414
default['cfncluster']['cfn_instance_slots'] = '1'

‎files/amazon-2/sge_preinstall.sh‎

Lines changed: 0 additions & 30 deletions
This file was deleted.

0 commit comments

Comments
 (0)