Skip to content

GitLab

Commit 88740949 authored by John Tourtellott's avatar John Tourtellott
Browse files

Add option to generate AWS parallel cluster files 

In progress
parent 597d0021
Hide whitespace changes
Inline Side-by-side
.gitignore 0 → 100644
View file @ 88740949
__pycache__
simulation-workflows/ACE3P.py
View file @ 88740949
......@@ -97,6 +97,7 @@ def ExportCMB(export_op):
scope = ExportScope()
scope.logger = export_op.log()
scope.warning_messages = list()
scope.export_att = export_op.parameters()
scope.sim_atts = smtk.attribute.Resource.CastTo(scope.export_att.find('attributes').value())
if scope.sim_atts is None:
......@@ -293,6 +294,13 @@ def ExportCMB(export_op):
# And any sym link
print('Symlink', scope.symlink)
# Check for AWSSimulation item
aws_item = scope.export_att.find('AWSSimulation')
if aws_item is not None and aws_item.isEnabled():
from internal.writers import aws
reload(aws)
aws_writer = aws.AWSWriter(scope)
completed &= aws_writer.export_ace3p(aws_item)
return completed
......
simulation-workflows/internal/ace3p-export.sbt
View file @ 88740949
......@@ -146,6 +146,10 @@
</Structure>
</DiscreteInfo>
</String>
<Directory Name="ProjectFolder" Label="Project Folder" AdvanceLevel="1"
Optional="true" IsEnabledByDefault="false" Version="0">
<BriefDescription>The project root folder on the local filesystem</BriefDescription>
</Directory>
<Group Name="NERSCSimulation" Label="Submit job to NERSC"
Optional="true" IsEnabledByDefault="false"
......@@ -257,8 +261,151 @@
</String>
--> </ItemDefinitions>
</Group>
<Group Name="AWSSimulation" Label="Specify AWS Job (Experimental)" AdvanceLevel="1"
Optional="true" IsEnabledByDefault="false"
Version="1" NumberOfRequiredGroups="1">
<ItemDefinitions>
<String Name="JobName" Label="Job name" Version="0">
<BriefDescription>Label you can use to track your job</BriefDescription>
<DefaultValue>ACE3P</DefaultValue>
</String>
<String Name="JobNotes" Label="Notes" Version="0" MultipleLines="true">
<BriefDescription>Optional notes you want to save with this job</BriefDescription>
<DefaultValue> </DefaultValue>
</String>
<File Name="AWSConfigFile" Label="AWS Config File" ShouldExist="true" Version="0">
<BriefDescription>AWS ParallelCluster config file to use as a template</BriefDescription>
</File>
<String Name="ClusterSection" Label="Cluster Section">
<DefaultValue>cluster default</DefaultValue>
</String>
<String Name="InstanceType" Label="AWS Instance Family">
<ChildrenDefinitions>
<String Name="c4" Label="Instance Type">
<DiscreteInfo>
<Value Enum="c4.large (2 vCPU, 3.75 GB, 500 Mbps)">c4.large</Value>
<Value Enum="c4.xlarge (4 vCPU, 7.5 GB, 750 Mbps)">c4.xlarge</Value>
<Value Enum="c4.2xlarge (8 vCPU, 15 GB, 1000 Mbps)">c4.2xlarge</Value>
<Value Enum="c4.4xlarge (16 vCPU, 30 GB, 2000 Mbps)">c4.4xlarge</Value>
<Value Enum="c4.8xlarge (36 vCPU, 60 GB, 4000 Mbps)">c4.8xlarge</Value>
</DiscreteInfo>
</String>
<String Name="c5" Label="Instance Type">
<DiscreteInfo>
<Value Enum="c5.large (2 vCPU, 4 GB, Up to 10 Gbps)">c5.large</Value>
<Value Enum="c5.xlarge (4 vCPU, 8 GB, Up to 10 Gbps)">c5.xlarge</Value>
<Value Enum="c5.2xlarge (8 vCPU, 16 GB, Up to 10 Gbps)">c5.2xlarge</Value>
<Value Enum="c5.4xlarge (16 vCPU, 32 GB, Up to 10 Gbps)">c5.4xlarge</Value>
<Value Enum="c5.9xlarge (36 vCPU, 72 GB, 10 Gbps)">c5.9xlarge</Value>
<Value Enum="c5.12xlarge (48 vCPU, 96 GB, 12 Gbps)">c5.12xlarge</Value>
<Value Enum="c5.18xlarge (72 vCPU, 144 GB, 25 Gbps)">c5.18xlarge</Value>
<Value Enum="c5.24xlarge (96 vCPU, 192 GB, 25 Gbps)">c5.24xlarge</Value>
</DiscreteInfo>
</String>
<String Name="c5n" Label="Instance Type">
<ChildrenDefinitions>
<Void Name="ElasticFabricAdapter" Label="Use Elastic Fabric Adapter (EFA)"
Optional="true" IsEnabledByDefault="false" />
</ChildrenDefinitions>
<DiscreteInfo>
<Value Enum="c5n.large (2 vCPU, 5.25 GB, Up to 25 Gbps)">c5n.large</Value>
<Value Enum="c5n.xlarge (4 vCPU, 10.5 GB, Up to 25 Gbps)">c5n.xlarge</Value>
<Value Enum="c5n.2xlarge (8 vCPU, 21 GB, Up to 25 Gbps)">c5n.2xlarge</Value>
<Value Enum="c5n.4xlarge (16 vCPU, 42 GB, Up to 25 Gbps)">c5n.4xlarge</Value>
<Value Enum="c5n.9xlarge (36 vCPU, 96 GB, 50 Gbps)">c5n.9xlarge</Value>
<Structure>
<Value Enum="c5n.18xlarge (72 vCPU, 192 GB, 100 Gbps)">c5n.18xlarge</Value>
<Items>
<Item>ElasticFabricAdapter</Item>
</Items>
</Structure>
</DiscreteInfo>
</String>
<String Name="t2" Label="Instance Type">
<DiscreteInfo>
<Value Enum="t2.micro (1 vCPU, 1 GB)">t2.micro</Value>
<Value Enum="t2.large (2 vCPU, 8 GB)">t2.large</Value>
<Value Enum="t2.xlarge (4 vCPU, 16GB)">t2.xlarge</Value>
<Value Enum="t2.2xlarge (8 vCPU, 32GB)">t2.2xlarge</Value>
</DiscreteInfo>
</String>
<String Name="t3" Label="Instance Type">
<DiscreteInfo>
<Value Enum="t3.large (2 vCPU, 8 GB)">t3.large</Value>
<Value Enum="t3.xlarge (4 vCPU, 16GB)">t3.xlarge</Value>
<Value Enum="t3.2xlarge (8 vCPU, 32GB)">t3.2xlarge</Value>
</DiscreteInfo>
</String>
<String Name="t3a" Label="Instance Type">
<DiscreteInfo>
<Value Enum="t3a.large (2 vCPU, 8 GB)">t3a.large</Value>
<Value Enum="t3a.xlarge (4 vCPU, 16GB)">t3a.xlarge</Value>
<Value Enum="t3a.2xlarge (8 vCPU, 32GB)">t3a.2xlarge</Value>
</DiscreteInfo>
</String>
<String Name="OtherType" Label="Other" />
</ChildrenDefinitions>
<DiscreteInfo>
<Structure>
<Value>c4</Value>
<Items>
<Item>c4</Item>
</Items>
</Structure>
<Structure>
<Value>c5</Value>
<Items>
<Item>c5</Item>
</Items>
</Structure>
<Structure>
<Value>c5n</Value>
<Items>
<Item>c5n</Item>
</Items>
</Structure>
<Structure>
<Value>t2</Value>
<Items>
<Item>t2</Item>
</Items>
</Structure>
<Structure>
<Value>t3</Value>
<Items>
<Item>t3</Item>
</Items>
</Structure>
<Structure>
<Value>t3a</Value>
<Items>
<Item>t3a</Item>
</Items>
</Structure>
<Structure>
<Value>Other</Value>
<Items>
<Item>OtherType</Item>
</Items>
</Structure>
</DiscreteInfo>
</String>
<Int Name="NumberOfNodes" Label="Number of nodes" Version="0">
<DefaultValue>1</DefaultValue>
<RangeInfo><Min Inclusive="true">1</Min></RangeInfo>
</Int>
<Int Name="Timeout" Label="Time limit" Units="min" Version="0">
<DefaultValue>5</DefaultValue>
<RangeInfo><Min Inclusive="true">1</Min></RangeInfo>
</Int>
</ItemDefinitions>
</Group>
</ItemDefinitions>
</AttDef>
</Definitions>
<Views>
<View Type="Instanced" Title="Export Settings" TopLevel="true" FilterByCategory="false" FilterByAdvanceLevel="true">
......
simulation-workflows/internal/writers/aws.py 0 → 100644
View file @ 88740949
#=============================================================================
#
# Copyright (c) Kitware, Inc.
# All rights reserved.
# See LICENSE.txt for details.
#
# This software is distributed WITHOUT ANY WARRANTY; without even
# the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
# PURPOSE. See the above copyright notice for more information.
#
#=============================================================================
import configparser
import datetime
import os
import warnings
print('loading', os.path.basename(__file__))
import smtk
import smtk.attribute
import smtk.io
# import smtk.model
# ---------------------------------------------------------------------
class AWSWriter:
""""""
def __init__(self, scope):
""""""
# Todo Check OS? (only support linux?)
# Check for genesis/exodus model files, which aren't supported
for path in scope.files_to_upload:
basename, ext = os.path.splitext(path)
if ext in ['.gen', '.exo', '.ex2']:
msg = 'AWS does not Genesis/Exodus models: {}'.format(ext)
print(msg)
RuntimeError(msg)
self.aws_item = None
self.scope = scope
def export_ace3p(self, aws_item):
self.aws_item = aws_item
# Create job folder
project_folder_item = self.scope.export_att.findDirectory('ProjectFolder')
if project_folder_item is None or \
not project_folder_item.isEnabled() or \
not project_folder_item.isSet():
msg = 'AWSWriter Error: ProjectFolder is not specified'
print(msg)
raise RuntimeError(msg)
dt = datetime.datetime.now()
folder_name = dt.strftime('aws-%y%m%d-%H%M%S')
# Generate cluster configuration, starting with specified default file
config_file_item = self.scope.export_att.itemAtPath('AWSSimulation/AWSConfigFile')
if config_file_item is None or not config_file_item.isSet():
msg = 'AWSWriter ERROR: AWS config file not specified'
print(msg)
raise RuntimeError(msg)
config_file_path = config_file_item.value()
if not os.path.exists(config_file_path):
msg = 'AWSWriter ERROR: AWS config file not found: {}'.format(config_file_path)
print(msg)
raise RuntimeError(msg)
# Get the cluster section to use
cluster_key_item = self.scope.export_att.itemAtPath('AWSSimulation/ClusterSection')
cluster_key = cluster_key_item.value()
# Load config file and set instance type
config = configparser.ConfigParser()
config.read_file(open(config_file_path))
instance_type_item = self.scope.export_att.itemAtPath('AWSSimulation/InstanceType')
# There must be one active child
if instance_type_item.numberOfActiveChildrenItems() != 1:
n = instance_type_item.numberOfActiveChildrenItems()
msg = 'AWSWriter ERROR: instance type should have 1 active child item, not {}'.format(n)
print(msg)
raise RuntimeError(msg)
instance_item = instance_type_item.activeChildItem(0)
config[cluster_key]['compute_instance_type'] = instance_item.value()
# Todo change master instance type to a compute node?
# config[cluster_key]['master_instance_type'] = instance_item.value()
# Todo Generate arguments for aws submit script, using:
# - NumberOfNodes item
# - Timeout (min) item
# Create job folder
folder_path = os.path.join(project_folder_item.value(), 'jobs', folder_name)
os.makedirs(folder_path)
# Write list of files to upload
upload_info_path = os.path.join(folder_path, 'files_to_upload.txt')
with open(upload_info_path, 'w') as upload_info_file:
for path in self.scope.files_to_upload:
upload_info_path.write(path)
upload_info_path.write('\n')
# Todo copy the ace3p input file (?)
# Write the config file
config_path = os.path.join(folder_path, 'parallelcluster.config')
with open(config_path, 'w') as configfile:
config.write(configfile)
# Write export attributes to the job folder
att_writer = smtk.io.AttributeWriter()
logger = smtk.io.Logger()
export_atts_filename = 'export.sbi'
export_atts_path = os.path.join(folder_path, export_atts_filename)
err = att_writer.write(self.scope.export_att.attributeResource(), export_atts_path, logger)
if err:
msg = 'AWSWriter Error: Unable to write export attribute file {}'.format(export_atts_path)
print(msg)
raise RuntimeError(msg)
# Create stamp subdirectory and touch "export" file
stamp_path = os.path.join(folder_path, 'stamp')
os.makedirs(stamp_path)
export_path = os.path.join(stamp_path, 'export')
with open(export_path, 'w') as st:
pass
print('Created job folder {}'.format(folder_path))
return True
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment