fix a potential new line char issue
This commit is contained in:
Azuki1029
Родитель
1e6ef21acd
Коммит
81f511683f
|
|
@ -101,14 +101,14 @@ elif $CGInstalled && ! $cgDisabled; then
|
|||
((maxLoop--))
|
||||
done
|
||||
else # processes would run away if pstree is not installed
|
||||
pid=$(pstree -l -p "$processId" | grep "([[:digit:]]*)" -o | tr -d '()')
|
||||
if [ -n "$pid" ]; then
|
||||
if [ "$forced" == "1" ]; then
|
||||
kill -s 9 $pid
|
||||
else
|
||||
kill -s SIGINT $pid
|
||||
fi
|
||||
fi
|
||||
pid=$(pstree -l -p "$processId" | grep "([[:digit:]]*)" -o | tr -d '()')
|
||||
if [ -n "$pid" ]; then
|
||||
if [ "$forced" == "1" ]; then
|
||||
kill -s 9 $pid
|
||||
else
|
||||
kill -s SIGINT $pid
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
|
||||
exit 0
|
||||
|
|
@ -56,7 +56,7 @@ if $isDockerTask; then
|
|||
|
||||
containerId=$(GetContainerId $taskFolder)
|
||||
docker exec $containerId useradd -m $userName
|
||||
docker exec $containerId chown $userName $taskFolder
|
||||
docker exec $containerId chown $userName $taskFolder
|
||||
if $isMpiTask && ! $skipSshSetup; then
|
||||
/bin/bash MpiContainerPreparation.sh $containerId $userName
|
||||
fi
|
||||
|
|
|
|||
|
|
@ -17,45 +17,45 @@ cp {TestMutualTrust.sh,WaitForTrust.sh} $taskFolder
|
|||
# Generate hostfile or machinefile for Intel MPI, Open MPI, MPICH or other MPI applications
|
||||
export CCP_MPI_HOSTFILE=$taskFolder/mpi_hostfile
|
||||
case "$CCP_MPI_HOSTFILE_FORMAT" in
|
||||
"1") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?":":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
"2") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?" slots=":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
"3") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?" ":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
*) echo $CCP_NODES_CORES | tr ' ' '\n' | sed -n 'n;p' > $CCP_MPI_HOSTFILE;;
|
||||
"1") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?":":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
"2") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?" slots=":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
"3") echo $CCP_NODES_CORES | awk '{ORS=NR%2==0?" ":"\n"}1' RS=" " | tail -n +2 > $CCP_MPI_HOSTFILE;;
|
||||
*) echo $CCP_NODES_CORES | tr ' ' '\n' | sed -n 'n;p' > $CCP_MPI_HOSTFILE;;
|
||||
esac
|
||||
|
||||
isDockerTask=$(CheckDockerEnvFileExist $taskFolder)
|
||||
if $isDockerTask; then
|
||||
containerId=$(GetContainerId $taskFolder)
|
||||
docker exec $containerId /bin/bash -c "$taskFolder/TestMutualTrust.sh $taskId $taskFolder $userName" &&\
|
||||
docker exec -u $userName -e CCP_MPI_HOSTFILE=$taskFolder/mpi_hostfile $containerId /bin/bash $runPath
|
||||
exit
|
||||
docker exec $containerId /bin/bash -c "$taskFolder/TestMutualTrust.sh $taskId $taskFolder $userName" &&\
|
||||
docker exec -u $userName -e CCP_MPI_HOSTFILE=$taskFolder/mpi_hostfile $containerId /bin/bash $runPath
|
||||
exit
|
||||
fi
|
||||
|
||||
cgDisabled=$(CheckCgroupDisabledInFlagFile $taskFolder)
|
||||
if ! $CGroupV1 && ! $cgDisabled; then
|
||||
groupName=$(GetCGroupName "$taskId")
|
||||
procsFile=$(GetCpusetTasksFileV2 "$groupName")
|
||||
echo $$ > "$procsFile"
|
||||
/bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
groupName=$(GetCGroupName "$taskId")
|
||||
procsFile=$(GetCpusetTasksFileV2 "$groupName")
|
||||
echo $$ > "$procsFile"
|
||||
/bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
elif $CGInstalled && ! $cgDisabled; then
|
||||
groupName=$(GetCGroupName "$taskId")
|
||||
group=$CGroupSubSys:$groupName
|
||||
cgexec -g "$group" /bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
cgexec -g "$group" su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
cgexec -g "$group" sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
groupName=$(GetCGroupName "$taskId")
|
||||
group=$CGroupSubSys:$groupName
|
||||
cgexec -g "$group" /bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
cgexec -g "$group" su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
cgexec -g "$group" sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
else
|
||||
/bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
/bin/bash $taskFolder/TestMutualTrust.sh "$taskId" "$taskFolder" "$userName" || exit
|
||||
if [ "$CCP_SWITCH_USER" == "1" ]; then
|
||||
su $userName -m -c "/bin/bash $runPath"
|
||||
else
|
||||
sudo -H -E -u $userName env "PATH=$PATH" /bin/bash $runPath
|
||||
fi
|
||||
fi
|
||||
|
|
@ -7,7 +7,7 @@ command -v cgexec > /dev/null 2>&1 && CGInstalled=true
|
|||
|
||||
CGroupV1=true
|
||||
if [ "$(stat -fc %T /sys/fs/cgroup/)" == "cgroup2fs" ]; then
|
||||
CGroupV1=false
|
||||
CGroupV1=false
|
||||
fi
|
||||
|
||||
CGroupV2Root="/sys/fs/cgroup"
|
||||
|
|
|
|||
|
|
@ -36,215 +36,215 @@ LogFile = '/opt/hpcnodemanager/logs/hpclinuxagent.log'
|
|||
RestartIntervalInSeconds = 60
|
||||
|
||||
def main():
|
||||
global config, logger
|
||||
logger = agentUtil.LoggerInit(LogFile, '/dev/stdout')
|
||||
logger.Log("The command line is: " + " ".join(sys.argv))
|
||||
config = configUtil.ConfigUtility(logger.Log, logger.Error)
|
||||
global config, logger
|
||||
logger = agentUtil.LoggerInit(LogFile, '/dev/stdout')
|
||||
logger.Log("The command line is: " + " ".join(sys.argv))
|
||||
config = configUtil.ConfigUtility(logger.Log, logger.Error)
|
||||
|
||||
ret = 1
|
||||
if re.match("^([-/]*)(disable)", sys.argv[1]):
|
||||
ret = disable()
|
||||
elif re.match("^([-/]*)(enable)", sys.argv[1]):
|
||||
ret = enable()
|
||||
elif re.match("^([-/]*)(daemon)", sys.argv[1]):
|
||||
ret = daemon()
|
||||
elif re.match("^([-/]*)(restart)", sys.argv[1]):
|
||||
disable()
|
||||
ret = enable()
|
||||
else:
|
||||
logger.Log("Invalid parameter %s" % sys.argv[1])
|
||||
sys.exit(ret)
|
||||
ret = 1
|
||||
if re.match("^([-/]*)(disable)", sys.argv[1]):
|
||||
ret = disable()
|
||||
elif re.match("^([-/]*)(enable)", sys.argv[1]):
|
||||
ret = enable()
|
||||
elif re.match("^([-/]*)(daemon)", sys.argv[1]):
|
||||
ret = daemon()
|
||||
elif re.match("^([-/]*)(restart)", sys.argv[1]):
|
||||
disable()
|
||||
ret = enable()
|
||||
else:
|
||||
logger.Log("Invalid parameter %s" % sys.argv[1])
|
||||
sys.exit(ret)
|
||||
|
||||
|
||||
def _is_nodemanager_daemon(pid):
|
||||
retcode, output = agentUtil.RunGetOutput("ps -p {0} -o cmd=".format(pid))
|
||||
if retcode == 0:
|
||||
logger.Log("The cmd for process {0} is {1}".format(pid, output))
|
||||
pattern = r'(.*[/\s])?{0}\s+[-/]*daemon$'.format(__file__)
|
||||
if re.match(pattern, output):
|
||||
return True
|
||||
logger.Log("The process {0} is not HPC Linux node manager daemon".format(pid))
|
||||
return False
|
||||
retcode, output = agentUtil.RunGetOutput("ps -p {0} -o cmd=".format(pid))
|
||||
if retcode == 0:
|
||||
logger.Log("The cmd for process {0} is {1}".format(pid, output))
|
||||
pattern = r'(.*[/\s])?{0}\s+[-/]*daemon$'.format(__file__)
|
||||
if re.match(pattern, output):
|
||||
return True
|
||||
logger.Log("The process {0} is not HPC Linux node manager daemon".format(pid))
|
||||
return False
|
||||
|
||||
|
||||
def _add_dns_search(domain_fqdn):
|
||||
need_update = False
|
||||
new_content = ''
|
||||
for line in (open('/etc/resolv.conf', 'r')).readlines():
|
||||
if re.match('^search.* {0}'.format(domain_fqdn), line):
|
||||
logger.Log('{0} was already added in /etc/resolv.conf'.format(domain_fqdn))
|
||||
return
|
||||
if re.match('^search', line):
|
||||
need_update = True
|
||||
new_content += line.replace('search', 'search {0}'.format(domain_fqdn))
|
||||
else:
|
||||
new_content += line
|
||||
if need_update:
|
||||
logger.Log('Adding {0} to /etc/resolv.conf'.format(domain_fqdn))
|
||||
agentUtil.SetFileContents('/etc/resolv.conf', new_content)
|
||||
need_update = False
|
||||
new_content = ''
|
||||
for line in (open('/etc/resolv.conf', 'r')).readlines():
|
||||
if re.match('^search.* {0}'.format(domain_fqdn), line):
|
||||
logger.Log('{0} was already added in /etc/resolv.conf'.format(domain_fqdn))
|
||||
return
|
||||
if re.match('^search', line):
|
||||
need_update = True
|
||||
new_content += line.replace('search', 'search {0}'.format(domain_fqdn))
|
||||
else:
|
||||
new_content += line
|
||||
if need_update:
|
||||
logger.Log('Adding {0} to /etc/resolv.conf'.format(domain_fqdn))
|
||||
agentUtil.SetFileContents('/etc/resolv.conf', new_content)
|
||||
|
||||
def _mount_cgroup():
|
||||
if not os.path.isdir('/cgroup'):
|
||||
os.mkdir('/cgroup')
|
||||
if not os.listdir('/cgroup'):
|
||||
retcode, mount_msg = agentUtil.RunGetOutput('mount -t cgroup cgroup /cgroup')
|
||||
logger.Log("mount /cgroup directory {0}:{1}".format(retcode, mount_msg))
|
||||
if retcode == 0:
|
||||
logger.Log("/cgroup directory is successfully mounted.")
|
||||
else:
|
||||
raise Exception("failed to mount /cgroup directory")
|
||||
else:
|
||||
logger.Log("/cgroup directory was already mounted.")
|
||||
if not os.path.isdir('/cgroup'):
|
||||
os.mkdir('/cgroup')
|
||||
if not os.listdir('/cgroup'):
|
||||
retcode, mount_msg = agentUtil.RunGetOutput('mount -t cgroup cgroup /cgroup')
|
||||
logger.Log("mount /cgroup directory {0}:{1}".format(retcode, mount_msg))
|
||||
if retcode == 0:
|
||||
logger.Log("/cgroup directory is successfully mounted.")
|
||||
else:
|
||||
raise Exception("failed to mount /cgroup directory")
|
||||
else:
|
||||
logger.Log("/cgroup directory was already mounted.")
|
||||
|
||||
def configure_iptables():
|
||||
if agentUtil.Run("command -v iptables", chk_err=False) == 0:
|
||||
agentUtil.Run("iptables -D INPUT -p tcp --dport 40000 -j ACCEPT", chk_err=False)
|
||||
agentUtil.Run("iptables -D INPUT -p tcp --dport 40002 -j ACCEPT", chk_err=False)
|
||||
agentUtil.Run("iptables -I INPUT -p tcp --dport 40000 -j ACCEPT")
|
||||
agentUtil.Run("iptables -I INPUT -p tcp --dport 40002 -j ACCEPT")
|
||||
if agentUtil.Run("command -v iptables", chk_err=False) == 0:
|
||||
agentUtil.Run("iptables -D INPUT -p tcp --dport 40000 -j ACCEPT", chk_err=False)
|
||||
agentUtil.Run("iptables -D INPUT -p tcp --dport 40002 -j ACCEPT", chk_err=False)
|
||||
agentUtil.Run("iptables -I INPUT -p tcp --dport 40000 -j ACCEPT")
|
||||
agentUtil.Run("iptables -I INPUT -p tcp --dport 40002 -j ACCEPT")
|
||||
|
||||
def enable():
|
||||
# Check whether monitor process is running.
|
||||
# If it does, return. Otherwise clear pid file
|
||||
if os.path.isfile(PidFilePath):
|
||||
pid = agentUtil.GetFileContents(PidFilePath)
|
||||
if os.path.isdir(os.path.join("/proc", pid)) and _is_nodemanager_daemon(pid):
|
||||
logger.Log('HPC Linux node manager daemon is already running')
|
||||
return 0
|
||||
os.remove(PidFilePath)
|
||||
# Check whether monitor process is running.
|
||||
# If it does, return. Otherwise clear pid file
|
||||
if os.path.isfile(PidFilePath):
|
||||
pid = agentUtil.GetFileContents(PidFilePath)
|
||||
if os.path.isdir(os.path.join("/proc", pid)) and _is_nodemanager_daemon(pid):
|
||||
logger.Log('HPC Linux node manager daemon is already running')
|
||||
return 0
|
||||
os.remove(PidFilePath)
|
||||
|
||||
args = [get_python_executor(), os.path.join(os.getcwd(), __file__), "daemon"]
|
||||
devnull = open(os.devnull, 'w')
|
||||
child = subprocess.Popen(args, stdout=devnull, stderr=devnull, preexec_fn=os.setsid)
|
||||
if child.pid is None or child.pid < 1:
|
||||
logger.Error('Failed to launch HPC Linux node manager daemon')
|
||||
return 1
|
||||
else:
|
||||
# Sleep 3 seconds to check if the process is still running
|
||||
time.sleep(3)
|
||||
if child.poll() is None:
|
||||
agentUtil.SetFileContents(PidFilePath, str(child.pid))
|
||||
logger.Log("Daemon pid: {0}".format(child.pid))
|
||||
logger.Log('HPC Linux node manager daemon is enabled')
|
||||
return 0
|
||||
else:
|
||||
logger.Log('Failed to launch HPC Linux node manager daemon')
|
||||
return 1
|
||||
args = [get_python_executor(), os.path.join(os.getcwd(), __file__), "daemon"]
|
||||
devnull = open(os.devnull, 'w')
|
||||
child = subprocess.Popen(args, stdout=devnull, stderr=devnull, preexec_fn=os.setsid)
|
||||
if child.pid is None or child.pid < 1:
|
||||
logger.Error('Failed to launch HPC Linux node manager daemon')
|
||||
return 1
|
||||
else:
|
||||
# Sleep 3 seconds to check if the process is still running
|
||||
time.sleep(3)
|
||||
if child.poll() is None:
|
||||
agentUtil.SetFileContents(PidFilePath, str(child.pid))
|
||||
logger.Log("Daemon pid: {0}".format(child.pid))
|
||||
logger.Log('HPC Linux node manager daemon is enabled')
|
||||
return 0
|
||||
else:
|
||||
logger.Log('Failed to launch HPC Linux node manager daemon')
|
||||
return 1
|
||||
|
||||
def get_dist_info():
|
||||
distroName = ''
|
||||
distroVersion = ''
|
||||
if 'linux_distribution' in dir(platform):
|
||||
distinfo = platform.linux_distribution(full_distribution_name=0)
|
||||
distroName = distinfo[0].strip()
|
||||
distroVersion = distinfo[1]
|
||||
# if the distroName is empty we get from /etc/*-release
|
||||
if not distroName:
|
||||
errCode, info = agentUtil.RunGetOutput("cat /etc/*-release")
|
||||
if errCode != 0:
|
||||
raise Exception('Failed to get Linux Distro info by running command "cat /etc/*release", error code: {}'.format(errCode))
|
||||
for line in info.splitlines():
|
||||
if line.startswith('PRETTY_NAME='):
|
||||
line = line.lower()
|
||||
if 'ubuntu' in line:
|
||||
distroName = 'ubuntu'
|
||||
elif 'centos' in line:
|
||||
distroName = 'centos'
|
||||
elif 'red hat' in line:
|
||||
distroName = 'redhat'
|
||||
elif 'suse' in line:
|
||||
distroName = 'suse'
|
||||
elif 'almalinux' in line:
|
||||
distroName = 'almalinux'
|
||||
elif 'rocky' in line:
|
||||
distroName = 'rocky'
|
||||
elif 'fedora' in line:
|
||||
distroName = 'fedora'
|
||||
elif 'freebsd' in line:
|
||||
distroName = 'freebsd'
|
||||
else:
|
||||
raise Exception('Unknown linux distribution with {}'.format(line))
|
||||
if line.startswith('VERSION_ID='):
|
||||
line = line.strip(' ')
|
||||
quoteIndex = line.index('"')
|
||||
if quoteIndex >= 0:
|
||||
distroVersion = line[quoteIndex+1:-1]
|
||||
return distroName.lower(), distroVersion
|
||||
distroName = ''
|
||||
distroVersion = ''
|
||||
if 'linux_distribution' in dir(platform):
|
||||
distinfo = platform.linux_distribution(full_distribution_name=0)
|
||||
distroName = distinfo[0].strip()
|
||||
distroVersion = distinfo[1]
|
||||
# if the distroName is empty we get from /etc/*-release
|
||||
if not distroName:
|
||||
errCode, info = agentUtil.RunGetOutput("cat /etc/*-release")
|
||||
if errCode != 0:
|
||||
raise Exception('Failed to get Linux Distro info by running command "cat /etc/*release", error code: {}'.format(errCode))
|
||||
for line in info.splitlines():
|
||||
if line.startswith('PRETTY_NAME='):
|
||||
line = line.lower()
|
||||
if 'ubuntu' in line:
|
||||
distroName = 'ubuntu'
|
||||
elif 'centos' in line:
|
||||
distroName = 'centos'
|
||||
elif 'red hat' in line:
|
||||
distroName = 'redhat'
|
||||
elif 'suse' in line:
|
||||
distroName = 'suse'
|
||||
elif 'almalinux' in line:
|
||||
distroName = 'almalinux'
|
||||
elif 'rocky' in line:
|
||||
distroName = 'rocky'
|
||||
elif 'fedora' in line:
|
||||
distroName = 'fedora'
|
||||
elif 'freebsd' in line:
|
||||
distroName = 'freebsd'
|
||||
else:
|
||||
raise Exception('Unknown linux distribution with {}'.format(line))
|
||||
if line.startswith('VERSION_ID='):
|
||||
line = line.strip(' ')
|
||||
quoteIndex = line.index('"')
|
||||
if quoteIndex >= 0:
|
||||
distroVersion = line[quoteIndex+1:-1]
|
||||
return distroName.lower(), distroVersion
|
||||
|
||||
def get_python_executor():
|
||||
cmd = ''
|
||||
if sys.version_info.major == 2:
|
||||
cmd = 'python2'
|
||||
elif sys.version_info.major == 3:
|
||||
cmd = 'python3'
|
||||
if agentUtil.Run("command -v {0}".format(cmd), chk_err=False) != 0:
|
||||
# If a user-installed python isn't available, check for a platform-python. This is typically only used in RHEL 8.0.
|
||||
if agentUtil.Run("command -v /usr/libexec/platform-python", chk_err=False) == 0:
|
||||
cmd = '/usr/libexec/platform-python'
|
||||
return cmd
|
||||
cmd = ''
|
||||
if sys.version_info.major == 2:
|
||||
cmd = 'python2'
|
||||
elif sys.version_info.major == 3:
|
||||
cmd = 'python3'
|
||||
if agentUtil.Run("command -v {0}".format(cmd), chk_err=False) != 0:
|
||||
# If a user-installed python isn't available, check for a platform-python. This is typically only used in RHEL 8.0.
|
||||
if agentUtil.Run("command -v /usr/libexec/platform-python", chk_err=False) == 0:
|
||||
cmd = '/usr/libexec/platform-python'
|
||||
return cmd
|
||||
|
||||
def daemon():
|
||||
conn_string = config.get_cluster_connectionstring()
|
||||
logger.Log("The connection string is " + conn_string)
|
||||
dns_name = conn_string.split(',')[0].strip()
|
||||
if dns_name.find('.') > 0:
|
||||
# The cluster name is FQDN, extract the domain FQDN
|
||||
domain_fqdn = dns_name.split(".", 1)[1]
|
||||
logger.Log("The domain FQDN is " + domain_fqdn)
|
||||
_add_dns_search(domain_fqdn)
|
||||
conn_string = config.get_cluster_connectionstring()
|
||||
logger.Log("The connection string is " + conn_string)
|
||||
dns_name = conn_string.split(',')[0].strip()
|
||||
if dns_name.find('.') > 0:
|
||||
# The cluster name is FQDN, extract the domain FQDN
|
||||
domain_fqdn = dns_name.split(".", 1)[1]
|
||||
logger.Log("The domain FQDN is " + domain_fqdn)
|
||||
_add_dns_search(domain_fqdn)
|
||||
|
||||
if os.path.exists("/sys/fs/cgroup/cgroup.controllers"):
|
||||
agentUtil.Run("echo \"+cpu +cpuset +memory\" > /sys/fs/cgroup/cgroup.subtree_control")
|
||||
logger.Log("Cgroup v2 subsystem enabled: cpu, cpuset, memory")
|
||||
if os.path.exists("/sys/fs/cgroup/cgroup.controllers"):
|
||||
agentUtil.Run("echo \"+cpu +cpuset +memory\" > /sys/fs/cgroup/cgroup.subtree_control")
|
||||
logger.Log("Cgroup v2 subsystem enabled: cpu, cpuset, memory")
|
||||
|
||||
try:
|
||||
# Mount the directory /cgroup for centos 6.*
|
||||
distroName, distroVersion = get_dist_info()
|
||||
major_version = int(distroVersion.split('.')[0])
|
||||
if (distroName == 'centos' or distroName == 'redhat') and major_version < 7:
|
||||
_mount_cgroup()
|
||||
logger.Log("Configure iptables to allow incoming tcp connection to 40000 and 40002.")
|
||||
configure_iptables()
|
||||
while True:
|
||||
exe_path = os.path.join(InstallRoot, "nodemanager")
|
||||
devnull = open(os.devnull, 'w')
|
||||
child_process = subprocess.Popen(exe_path, stdout=devnull, stderr=devnull, cwd=InstallRoot)
|
||||
if child_process.pid is None or child_process.pid < 1:
|
||||
logger.Log('Failed to start HPC node manager process')
|
||||
return 1
|
||||
else:
|
||||
# Sleep 1 second to check if the process is still running
|
||||
time.sleep(1)
|
||||
if child_process.poll() is None:
|
||||
logger.Log('HPC node manager process started')
|
||||
exit_code = child_process.wait()
|
||||
exit_msg = "HPC node manager process exits: {0}".format(exit_code)
|
||||
logger.Warn(exit_msg)
|
||||
else:
|
||||
exit_msg = "HPC node manager process crashes: {0}".format(child_process.returncode)
|
||||
logger.Error(exit_msg)
|
||||
logger.Log("Restart HPC node manager process after {0} seconds".format(RestartIntervalInSeconds))
|
||||
time.sleep(RestartIntervalInSeconds)
|
||||
try:
|
||||
# Mount the directory /cgroup for centos 6.*
|
||||
distroName, distroVersion = get_dist_info()
|
||||
major_version = int(distroVersion.split('.')[0])
|
||||
if (distroName == 'centos' or distroName == 'redhat') and major_version < 7:
|
||||
_mount_cgroup()
|
||||
logger.Log("Configure iptables to allow incoming tcp connection to 40000 and 40002.")
|
||||
configure_iptables()
|
||||
while True:
|
||||
exe_path = os.path.join(InstallRoot, "nodemanager")
|
||||
devnull = open(os.devnull, 'w')
|
||||
child_process = subprocess.Popen(exe_path, stdout=devnull, stderr=devnull, cwd=InstallRoot)
|
||||
if child_process.pid is None or child_process.pid < 1:
|
||||
logger.Log('Failed to start HPC node manager process')
|
||||
return 1
|
||||
else:
|
||||
# Sleep 1 second to check if the process is still running
|
||||
time.sleep(1)
|
||||
if child_process.poll() is None:
|
||||
logger.Log('HPC node manager process started')
|
||||
exit_code = child_process.wait()
|
||||
exit_msg = "HPC node manager process exits: {0}".format(exit_code)
|
||||
logger.Warn(exit_msg)
|
||||
else:
|
||||
exit_msg = "HPC node manager process crashes: {0}".format(child_process.returncode)
|
||||
logger.Error(exit_msg)
|
||||
logger.Log("Restart HPC node manager process after {0} seconds".format(RestartIntervalInSeconds))
|
||||
time.sleep(RestartIntervalInSeconds)
|
||||
|
||||
except Exception as e:
|
||||
logger.Error("Failed to start the daemon with error:{0}, stack trace: {1}".format(e, traceback.format_exc()))
|
||||
return 1
|
||||
except Exception as e:
|
||||
logger.Error("Failed to start the daemon with error:{0}, stack trace: {1}".format(e, traceback.format_exc()))
|
||||
return 1
|
||||
|
||||
|
||||
def disable():
|
||||
# Check whether daemon process is running.
|
||||
# If it does, kill it. Otherwise clear pid file
|
||||
if os.path.isfile(PidFilePath):
|
||||
pid = agentUtil.GetFileContents(PidFilePath)
|
||||
if os.path.isdir(os.path.join("/proc", pid)) and _is_nodemanager_daemon(pid):
|
||||
logger.Log("Stop HPC node manager daemon: {0}".format(pid))
|
||||
os.killpg(int(pid), 9)
|
||||
logger.Log('HPC node manager daemon is disabled')
|
||||
os.remove(PidFilePath)
|
||||
else:
|
||||
logger.Log('HPC node manager daemon is not running')
|
||||
return 0
|
||||
# Check whether daemon process is running.
|
||||
# If it does, kill it. Otherwise clear pid file
|
||||
if os.path.isfile(PidFilePath):
|
||||
pid = agentUtil.GetFileContents(PidFilePath)
|
||||
if os.path.isdir(os.path.join("/proc", pid)) and _is_nodemanager_daemon(pid):
|
||||
logger.Log("Stop HPC node manager daemon: {0}".format(pid))
|
||||
os.killpg(int(pid), 9)
|
||||
logger.Log('HPC node manager daemon is disabled')
|
||||
os.remove(PidFilePath)
|
||||
else:
|
||||
logger.Log('HPC node manager daemon is not running')
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
||||
main()
|
||||
|
|
@ -26,8 +26,8 @@ PIDFILE=/var/run/hpcnmdaemon.pid
|
|||
|
||||
# Exit if not run as root
|
||||
if [[ $EUID != 0 ]]; then
|
||||
echo "This script must be run as root"
|
||||
exit 1
|
||||
echo "This script must be run as root"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Exit if the package is not installed
|
||||
|
|
@ -37,11 +37,11 @@ fi
|
|||
[ -r /etc/default/$NAME ] && . /etc/default/$NAME
|
||||
|
||||
if command -v python3 >/dev/null 2>&1 ; then
|
||||
PYTHONEXECUTOR="python3"
|
||||
PYTHONEXECUTOR="python3"
|
||||
elif command -v python2 >/dev/null 2>&1 ; then
|
||||
PYTHONEXECUTOR="python2"
|
||||
PYTHONEXECUTOR="python2"
|
||||
elif command -v /usr/libexec/platform-python >/dev/null 2>&1 ; then
|
||||
PYTHONEXECUTOR="/usr/libexec/platform-python"
|
||||
PYTHONEXECUTOR="/usr/libexec/platform-python"
|
||||
fi
|
||||
|
||||
#
|
||||
|
|
@ -49,16 +49,16 @@ fi
|
|||
#
|
||||
do_start()
|
||||
{
|
||||
local RC=0
|
||||
echo "Starting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath enable
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was started"
|
||||
else
|
||||
echo "Failed to start $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
local RC=0
|
||||
echo "Starting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath enable
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was started"
|
||||
else
|
||||
echo "Failed to start $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
}
|
||||
|
||||
#
|
||||
|
|
@ -66,16 +66,16 @@ do_start()
|
|||
#
|
||||
do_stop()
|
||||
{
|
||||
local RC=0
|
||||
echo "Starting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath disable
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was stopped"
|
||||
else
|
||||
echo "Failed to stop $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
local RC=0
|
||||
echo "Starting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath disable
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was stopped"
|
||||
else
|
||||
echo "Failed to stop $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
}
|
||||
|
||||
#
|
||||
|
|
@ -83,54 +83,54 @@ do_stop()
|
|||
#
|
||||
do_restart()
|
||||
{
|
||||
local RC=0
|
||||
echo "restarting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath restart
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was restarted"
|
||||
else
|
||||
echo "Failed to restart $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
local RC=0
|
||||
echo "restarting $FriendlyName"
|
||||
$PYTHONEXECUTOR $AgentPath restart
|
||||
RC=$?
|
||||
if [ $RC = 0 ]; then
|
||||
echo "$FriendlyName was restarted"
|
||||
else
|
||||
echo "Failed to restart $FriendlyName : $RC"
|
||||
fi
|
||||
return $RC
|
||||
}
|
||||
|
||||
do_status()
|
||||
{
|
||||
local daemon_pid=0
|
||||
[ -r $PIDFILE ] && read daemon_pid < $PIDFILE
|
||||
if [ $daemon_pid != 0 ]; then
|
||||
local cmdline=`ps -p $daemon_pid -o cmd=`
|
||||
[[ "$cmdline" = *"$AgentPath"* ]] || daemon_pid=0
|
||||
fi
|
||||
if [ $daemon_pid = 0 ]; then
|
||||
echo "$FriendlyName is not running"
|
||||
else
|
||||
echo "$FriendlyName is running"
|
||||
fi
|
||||
return 0
|
||||
local daemon_pid=0
|
||||
[ -r $PIDFILE ] && read daemon_pid < $PIDFILE
|
||||
if [ $daemon_pid != 0 ]; then
|
||||
local cmdline=`ps -p $daemon_pid -o cmd=`
|
||||
[[ "$cmdline" = *"$AgentPath"* ]] || daemon_pid=0
|
||||
fi
|
||||
if [ $daemon_pid = 0 ]; then
|
||||
echo "$FriendlyName is not running"
|
||||
else
|
||||
echo "$FriendlyName is running"
|
||||
fi
|
||||
return 0
|
||||
}
|
||||
|
||||
case "$1" in
|
||||
start)
|
||||
do_start
|
||||
RETVAL=$?
|
||||
;;
|
||||
do_start
|
||||
RETVAL=$?
|
||||
;;
|
||||
stop)
|
||||
do_stop
|
||||
RETVAL=$?
|
||||
;;
|
||||
do_stop
|
||||
RETVAL=$?
|
||||
;;
|
||||
status)
|
||||
do_status
|
||||
RETVAL=$?
|
||||
;;
|
||||
do_status
|
||||
RETVAL=$?
|
||||
;;
|
||||
restart|reload|force-reload)
|
||||
do_restart
|
||||
RETVAL=$?
|
||||
;;
|
||||
do_restart
|
||||
RETVAL=$?
|
||||
;;
|
||||
*)
|
||||
echo "Usage: $SCRIPTNAME {start|stop|status|restart|force-reload}" >&2
|
||||
RETVAL=3
|
||||
;;
|
||||
echo "Usage: $SCRIPTNAME {start|stop|status|restart|force-reload}" >&2
|
||||
RETVAL=3
|
||||
;;
|
||||
esac
|
||||
exit $RETVAL
|
||||
Разница между файлами не показана из-за своего большого размера
Загрузить разницу
Загрузка…
Ссылка в новой задаче