def configure_htcondor(self):
"""
Configure environment for running HTCondor service over a node.
"""
all_done = False
try:
htcondor_params = {}
if self.srv_type == "master":
condor_template = conf_manager.load_conf_template(conf_manager.HTCONDOR_MASTER_CONF_TEMPLATE)
# log.debug("Condor template: {0}".format(condor_template))
htcondor_params["flock_host"] = self.flock_to
else:
condor_template = conf_manager.load_conf_template(conf_manager.HTCONDOR_WOORKER_CONF_TEMPLATE)
htcondor_params = {
"host": self.host
}
log.debug("HTCondor params: {0}".format(str(htcondor_params)))
condor_template = condor_template.substitute(htcondor_params)
if os.path.exists(paths.P_HTCONDOR_CONFIG_PATH):
with open(paths.P_HTCONDOR_CONFIG_PATH, 'a') as f:
print >> f, condor_template
misc.run(paths.P_HTCONDOR_HOME + "/condor restart")
all_done = True
self.state = service_states.RUNNING
else:
log.error("HTCondor config file {0} not found!"
.format(paths.P_HTCONDOR_CONFIG_PATH))
except Exception, e:
log.debug("Error while configuring HTCondor: {0}".format(e))
self.state = service_states.ERROR
all_done = False
def configure_htcondor(self):
"""
Configure environment for running HTCondor service over a node.
"""
all_done = False
try:
htcondor_params = {}
if self.srv_type == "master":
condor_template = conf_manager.load_conf_template(
conf_manager.HTCONDOR_MASTER_CONF_TEMPLATE)
# log.debug("Condor template: {0}".format(condor_template))
htcondor_params["flock_host"] = self.flock_to
else:
condor_template = conf_manager.load_conf_template(
conf_manager.HTCONDOR_WOORKER_CONF_TEMPLATE)
htcondor_params = {"host": self.host}
log.debug("HTCondor params: {0}".format(str(htcondor_params)))
condor_template = condor_template.substitute(htcondor_params)
if os.path.exists(paths.P_HTCONDOR_CONFIG_PATH):
with open(paths.P_HTCONDOR_CONFIG_PATH, 'a') as f:
print >> f, condor_template
misc.run(paths.P_HTCONDOR_HOME + "/condor restart")
all_done = True
self.state = service_states.RUNNING
else:
log.error("HTCondor config file {0} not found!".format(
paths.P_HTCONDOR_CONFIG_PATH))
except Exception, e:
log.debug("Error while configuring HTCondor: {0}".format(e))
self.state = service_states.ERROR
all_done = False
def _get_sge_install_conf(app, host):
"""Fills and returns the SGE_INSTALL_TEMPLATE as a string.
Positional arguments:
app -- The cloudman app
host -- the host param string is used for admin, submit and exec hosts
"""
# Add master as an execution host
# Additional execution hosts will be added later, as they start
sge_install_template = conf_manager.load_conf_template(
conf_manager.SGE_INSTALL_TEMPLATE)
sge_params = {
"cluster_name": "GalaxyEC2",
"admin_host_list": host,
"submit_host_list": host,
"exec_host_list": host,
"hostname_resolving": "true",
}
for key, value in app.config.iteritems():
if key.startswith("sge_"):
key = key[len("sge_"):]
sge_params[key] = value
return sge_install_template.substitute(sge_params)
def configure_proftpd(self):
"""
Configure environment for running ProFTPd service.
"""
log.debug("Configuring ProFTPd")
# Because we're rewriting the proftpd config file below, update the
# password for PostgreSQL galaxyftp user
gftp_pwd = self.app.path_resolver.proftpd_galaxyftp_user_pwd
log.debug("Setting psql password for galaxyftp role to {0}".format(gftp_pwd))
# Check if galaxtftp role already exists
cmd = ('{0} - postgres -c"{1} -p {2} -tAc\\\"SELECT 1 FROM pg_roles WHERE rolname=\'galaxyftp\'\\\""'
.format(paths.P_SU, self.app.path_resolver.psql_cmd,
self.app.path_resolver.psql_db_port))
role = 'ALTER' if misc.getoutput(cmd) == '1' else 'CREATE'
# Now either CREATE or ALTER the galaxyftp role to set the permissions
cmd = ('{0} - postgres -c"{1} -p {4} -c\\\"{2} ROLE galaxyftp LOGIN PASSWORD \'{3}\'\\\""'
.format(paths.P_SU, self.app.path_resolver.psql_cmd, role,
gftp_pwd, self.app.path_resolver.psql_db_port))
misc.run(cmd)
# Update the config to match the current environment
proftpd_tmplt = conf_manager.PROFTPD_CONF_TEMPLATE
proftpd_conf_template = conf_manager.load_conf_template(proftpd_tmplt)
params = {
'galaxy_user_name': paths.GALAXY_USER_NAME,
'galaxyftp_user_name': 'galaxyftp',
'psql_galaxyftp_password': gftp_pwd,
'galaxy_db_port': self.app.path_resolver.psql_db_port,
'galaxyFS_base_path': self.app.path_resolver.galaxy_data,
'public_ip_address': self.app.cloud_interface.get_public_ip()
}
template = proftpd_conf_template.substitute(params)
# Write out the config file
with open(self.app.path_resolver.proftpd_conf_file, 'w') as f:
print >> f, template
log.debug("Updated ProFTPd conf file {0}".format(
self.app.path_resolver.proftpd_conf_file))
# Place the FTP welcome message file
urllib.urlretrieve("https://s3.amazonaws.com/cloudman/files/proftpd_welcome.txt",
"/usr/proftpd/etc/welcome_msg.txt")
# Setup the Galaxy data dir for FTP
ftp_data_dir = '%s/tmp/ftp' % self.app.path_resolver.galaxy_data
if not os.path.exists(ftp_data_dir):
os.makedirs(ftp_data_dir)
attempt_chown_galaxy(ftp_data_dir)
# Some images have vsFTPd server included so stop it first
vsFTPd_exists = misc.run('status vsftpd', quiet=True)
if vsFTPd_exists and 'start/running' in vsFTPd_exists:
log.debug("Stopping vsFTPd")
misc.run('stop vsftpd')
# Start the server now
if misc.run('/etc/init.d/proftpd start'):
self.state = service_states.RUNNING
return True
else:
log.error("Trouble starting ProFTPd")
self.state = service_states.ERROR
return False
def _build_slurm_conf():
log.debug("Setting slurm.conf parameters")
# Make sure the slurm root dir exists and is owned by slurm user
misc.make_dir(self.app.path_resolver.slurm_root_tmp)
os.chown(self.app.path_resolver.slurm_root_tmp,
pwd.getpwnam("slurm")[2], grp.getgrnam("slurm")[2])
worker_nodes, worker_names = _worker_nodes_conf()
slurm_conf_template = conf_manager.load_conf_template(conf_manager.SLURM_CONF_TEMPLATE)
slurm_conf_params = {
"master_hostname": misc.get_hostname(),
"num_cpus": max(self.app.manager.num_cpus - 1, 1), # Reserve 1 CPU
"total_memory": max(1, self.app.manager.total_memory / 1024),
"slurm_root_tmp": self.app.path_resolver.slurm_root_tmp,
"worker_nodes": worker_nodes,
"worker_names": worker_names
}
return slurm_conf_template.substitute(slurm_conf_params)
def _build_slurm_conf():
log.debug("Setting slurm.conf parameters")
# Make sure the slurm root dir exists and is owned by slurm user
misc.make_dir(self.app.path_resolver.slurm_root_tmp)
os.chown(self.app.path_resolver.slurm_root_tmp,
pwd.getpwnam("slurm")[2], grp.getgrnam("slurm")[2])
worker_nodes, worker_names = _worker_nodes_conf()
slurm_conf_template = conf_manager.load_conf_template(conf_manager.SLURM_CONF_TEMPLATE)
slurm_conf_params = {
"master_hostname": misc.get_hostname(),
"num_cpus": max(self.app.manager.num_cpus - 1, 1), # Reserve 1 CPU
"total_memory": max(1, self.app.manager.total_memory / 1024),
"slurm_root_tmp": self.app.path_resolver.slurm_root_tmp,
"worker_nodes": worker_nodes,
"worker_names": worker_names
}
return slurm_conf_template.substitute(slurm_conf_params)
def _configure(self):
"""
Setup Supervisor for running via CloudMan by creating
``supervisord.conf`` file.
"""
log.debug("Configuring supervisord")
# Create supervisord config file
sv_vars = {
'supervisord_pid': self.pid_file,
'sv_port': self.sv_port,
'conf_dir': self.conf_dir,
'supervisord_log': self.log_file
}
template = conf_manager.load_conf_template(conf_manager.SUPERVISOR_TEMPLATE)
misc.write_template_file(template, sv_vars, self.main_conf_file)
# Make sure the config dir exists for programs managed by supervisor
misc.make_dir(self.conf_dir)
def _configure(self):
"""
Setup Supervisor for running via CloudMan by creating
``supervisord.conf`` file.
"""
log.debug("Configuring supervisord")
# Create supervisord config file
sv_vars = {
'supervisord_pid': self.pid_file,
'sv_port': self.sv_port,
'conf_dir': self.conf_dir,
'supervisord_log': self.log_file
}
template = conf_manager.load_conf_template(
conf_manager.SUPERVISOR_TEMPLATE)
misc.write_template_file(template, sv_vars, self.main_conf_file)
# Make sure the config dir exists for programs managed by supervisor
misc.make_dir(self.conf_dir)
def _update_job_conf(app, num_handlers, plugin_id='slurm'):
"""
Update Galaxy job_conf.xml.
At the moment, only job handlers are set. Number of handlers are added that
correspond to ``num_handlers``.
"""
galaxy_config_dir = app.path_resolver.galaxy_config_dir
job_conf_file_path = join(galaxy_config_dir, 'job_conf.xml')
log.debug("Updating Galaxy's job conf file {0}".format(job_conf_file_path))
template = conf_manager.load_conf_template(
conf_manager.GALAXY_JOB_CONF_TEMPLATE)
handler_xml = ""
for i in range(num_handlers):
handler_xml += (
'\t\t<handler id="handler{0}" tags="handlers" />\n'.format(i))
params = {'cloudman_handlers': handler_xml}
misc.write_template_file(template, params, job_conf_file_path)
def _write_template_file(self, template_file, parameters, conf_file):
"""
Given a plain text `template_file` path and appropriate `parameters`,
load the file as a `string.Template`, substitute the `parameters` and
write out the file to the `conf_file` path.
"""
template = conf_manager.load_conf_template(template_file)
try:
t = template.substitute(parameters)
# create conf directory if required
if not os.path.exists(os.path.dirname(conf_file)):
log.debug("Configuration path does not exist. Creating path: {0}".format(os.path.dirname(conf_file)))
os.makedirs(os.path.dirname(conf_file))
# Write out the file
with open(conf_file, 'w') as f:
print >> f, t
log.debug("Wrote Nginx config file {0}".format(conf_file))
except KeyError, kexc:
log.error("KeyError filling template {0}: {1}".format(template_file,
kexc))
def _write_template_file(self, template_file, parameters, conf_file):
"""
Given a plain text `template_file` path and appropriate `parameters`,
load the file as a `string.Template`, substitute the `parameters` and
write out the file to the `conf_file` path.
"""
template = conf_manager.load_conf_template(template_file)
try:
t = template.substitute(parameters)
# create conf directory if required
if not os.path.exists(os.path.dirname(conf_file)):
log.debug("Configuration path does not exist. Creating path: {0}".format(os.path.dirname(conf_file)))
os.makedirs(os.path.dirname(conf_file))
# Write out the file
with open(conf_file, 'w') as f:
print >> f, t
log.debug("Wrote Nginx config file {0}".format(conf_file))
except KeyError, kexc:
log.error("KeyError filling template {0}: {1}".format(template_file,
kexc))
def _setup(self):
log.debug("Running GalaxyReportsService _setup")
# WORKAROUND: The run_reports.sh command refers to a parameter
# named --safe-pidfile which is not supported by the uwsgi binary.
# Replace it with --pidfile instead.
patch_start_command = ("sudo sed -i \"s/--safe-pidfile/--pidfile/g"
"\" %s/scripts/common_startup_functions.sh"
% self.galaxy_home)
misc.run(patch_start_command)
# Create default output dir for files
file_path = os.path.join(self.app.path_resolver.galaxy_home, "database/files")
misc.make_dir(file_path, owner='galaxy')
tmp_file_path = os.path.join(self.app.path_resolver.galaxy_home, "database/tmp")
misc.make_dir(tmp_file_path, owner='galaxy')
# Create the new reports config
params = {
'galaxy_db_port': self.app.path_resolver.psql_db_port
}
template = conf_manager.load_conf_template(conf_manager.GALAXY_REPORTS_TEMPLATE)
misc.write_template_file(template, params, self.conf_file)
attempt_chown_galaxy(self.conf_file)
def _get_sge_install_conf(app, host):
"""Fills and returns the SGE_INSTALL_TEMPLATE as a string.
Positional arguments:
app -- The cloudman app
host -- the host param string is used for admin, submit and exec hosts
"""
# Add master as an execution host
# Additional execution hosts will be added later, as they start
sge_install_template = conf_manager.load_conf_template(conf_manager.SGE_INSTALL_TEMPLATE)
sge_params = {
"cluster_name": "GalaxyEC2",
"admin_host_list": host,
"submit_host_list": host,
"exec_host_list": host,
"hostname_resolving": "true",
}
for key, value in app.config.iteritems():
if key.startswith("sge_"):
key = key[len("sge_"):]
sge_params[key] = value
return sge_install_template.substitute(sge_params)
def _configure_sge(self):
log.info("Setting up SGE...")
SGE_config_file = '%s/galaxyEC2.conf' % self.app.path_resolver.sge_root
with open(SGE_config_file, 'w') as f:
print >> f, _get_sge_install_conf(
self.app, self.app.cloud_interface.get_private_ip())
os.chown(SGE_config_file,
pwd.getpwnam("sgeadmin")[2],
grp.getgrnam("sgeadmin")[2])
log.debug("Created SGE install template as file '%s'" %
SGE_config_file)
fix_libc()
log.debug("Setting up SGE.")
self._fix_util_arch()
if misc.run(
'cd %s; ./inst_sge -m -x -auto %s' %
(self.app.path_resolver.sge_root, SGE_config_file),
"Setting up SGE did not go smoothly",
"Successfully set up SGE"):
log.debug("Successfully setup SGE; configuring SGE")
log.debug("Adding parallel environments")
pes = ['SGE_SMP_PE', 'SGE_MPI_PE']
for pe in pes:
pe_file_path = os.path.join('/tmp', pe)
with open(pe_file_path, 'w') as f:
print >> f, conf_manager.load_conf_template(
getattr(conf_manager, pe)).safe_substitute()
misc.run('cd %s; ./bin/lx24-amd64/qconf -Ap %s' %
(self.app.path_resolver.sge_root, pe_file_path))
log.debug("Creating queue 'all.q'")
SGE_allq_file = '%s/all.q.conf' % self.app.path_resolver.sge_root
all_q_template = conf_manager.load_conf_template(
conf_manager.SGE_ALL_Q_TEMPLATE)
if self.app.config.hadoop_enabled:
all_q_params = {
"slots":
int(commands.getoutput("nproc")),
"prolog_path":
os.path.join(
paths.P_HADOOP_HOME,
paths.P_HADOOP_INTEGRATION_FOLDER + "/hdfsstart.sh"),
"epilog_path":
os.path.join(
paths.P_HADOOP_HOME,
paths.P_HADOOP_INTEGRATION_FOLDER + "/hdfsstop.sh")
}
else:
all_q_params = {
"slots": int(commands.getoutput("nproc")),
"prolog_path": 'NONE',
"epilog_path": 'NONE'
}
with open(SGE_allq_file, 'w') as f:
print >> f, all_q_template.substitute(all_q_params)
os.chown(SGE_allq_file,
pwd.getpwnam("sgeadmin")[2],
grp.getgrnam("sgeadmin")[2])
log.debug("Created SGE all.q template as file '%s'" %
SGE_allq_file)
misc.run(
'cd %s; ./bin/lx24-amd64/qconf -Mq %s' %
(self.app.path_resolver.sge_root, SGE_allq_file),
"Error modifying all.q", "Successfully modified all.q")
log.debug("Configuring users' SGE profiles")
misc.append_to_file(
paths.LOGIN_SHELL_SCRIPT,
"\nexport SGE_ROOT=%s" % self.app.path_resolver.sge_root)
misc.append_to_file(paths.LOGIN_SHELL_SCRIPT,
"\n. $SGE_ROOT/default/common/settings.sh\n")
# Write out the .sge_request file for individual users
sge_request_template = conf_manager.load_conf_template(
conf_manager.SGE_REQUEST_TEMPLATE)
sge_request_params = {
'psql_home': self.app.path_resolver.pg_home,
'galaxy_tools_dir': self.app.path_resolver.galaxy_tools,
}
users = ['galaxy', 'ubuntu']
for user in users:
sge_request_file = os.path.join('/home', user, '.sge_request')
with open(sge_request_file, 'w') as f:
print >> f, sge_request_template.substitute(
sge_request_params)
os.chown(sge_request_file,
pwd.getpwnam(user)[2],
grp.getgrnam(user)[2])
return True
return False
def _add_instance_as_exec_host(self, inst_alias, inst_private_ip):
"""
Add instance with ``inst_alias`` and ``inst_private_ip`` to the SGE
execution host list.
``inst_alias`` is used only in log statements while the the ``inst_private_ip``
is the IP address (or hostname) of the given instance, which must be
visible (i.e., accessible) to the other nodes in the clusters.
"""
ok = True
# Check if host is already in the exec host list
cmd = "export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; %s/bin/lx24-amd64/qconf -sel" \
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root)
proc = subprocess.Popen(
cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout, stderr = proc.communicate()
if inst_private_ip in stdout:
log.debug(
"Instance '%s' already in SGE execution host list" % inst_alias)
else:
log.debug(
"Adding instance '%s' to SGE execution host list." % inst_alias)
# Create a dir to hold all of workers host configuration files
host_conf_dir = "%s/host_confs" % self.app.path_resolver.sge_root
if not os.path.exists(host_conf_dir):
subprocess.call('mkdir -p %s' % host_conf_dir, shell=True)
os.chown(host_conf_dir, pwd.getpwnam(
"sgeadmin")[2], grp.getgrnam("sgeadmin")[2])
host_conf_file = os.path.join(host_conf_dir, str(inst_alias))
with open(host_conf_file, 'w') as f:
print >> f, conf_manager.load_conf_template(conf_manager.SGE_HOST_CONF_TEMPLATE).substitute({'hostname': inst_private_ip})
os.chown(host_conf_file, pwd.getpwnam("sgeadmin")[
2], grp.getgrnam("sgeadmin")[2])
log.debug(
"Created SGE host configuration template as file '%s'." % host_conf_file)
# Add worker instance as execution host to SGE
cmd = 'export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; %s/bin/lx24-amd64/qconf -Ae %s' \
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root, host_conf_file)
log.debug("Add SGE exec host cmd: {0}".format(cmd))
proc = subprocess.Popen(cmd, shell=True,
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
if proc.wait() == 0:
log.debug("Successfully added instance '%s' w/ private IP '%s' as an execution host."
% (inst_alias, inst_private_ip))
else:
ok = False
log.error("Process encountered problems adding instance '%s' as an SGE execution host. "
"Process returned code %s" % (inst_alias, proc.returncode))
stderr = stdout = None
stdout, stderr = proc.communicate()
log.debug(" - adding instance '%s' SGE execution host stdout (private IP: %s): '%s'"
% (inst_alias, inst_private_ip, stdout))
log.debug(" - adding instance '%s' SGE execution host stderr (private IP: %s): '%s'"
% (inst_alias, inst_private_ip, stderr))
# == Add given instance's hostname to @allhosts
# Check if instance is already in allhosts file and do not recreate the
# file if so.
# Additional documentation: allhosts file can be generated by CloudMan
# each time an instance is added or removed. The file is generated based
# on the Instance object CloudMan keeps track of and, as a result, it
# includes all of the instances listed. So, some instances, although they
# have yet to go through the addition process, might have had their IPs
# already included in the allhosts file. This approach ensures consistency
# between SGE and CloudMan and has been working much better than trying
# to sync the two via other methods.
proc = subprocess.Popen("export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; "
"%s/bin/lx24-amd64/qconf -shgrp @allhosts"
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root), shell=True, stdout=subprocess.PIPE)
allhosts_out = proc.communicate()[0]
if inst_private_ip not in allhosts_out:
now = datetime.datetime.utcnow()
ah_file = '/tmp/ah_add_' + now.strftime("%H_%M_%S")
self._write_allhosts_file(filename=ah_file, to_add=inst_private_ip)
if not misc.run('export SGE_ROOT=%s;. $SGE_ROOT/default/common/settings.sh; '
'%s/bin/lx24-amd64/qconf -Mhgrp %s'
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root, ah_file),
"Problems updating @allhosts aimed at adding '%s'" % inst_alias,
"Successfully updated @allhosts to add '%s' with address '%s'" % (inst_alias, inst_private_ip)):
ok = False
else:
log.debug("Instance '%s' IP is already in SGE's @allhosts" % inst_alias)
# On instance reboot, SGE might have already been configured for a given
# instance and this method will fail along the way although the instance
# will still operate within SGE so don't explicitly state it was added.
if ok:
log.debug("Successfully added instance '%s' to SGE" % inst_alias)
return ok
def _configure_sge(self):
log.info("Setting up SGE...")
SGE_config_file = '%s/galaxyEC2.conf' % self.app.path_resolver.sge_root
with open(SGE_config_file, 'w') as f:
print >> f, _get_sge_install_conf(self.app, self.app.cloud_interface.get_private_ip())
os.chown(SGE_config_file, pwd.getpwnam("sgeadmin")[2],
grp.getgrnam("sgeadmin")[2])
log.debug(
"Created SGE install template as file '%s'" % SGE_config_file)
fix_libc()
log.debug("Setting up SGE.")
self._fix_util_arch()
if misc.run('cd %s; ./inst_sge -m -x -auto %s' % (self.app.path_resolver.sge_root, SGE_config_file), "Setting up SGE did not go smoothly", "Successfully set up SGE"):
log.debug("Successfully setup SGE; configuring SGE")
log.debug("Adding parallel environments")
pes = ['SGE_SMP_PE', 'SGE_MPI_PE']
for pe in pes:
pe_file_path = os.path.join('/tmp', pe)
with open(pe_file_path, 'w') as f:
print >> f, conf_manager.load_conf_template(getattr(conf_manager, pe)).safe_substitute()
misc.run('cd %s; ./bin/lx24-amd64/qconf -Ap %s' % (
self.app.path_resolver.sge_root, pe_file_path))
log.debug("Creating queue 'all.q'")
SGE_allq_file = '%s/all.q.conf' % self.app.path_resolver.sge_root
all_q_template = conf_manager.load_conf_template(conf_manager.SGE_ALL_Q_TEMPLATE)
if self.app.config.hadoop_enabled:
all_q_params = {
"slots": int(commands.getoutput("nproc")),
"prolog_path": os.path.join(paths.P_HADOOP_HOME, paths.P_HADOOP_INTEGRATION_FOLDER + "/hdfsstart.sh"),
"epilog_path": os.path.join(paths.P_HADOOP_HOME, paths.P_HADOOP_INTEGRATION_FOLDER + "/hdfsstop.sh")
}
else:
all_q_params = {
"slots": int(commands.getoutput("nproc")),
"prolog_path": 'NONE',
"epilog_path": 'NONE'
}
with open(SGE_allq_file, 'w') as f:
print >> f, all_q_template.substitute(
all_q_params)
os.chown(SGE_allq_file, pwd.getpwnam("sgeadmin")[
2], grp.getgrnam("sgeadmin")[2])
log.debug(
"Created SGE all.q template as file '%s'" % SGE_allq_file)
misc.run(
'cd %s; ./bin/lx24-amd64/qconf -Mq %s' % (
self.app.path_resolver.sge_root, SGE_allq_file),
"Error modifying all.q", "Successfully modified all.q")
log.debug("Configuring users' SGE profiles")
misc.append_to_file(paths.LOGIN_SHELL_SCRIPT,
"\nexport SGE_ROOT=%s" % self.app.path_resolver.sge_root)
misc.append_to_file(paths.LOGIN_SHELL_SCRIPT,
"\n. $SGE_ROOT/default/common/settings.sh\n")
# Write out the .sge_request file for individual users
sge_request_template = conf_manager.load_conf_template(conf_manager.SGE_REQUEST_TEMPLATE)
sge_request_params = {
'psql_home': self.app.path_resolver.pg_home,
'galaxy_tools_dir': self.app.path_resolver.galaxy_tools,
}
users = ['galaxy', 'ubuntu']
for user in users:
sge_request_file = os.path.join('/home', user, '.sge_request')
with open(sge_request_file, 'w') as f:
print >> f, sge_request_template.substitute(sge_request_params)
os.chown(sge_request_file, pwd.getpwnam(user)[2], grp.getgrnam(user)[2])
return True
return False
def _add_instance_as_exec_host(self, inst_alias, inst_private_ip):
"""
Add instance with ``inst_alias`` and ``inst_private_ip`` to the SGE
execution host list.
``inst_alias`` is used only in log statements while the the ``inst_private_ip``
is the IP address (or hostname) of the given instance, which must be
visible (i.e., accessible) to the other nodes in the clusters.
"""
ok = True
# Check if host is already in the exec host list
cmd = "export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; %s/bin/lx24-amd64/qconf -sel" \
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root)
proc = subprocess.Popen(cmd,
shell=True,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
stdout, stderr = proc.communicate()
if inst_private_ip in stdout:
log.debug("Instance '%s' already in SGE execution host list" %
inst_alias)
else:
log.debug("Adding instance '%s' to SGE execution host list." %
inst_alias)
# Create a dir to hold all of workers host configuration files
host_conf_dir = "%s/host_confs" % self.app.path_resolver.sge_root
if not os.path.exists(host_conf_dir):
subprocess.call('mkdir -p %s' % host_conf_dir, shell=True)
os.chown(host_conf_dir,
pwd.getpwnam("sgeadmin")[2],
grp.getgrnam("sgeadmin")[2])
host_conf_file = os.path.join(host_conf_dir, str(inst_alias))
with open(host_conf_file, 'w') as f:
print >> f, conf_manager.load_conf_template(
conf_manager.SGE_HOST_CONF_TEMPLATE).substitute(
{'hostname': inst_private_ip})
os.chown(host_conf_file,
pwd.getpwnam("sgeadmin")[2],
grp.getgrnam("sgeadmin")[2])
log.debug("Created SGE host configuration template as file '%s'." %
host_conf_file)
# Add worker instance as execution host to SGE
cmd = 'export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; %s/bin/lx24-amd64/qconf -Ae %s' \
% (self.app.path_resolver.sge_root, self.app.path_resolver.sge_root, host_conf_file)
log.debug("Add SGE exec host cmd: {0}".format(cmd))
proc = subprocess.Popen(cmd,
shell=True,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
if proc.wait() == 0:
log.debug(
"Successfully added instance '%s' w/ private IP '%s' as an execution host."
% (inst_alias, inst_private_ip))
else:
ok = False
log.error(
"Process encountered problems adding instance '%s' as an SGE execution host. "
"Process returned code %s" % (inst_alias, proc.returncode))
stderr = stdout = None
stdout, stderr = proc.communicate()
log.debug(
" - adding instance '%s' SGE execution host stdout (private IP: %s): '%s'"
% (inst_alias, inst_private_ip, stdout))
log.debug(
" - adding instance '%s' SGE execution host stderr (private IP: %s): '%s'"
% (inst_alias, inst_private_ip, stderr))
# == Add given instance's hostname to @allhosts
# Check if instance is already in allhosts file and do not recreate the
# file if so.
# Additional documentation: allhosts file can be generated by CloudMan
# each time an instance is added or removed. The file is generated based
# on the Instance object CloudMan keeps track of and, as a result, it
# includes all of the instances listed. So, some instances, although they
# have yet to go through the addition process, might have had their IPs
# already included in the allhosts file. This approach ensures consistency
# between SGE and CloudMan and has been working much better than trying
# to sync the two via other methods.
proc = subprocess.Popen(
"export SGE_ROOT=%s; . $SGE_ROOT/default/common/settings.sh; "
"%s/bin/lx24-amd64/qconf -shgrp @allhosts" %
(self.app.path_resolver.sge_root, self.app.path_resolver.sge_root),
shell=True,
stdout=subprocess.PIPE)
allhosts_out = proc.communicate()[0]
if inst_private_ip not in allhosts_out:
now = datetime.datetime.utcnow()
ah_file = '/tmp/ah_add_' + now.strftime("%H_%M_%S")
self._write_allhosts_file(filename=ah_file, to_add=inst_private_ip)
if not misc.run(
'export SGE_ROOT=%s;. $SGE_ROOT/default/common/settings.sh; '
'%s/bin/lx24-amd64/qconf -Mhgrp %s' %
(self.app.path_resolver.sge_root,
self.app.path_resolver.sge_root, ah_file),
"Problems updating @allhosts aimed at adding '%s'" %
inst_alias,
"Successfully updated @allhosts to add '%s' with address '%s'"
% (inst_alias, inst_private_ip)):
ok = False
else:
log.debug("Instance '%s' IP is already in SGE's @allhosts" %
inst_alias)
# On instance reboot, SGE might have already been configured for a given
# instance and this method will fail along the way although the instance
# will still operate within SGE so don't explicitly state it was added.
if ok:
log.debug("Successfully added instance '%s' to SGE" % inst_alias)
return ok
def configure_proftpd(self):
"""
Configure environment for running ProFTPd service.
"""
log.debug("Configuring ProFTPd")
# Because we're rewriting the proftpd config file below, update the
# password for PostgreSQL galaxyftp user
gftp_pwd = self.app.path_resolver.proftpd_galaxyftp_user_pwd
log.debug(
"Setting psql password for galaxyftp role to {0}".format(gftp_pwd))
# Check if galaxtftp role already exists
cmd = (
'{0} - postgres -c"{1} -p {2} -tAc\\\"SELECT 1 FROM pg_roles WHERE rolname=\'galaxyftp\'\\\""'
.format(paths.P_SU, self.app.path_resolver.psql_cmd,
self.app.path_resolver.psql_db_port))
role = 'ALTER' if misc.getoutput(cmd) == '1' else 'CREATE'
# Now either CREATE or ALTER the galaxyftp role to set the permissions
cmd = (
'{0} - postgres -c"{1} -p {4} -c\\\"{2} ROLE galaxyftp LOGIN PASSWORD \'{3}\'\\\""'
.format(paths.P_SU, self.app.path_resolver.psql_cmd, role,
gftp_pwd, self.app.path_resolver.psql_db_port))
misc.run(cmd)
# Update the config to match the current environment
proftpd_tmplt = conf_manager.PROFTPD_CONF_TEMPLATE
proftpd_conf_template = conf_manager.load_conf_template(proftpd_tmplt)
params = {
'galaxy_user_name': paths.GALAXY_USER_NAME,
'galaxyftp_user_name': 'galaxyftp',
'psql_galaxyftp_password': gftp_pwd,
'galaxy_db_port': self.app.path_resolver.psql_db_port,
'galaxyFS_base_path': self.app.path_resolver.galaxy_data,
'public_ip_address': self.app.cloud_interface.get_public_ip()
}
template = proftpd_conf_template.substitute(params)
# Write out the config file
with open(self.app.path_resolver.proftpd_conf_file, 'w') as f:
print >> f, template
log.debug("Updated ProFTPd conf file {0}".format(
self.app.path_resolver.proftpd_conf_file))
# Place the FTP welcome message file
urllib.urlretrieve(
"https://s3.amazonaws.com/cloudman/files/proftpd_welcome.txt",
"/usr/proftpd/etc/welcome_msg.txt")
# Setup the Galaxy data dir for FTP
ftp_data_dir = '%s/tmp/ftp' % self.app.path_resolver.galaxy_data
if not os.path.exists(ftp_data_dir):
os.makedirs(ftp_data_dir)
attempt_chown_galaxy(ftp_data_dir)
# Some images have vsFTPd server included so stop it first
vsFTPd_exists = misc.run('status vsftpd', quiet=True)
if vsFTPd_exists and 'start/running' in vsFTPd_exists:
log.debug("Stopping vsFTPd")
misc.run('stop vsftpd')
# Start the server now
if misc.run('/etc/init.d/proftpd start'):
self.state = service_states.RUNNING
return True
else:
log.error("Trouble starting ProFTPd")
self.state = service_states.ERROR
return False
