def cancel_unmigrated_requests(tenants):
"""
There are two cases to handle here. Either there are no idle jobs, so
all requests should be cancelled.
Or there are idle jobs but the existing requests could not be migrated
to them. In this case, any orphaned requests should also be cancelled.
"""
for tenant in tenants:
# start by grabbing all of the open spot requests for this tenant
ids_to_check = ProvisionerConfig().simulator.get_open_requests()
# Get the set of idle job numbers
idle_job_numbers = []
for job in tenant.jobs:
if job.sim_status == 'IDLE':
idle_job_numbers.append(job.id)
# now get all of the orphaned requests
reqs = get_orphaned_requests(tenant, ids_to_check, idle_job_numbers)
reqs_to_cancel = []
# build a nice list we can pass to boto
for req in reqs:
reqs_to_cancel.append(req['request_id'])
# now cancel all of these requests
try:
if len(reqs_to_cancel) > 0:
logger.debug("Cancelling unmigrated requests: %s" %
reqs_to_cancel)
ProvisionerConfig().simulator.cancel_spot_instance_requests(
ids_to_check)
except Exception as e:
logger.exception("Error removing spot instance requests.")
raise e
def update_database(tenants):
"""
Record when an instance is started in the database. This should also
try and record when an instance is terminated.
In future work this should probably calculate the cost of the instance
as well.
"""
for tenant in tenants:
conn = boto.connect_ec2(tenant.access_key, tenant.secret_key)
try:
# First get all operating instances (instances probably are not
# yet tagged, so don't filter them yet.)
reservations = conn.get_all_instances()
instance_spot_ids = []
# Go over the fulfilled spot requests
for r in reservations:
for i in r.instances:
if i.spot_instance_request_id is not None:
instance_spot_ids.append("'%s'" %
i.spot_instance_request_id)
# Also include ondemand instances which tag as the id.
else:
instance_spot_ids.append("'%s'" % i.id)
# Get the entry in the instance_request table for each of these
# requests
check_for_new_instances(reservations, instance_spot_ids, conn,
tenant)
check_for_terminated_instances(reservations)
except:
logger.exception("Error updating database. Or, more likely, the " +
"instance wasn't yet registered by amazon, so " +
"skip this error this time.")
def launch_spot_request(conn, request, tenant, job):
try:
cost_aware_req = job.cost_aware
drafts_req = job.DrAFTS
drafts_avg = job.DrAFTSAvg
cost_aware_req = job.cost_aware
drafts_req = job.cost_aware
drafts_avg = job.cost_aware
mapping = None
my_req_ids = ProvisionerConfig().simulator.request_spot_instances(
price=request.bid, image_id=request.ami,
subnet_id=tenant.subnets[request.zone],
count=request.count,
key_name=tenant.key_pair,
security_group_ids=[tenant.security_group],
instance_type=request.instance_type,
user_data=customise_cloudinit(tenant, job),
block_device_map=mapping,
job=job)
for req in my_req_ids:
# tag each request
tag_requests(req, tenant.name, conn)
ProvisionerConfig().dbconn.execute(
("insert into instance_request (tenant, instance_type, " +
"price, job_runner_id, request_type, request_id, " +
"subnet, cost_aware_ins, cost_aware_bid, " +
"cost_aware_subnet, " +
"drafts_ins, drafts_bid, drafts_subnet, selected_avg_price, "
"cost_aware_avg_price, drafts_avg_price, drafts_avg_ins, " +
"drafts_avg_bid, drafts_avg_subnet, drafts_avg_avg_price) " +
"values ('%s', '%s', %s, '%s', '%s', '%s', %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)") %
(tenant.db_id, request.instance.db_id, request.OraclePrice,
job.id,
"spot", req, tenant.subnets_db_id[request.zone],
cost_aware_req.instance.db_id, cost_aware_req.bid,
tenant.subnets_db_id[cost_aware_req.zone],
drafts_req.instance.db_id,
drafts_req.DrAFTS, tenant.subnets_db_id[drafts_req.zone],
request.AvgPrice, cost_aware_req.AvgPrice,
drafts_req.AvgPrice,
drafts_avg.instance.db_id, drafts_avg.DrAFTS,
tenant.subnets_db_id[
drafts_avg.zone],
drafts_avg.AvgPrice))
return my_req_ids
except boto.exception.EC2ResponseError:
logger.exception("There was an error communicating with EC2.")
def get_orphaned_requests(tenant, ids_to_check, idle_job_numbers):
"""
Check if there are any requests that don't belong to a job in the idle
queue
"""
res = []
if len(ids_to_check) > 0:
try:
# Add quotes and commas to the list items for psql.
sir_ids = (', '.join('\'' + item + '\'' for item in ids_to_check))
# Get any requests that do not belong to an idle job
rows = []
print sir_ids
print idle_job_numbers
if len(idle_job_numbers) > 0:
rows = ProvisionerConfig().dbconn.execute(
("select instance_request.id, instance_type.type, " +
"instance_request.job_runner_id, " +
"instance_request.request_id from instance_request, " +
"instance_type where instance_request.instance_type = " +
"instance_type.id and job_runner_id not in (%s) and " +
"request_id in (%s) and request_type = 'spot' and " +
"tenant = %s") % (",".join(
'\'' + num + '\''
for num in idle_job_numbers), sir_ids, tenant.db_id))
else:
rows = ProvisionerConfig().dbconn.execute(
("select instance_request.id, instance_type.type, " +
"instance_request.job_runner_id, " +
"instance_request.request_id from instance_request, " +
"instance_type where instance_request.instance_type = " +
"instance_type.id and " +
"request_id in (%s) and request_type = 'spot' and " +
"tenant = %s") % (sir_ids, tenant.db_id))
# I had some issues with the rows object closing after
# returning it, so this just builds a dict for it
for row in rows:
res.append({
'id': row['id'],
'type': row['type'],
'job_runner_id': row['job_runner_id'],
'request_id': row['request_id']
})
logger.warn("Orphaned request %s" % row['request_id'])
except psycopg2.Error:
logger.exception("Error migrating instances.")
return res
def get_instance_types():
"""
Get the set of instances from the database
"""
instances = []
try:
rows = self.dbconn.execute(
"select * from instance_type where available = True")
for row in rows:
instances.append(aws.Instance(
row['id'], row['type'], row['ondemand_price'],
row['cpus'], row['memory'], row['disk'],
row['ami']))
except psycopg2.Error:
logger.exception("Error getting instance types from database.")
return instances
def launch_ondemand_request(conn, request, tenant, job):
try:
mapping = BlockDeviceMapping()
sda1 = BlockDeviceType()
eph0 = BlockDeviceType()
eph1 = BlockDeviceType()
eph2 = BlockDeviceType()
eph3 = BlockDeviceType()
sda1.size = 10
eph0.ephemeral_name = 'ephemeral0'
eph1.ephemeral_name = 'ephemeral1'
eph2.ephemeral_name = 'ephemeral2'
eph3.ephemeral_name = 'ephemeral3'
mapping['/dev/sda1'] = sda1
mapping['/dev/sdb'] = eph0
mapping['/dev/sdc'] = eph1
mapping['/dev/sdd'] = eph2
mapping['/dev/sde'] = eph3
# issue a run_instances command for this request
res = conn.run_instances(min_count=request.count,
max_count=request.count,
key_name=tenant.key_pair,
image_id=request.ami,
security_group_ids=[tenant.security_group],
user_data=customise_cloudinit(tenant, job),
instance_type=request.instance_type,
subnet_id=tenant.subnet,
block_device_map=mapping)
my_req_ids = [req.id for req in res.instances]
# address = ""
for req in my_req_ids:
# tag each request
tag_requests(req, tenant.name, conn)
# update the database to include the new request
ProvisionerConfig().dbconn.execute(
("insert into instance_request (tenant, instance_type, " +
"price, job_runner_id, request_type, request_id, " +
"subnet) values ('%s', '%s', %s, %s, '%s', '%s', %s)") %
(tenant.db_id, request.instance.db_id,
request.instance.ondemand, job.id, "ondemand", req,
tenant.subnet_id))
# ProvisionerConfig().dbconn.commit()
return
except boto.exception.EC2ResponseError:
logger.exception("There was an error communicating with EC2.")
def tag_requests(req, tag, conn):
"""
Tag any requests that have just been made with the tenant name
"""
for x in range(0, 3):
try:
conn.create_tags([req], {"tenant": tag})
conn.create_tags([req], {"Name": 'worker@%s' % tag})
break
except boto.exception.BotoClientError:
time.sleep(2)
pass
except boto.exception.BotoServerError:
time.sleep(2)
pass
except boto.exception.EC2ResponseError:
logger.exception("There was an error communicating with EC2.")
def cancel_unmigrated_requests(tenants):
"""
There are two cases to handle here. Either there are no idle jobs, so
all requests should be cancelled.
Or there are idle jobs but the existing requests could not be migrated
to them. In this case, any orphaned requests should also be cancelled.
"""
for tenant in tenants:
# start by grabbing all of the open spot requests for this tenant
conn = boto.connect_ec2(tenant.access_key, tenant.secret_key)
reqs = conn.get_all_spot_instance_requests(filters={
"tag-value": tenant.name,
"state": "open"
})
# Get a list of ids that can be used in a db query
ids_to_check = []
for r in reqs:
ids_to_check.append("%s" % r.id)
# Get the set of idle job numbers
idle_job_numbers = []
for job in tenant.jobs:
if job.status == '1':
idle_job_numbers.append(job.id)
# now get all of the orphaned requests
reqs = get_orphaned_requests(tenant, ids_to_check, idle_job_numbers)
reqs_to_cancel = []
# build a nice list we can pass to boto
for req in reqs:
reqs_to_cancel.append(req['request_id'])
# now cancel all of these requests
try:
if len(reqs_to_cancel) > 0:
logger.debug("Cancelling unmigrated requests: %s" %
reqs_to_cancel)
conn.cancel_spot_instance_requests(ids_to_check)
except Exception as e:
logger.exception("Error removing spot instance requests.")
raise e
def update_database(tenants):
"""
Record when an instance is started in the database. This should also
try and record when an instance is terminated.
In future work this should probably calculate the cost of the instance
as well.
"""
for tenant in tenants:
try:
# First get all operating instances (instances probably are not
# yet tagged, so don't filter them yet.)
reservations = ProvisionerConfig().simulator.get_all_instances()
instance_spot_ids = ProvisionerConfig(
).simulator.get_spot_instances()
check_for_terminated_instances(reservations)
except psycopg2.Error:
logger.exception("Error updating database.")
def get_existing_requests(self, tenant, job):
# Get all of the outstanding requests from the db for this instance
existing_requests = []
try:
rows = ProvisionerConfig().dbconn.execute(
("select instance_request.instance_type, "
"instance_request.request_type, "
"instance_type.type, "
"instance_request.subnet, subnet_mapping.zone "
"from instance_request, subnet_mapping, instance_type "
"where job_runner_id = '%s' and "
"instance_request.tenant = %s and "
"instance_request.instance_type = instance_type.id and "
"subnet_mapping.id = instance_request.subnet") %
(job.id, tenant.db_id))
for row in rows:
existing_requests.append(aws.Request(
None, row['type'],
row['zone'], None, None))
except psycopg2.Error:
logger.exception("Error getting number of outstanding")
return existing_requests
def migrate_request_to_job(request, job):
"""
Check if an instance can be repurposed to another job and update the
database.
"""
# Check to see if the job can be fulfilled by the requested instance
if check_requirements(request['type'], job):
next_idle_job_id = job.id
try:
logger.debug(
("Migrating instance request %s, from job " + "%s to job %s.")
% (request['id'], request['job_runner_id'], next_idle_job_id))
ProvisionerConfig().dbconn.execute(
("update instance_request set job_runner_id = '%s' " +
"where id = %s") % (next_idle_job_id, request['id']))
ProvisionerConfig().dbconn.execute(
("insert into request_migration " +
"(request_id, from_job, to_job, migration_time) " +
"values (%s, %s, %s, NOW())") %
(request['id'], request['job_runner_id'], next_idle_job_id))
return True
except psycopg2.Error:
logger.exception("Error performing migration in database.")
def get_global_queue(self):
"""
Poll condor_q -global and return a set of Jobs.
"""
cmd = [
'condor_q', '-global', '-format', '%s:', 'GlobalJobId', '-format',
'%s:', 'ClusterId', '-format', '%s:', 'JobStatus', '-format',
'%s:', 'QDate', '-format', '%s:', 'RequestCpus', '-format', '%s:',
'RequestMemory', '-format', '%s:', 'RequestDisk', '-format', '%s',
'JobDescription', '-format', '%s\n', 'ExitStatus'
]
#output = subprocess.check_output(cmd)
output = subprocess.Popen(cmd, stdout=subprocess.PIPE).communicate()[0]
queue = output.split("\n")
queue = filter(None, queue)
jobs = []
if len(queue) > 0:
# set the time of the first job if this is it
if ProvisionerConfig().first_job_time is None:
logger.debug("Simulation: first job time set")
utc = timezone('UTC')
ProvisionerConfig().first_job_time = datetime.datetime.now(utc)
for line in queue:
if "All queues are empty" in line:
break
try:
split = line.split(":")
tenant_addr = ""
# Grab the address of the tenant from the global id
if "#" in split[0]:
tenant_addr = split[0].split("#")[0]
# Req memory is either a number or a string talking about
# requested memory, so check if it is a number
req_memory = 0
try:
req_memory = int(split[5])
if req_memory > 1024:
# change it to use GB like instance types.
req_memory = req_memory / 1024
except Exception, e:
pass
# Req disk is the same as memory. Again it is
# in mb I believe
req_disk = 0
try:
req_disk = int(split[6])
if req_disk > 1024:
# change it to use GB like instance types.
req_disk = req_disk / 1024
except Exception, e:
pass
# Decipher the description of the job as well (name, etc.)
description = {}
if "=" in split[7]:
description = self.process_job_description(split[7])
# Create the job: tenant address, job id, queue time,
# requested cpus, requested memory
j = Job(tenant_addr, split[1], split[2], split[3],
split[4], req_memory, req_disk, description)
jobs.append(j)
except Exception, e:
logger.exception("Something has gone wrong while"
" processing "
"the job queue.")
raise e
def __init__(self, *args, **kwargs):
"""
Load provisioner configuration based on the settings in a config file.
"""
# override defaults with kwargs
config_file = 'ggprovisioner/provisioner.ini'
cloudinit_file = "cloudinit.cfg"
if 'config_file' in kwargs:
config_file = kwargs['config_file']
if 'cloudinit_file' in kwargs:
cloudinit_file = kwargs['cloudinit_file']
# we need to pull cloudinit from the DB in the future
self.cloudinit_file = cloudinit_file
# read config from a file
config = ConfigParser.ConfigParser()
config.read(config_file)
# get DB connection info
user = config.get('Database', 'user')
password = config.get('Database', 'password')
host = config.get('Database', 'host')
port = config.get('Database', 'port')
database = config.get('Database', 'database')
# create a connection and keep it as a config attribute
try:
engine = sqlalchemy.create_engine(
'postgresql://%s:%s@%s:%s/%s' %
(user, password, host, port, database),
isolation_level="AUTOCOMMIT")
self.dbconn = engine.connect()
except psycopg2.Error:
logger.exception("Failed to connect to database.")
# Get some provisioner specific config settings
self.ondemand_price_threshold = float(
config.get('Provision', 'ondemand_price_threshold'))
self.max_requests = int(config.get('Provision', 'max_requests'))
self.run_rate = int(config.get('Provision', 'run_rate'))
self.DrAFTS = config.get('Provision', 'DrAFTS')
self.DrAFTSAvgPrice = config.get('Provision', 'DrAFTSAvgPrice')
self.DrAFTSProfiles = config.get('Provision', 'DrAFTSProfiles')
self.instance_types = []
if self.DrAFTS == 'True':
self.DrAFTS = True
else:
self.DrAFTS = False
if self.DrAFTSProfiles == 'True':
self.DrAFTSProfiles = True
else:
self.DrAFTSProfiles = False
if self.DrAFTSAvgPrice == 'True':
self.DrAFTSAvgPrice = True
else:
self.DrAFTSAvgPrice = False
self.simulate = config.get('Simulation', 'Simulate')
if self.simulate == 'True':
self.simulate = True
else:
self.simulate = False
self.neg_time = int(config.get('Simulation', 'NegTime'))
self.job_number = int(config.get('Simulation', 'JobNumber'))
self.idle_time = int(config.get('Simulation', 'IdleTime'))
self.terminate = config.get('Simulation', 'Terminate')
self.overhead_time = int(config.get('Simulation', 'OverheadTime'))
self.simulate_jobs = (config.get('Simulation', 'JobFile'))
self.run_name = config.get('Simulation', 'RunName')
# things for the simulator
self.first_job_time = None
# drafts_stored_db = True
# self.simulate = True
self.sim_time = datetime.datetime.strptime('2017-03-25T03:14:00Z',
'%Y-%m-%dT%H:%M:%SZ')
self.jobs_file = self.simulate_jobs.split("/")[-1]
if self.job_number == 100:
# 100 jobs
self.jobs_file = '100jobs.json'
self.simulate_jobs = '/home/ubuntu/100jobs.json'
self.sim_time = datetime.datetime.strptime('2017-03-23T21:35:00Z',
'%Y-%m-%dT%H:%M:%SZ')
elif self.job_number == 200:
# 200 jobs
self.jobs_file = '200jobs.json'
self.simulate_jobs = '/home/ubuntu/200jobs.json'
self.sim_time = datetime.datetime.strptime('2017-03-27T20:45:00Z',
'%Y-%m-%dT%H:%M:%SZ')
elif self.job_number == 500:
# 500 jobs
self.jobs_file = '500jobs.json'
self.simulate_jobs = '/home/ubuntu/500jobs.json'
self.sim_time = datetime.datetime.strptime('2017-03-28T21:55:00Z',
'%Y-%m-%dT%H:%M:%SZ')
elif self.job_number == 1000:
# 1000 jobs
self.jobs_file = '1000jobs.json'
self.simulate_jobs = '/home/ubuntu/1000jobs.json'
self.sim_time = datetime.datetime.strptime('2017-03-25T03:14:00Z',
'%Y-%m-%dT%H:%M:%SZ')
if self.idle_time == 120:
self.idle_time = 128
self.sim_time = self.sim_time.replace(tzinfo=pytz.utc)
self.run_id = random.randint(500, 10000)
self.relative_time = None
self.simulate_time = self.sim_time
if self.simulate:
self.simulator = ggprovisioner.cloud.simaws.aws_simulator.AWSSimulator()
def load_from_db():
"""
Load all of the tenant data. This should let us iterate over the
tenants and apply their preferences to the idle queue etc..
It should also let us shut down unnecessary requests across the board
as this will load their aws credentials.
"""
tenant_list = []
# Pull all of the tenant data from the database
try:
# Only get those that are subscribed
tenant_columns = [
'id', 'name', 'public_address', 'condor_address', 'public_ip',
'zone', 'vpc', 'security_group', 'domain'
]
tenant_settings_columns = [
'max_bid_price', 'bid_percent', 'timeout_threshold'
]
tenant_selection = ', '.join('tenant.' + c for c in tenant_columns)
tenant_settings_selection = ', '.join('tenant_settings.' + c
for c in tenant_settings_columns)
aws_creds_columns = ['access_key_id', 'secret_key', 'key_pair']
aws_creds_selection = ', '.join('aws_credentials.' + c
for c in aws_creds_columns)
subnet_columns = ['subnet', 'id as subnet_id']
subnet_selection = ', '.join('subnet_mapping.' + c
for c in subnet_columns)
full_selection = ', '.join(
(tenant_selection, tenant_settings_selection, aws_creds_selection,
subnet_selection))
rows = ProvisionerConfig().dbconn.execute(
"SELECT " + full_selection +
" FROM tenant, tenant_settings, aws_credentials, subnet_mapping"
" WHERE tenant_settings.tenant = tenant.id AND"
" tenant.credentials = aws_credentials.id AND"
" tenant.subscribed = TRUE AND subnet_mapping.tenant"
" = tenant.id AND subnet_mapping.zone = tenant.zone")
# Create a tenant object for each row returned
for row in rows:
t = Tenant(row['id'], row['name'], row['public_address'],
row['condor_address'], row['public_ip'], row['zone'],
row['subnet'], row['subnet_id'], row['vpc'],
row['security_group'], row['domain'],
row['max_bid_price'], row['bid_percent'],
row['timeout_threshold'], row['access_key_id'],
row['secret_key'], row['key_pair'])
# Pull the subnets for the tenant too
subnets = {}
subnets_db_id = {}
subs = ProvisionerConfig().dbconn.execute("select * "
"from subnet_mapping "
"where tenant = %s" %
t.db_id)
# Create a dict for the subnets and add that to the tenant
# I later realised that I need the database id of the subnet to
# store the instance request in the database:
# Hello, subnets_db_id.
for sn in subs:
subnets.update({sn['zone']: sn['subnet']})
subnets_db_id.update({sn['zone']: sn['id']})
t.subnets = subnets
t.subnets_db_id = subnets_db_id
tenant_list.append(t)
except psycopg2.Error:
logger.exception("Failed to get tenant data.")
return tenant_list
def launch_spot_request(conn, request, tenant, job):
try:
logger.debug("%s = %s. tenants vpc = %s" %
(request.zone, tenant.subnets[request.zone], tenant.vpc))
cost_aware_req = job.cost_aware
drafts_req = job.cost_aware
drafts_avg = job.cost_aware
mapping = BlockDeviceMapping()
sda1 = BlockDeviceType()
eph0 = BlockDeviceType()
eph1 = BlockDeviceType()
eph2 = BlockDeviceType()
eph3 = BlockDeviceType()
sda1.size = 10
eph0.ephemeral_name = 'ephemeral0'
eph1.ephemeral_name = 'ephemeral1'
eph2.ephemeral_name = 'ephemeral2'
eph3.ephemeral_name = 'ephemeral3'
mapping['/dev/sda1'] = sda1
mapping['/dev/sdb'] = eph0
mapping['/dev/sdc'] = eph1
mapping['/dev/sdd'] = eph2
mapping['/dev/sde'] = eph3
inst_req = None
inst_req = conn.request_spot_instances(
price=request.bid,
image_id=request.ami,
subnet_id=tenant.subnets[request.zone],
count=request.count,
key_name=tenant.key_pair,
security_group_ids=[tenant.security_group],
instance_type=request.instance_type,
user_data=customise_cloudinit(tenant, job),
block_device_map=mapping)
my_req_ids = [req.id for req in inst_req]
# address = ""
for req in my_req_ids:
insert_launch_stats(req, request, tenant)
# tag each request
tag_requests(req, tenant.name, conn)
ProvisionerConfig().dbconn.execute((
"insert into instance_request (tenant, instance_type, " +
"price, job_runner_id, request_type, request_id, " +
"subnet, cost_aware_ins, cost_aware_bid, cost_aware_subnet," +
" drafts_ins, drafts_bid, drafts_subnet, selected_avg_price,"
" cost_aware_avg_price, drafts_avg_price, drafts_avg_ins, " +
"drafts_avg_bid, drafts_avg_subnet, drafts_avg_avg_price) " +
"values ('%s', '%s', %s, %s, '%s', '%s', %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)"
) % (tenant.db_id, request.instance.db_id, request.price, job.id,
"spot", req, tenant.subnets_db_id[request.zone],
cost_aware_req.instance.db_id, cost_aware_req.bid,
tenant.subnets_db_id[cost_aware_req.zone],
drafts_req.instance.db_id, drafts_req.DrAFTS,
tenant.subnets_db_id[drafts_req.zone], request.AvgPrice,
cost_aware_req.AvgPrice, drafts_req.AvgPrice,
drafts_avg.instance.db_id, drafts_avg.DrAFTS,
tenant.subnets_db_id[drafts_avg.zone], drafts_avg.AvgPrice))
return my_req_ids
except boto.exception.EC2ResponseError:
logger.exception("There was an error communicating with EC2.")
def stop_over_requesting(tenants):
"""
Stop too many requests being made for an individual job. This is the
frequency of new requests being made for an individual job.
Future work would be to look at launching many requests instantly, and
then cancelling requests once one is fulfilled.
"""
for tenant in tenants:
# Stop excess instances being requested in a five minute round
logger.debug("Tenant: %s. Request rate: %s" %
(tenant.name, tenant.request_rate))
if ProvisionerConfig().simulate:
open_reqs = ProvisionerConfig().simulator.requests
for job in list(tenant.idle_jobs):
if ProvisionerConfig().simulate:
if job.sim_status != "IDLE":
continue
# check to see if we are requesting too frequently
count = 0
try:
if ProvisionerConfig().simulate:
open_reqs = ProvisionerConfig().simulator.requests
for openreq in open_reqs:
if openreq.job_runner_id == job.id:
# found an existing job
diff = (ProvisionerConfig().simulate_time -
openreq.request_time).total_seconds()
if diff <= tenant.request_rate:
count = 1
else:
rows = ProvisionerConfig().dbconn.execute(
("select count(*) from instance_request " +
"where job_runner_id = '%s' and " +
"request_time >= Now() - " +
"'%s second'::interval and tenant = %s;") %
(job.id, tenant.request_rate, tenant.db_id))
for row in rows:
count = row['count']
except psycopg2.Error:
logger.exception("Error getting number of outstanding "
"requests within time frame.")
if count > 0:
tenant.idle_jobs.remove(job)
logger.debug("Too many requests. Removed job %s" % job.id)
continue
# now check to see if we already have too many requests for
# this job
try:
if ProvisionerConfig().simulate:
count = 0
open_reqs = ProvisionerConfig().simulator.requests
for openreq in open_reqs:
if openreq.job_runner_id == job.id:
count = count + 1
if count > ProvisionerConfig().max_requests:
logger.warn("Too many outstanding requests, " +
"removing idle job: %s" % repr(job))
tenant.idle_jobs.remove(job)
else:
rows = ProvisionerConfig().dbconn.execute(
("select count(*) from instance_request "
"where job_runner_id = '%s' and tenant = %s;") %
(job.id, tenant.db_id))
count = 0
for row in rows:
count = row['count']
if count > ProvisionerConfig().max_requests:
logger.warn("Too many outstanding requests, "
"removing idle job: %s" % repr(job))
tenant.idle_jobs.remove(job)
except psycopg2.Error:
logger.exception("Error getting number of outstanding "
"requests.")
