def schedule_deletions(self,
replica_list,
operation_id,
comments=''): #override
LOG.info('Ignoring deletion schedule of %d replicas (operation %d)',
len(replica_list), operation_id)
result = []
for replica, block_replicas in replica_list:
clone_replica = DatasetReplica(replica.dataset, replica.site)
clone_replica.copy(replica)
if block_replicas is None:
result.append((clone_replica, None))
else:
clone_block_replicas = []
for block_replica in block_replicas:
clone_block_replica = BlockReplica(block_replica.block,
block_replica.site,
block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clone_block_replicas.append(clone_block_replica)
result.append((clone_replica, clone_block_replicas))
return result
def schedule_copies(self, replica_list, operation_id, comments = ''): #override
sites = set(r.site for r in replica_list)
if len(sites) != 1:
raise OperationalError('schedule_copies should be called with a list of replicas at a single site.')
LOG.info('Scheduling copy of %d replicas to %s using RLFSM (operation %d)', len(replica_list), list(sites)[0], operation_id)
result = []
for replica in replica_list:
# Function spec is to return clones (so that if specific block fails to copy, we can return a dataset replica without the block)
clone_replica = DatasetReplica(replica.dataset, replica.site)
clone_replica.copy(replica)
result.append(clone_replica)
for block_replica in replica.block_replicas:
LOG.debug('Subscribing files for %s', str(block_replica))
if block_replica.file_ids is None:
LOG.debug('No file to subscribe for %s', str(block_replica))
return
all_files = block_replica.block.files
missing_files = all_files - block_replica.files()
for lfile in missing_files:
self.rlfsm.subscribe_file(block_replica.site, lfile)
clone_block_replica = BlockReplica(block_replica.block, block_replica.site, block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clone_replica.block_replicas.add(clone_block_replica)
# no external dependency - everything is a success
return result
def new_replica_from_blocks(blocks, site, group):
dataset = blocks[0].dataset
replica = DatasetReplica(dataset, site)
for block in blocks:
replica.block_replicas.add(BlockReplica(block, site, group, size = 0))
return replica
def schedule_copies(self, replica_list, operation_id, comments = ''): #override
LOG.info('Ignoring copy schedule of %d replicas (operation %d)', len(replica_list), operation_id)
result = []
for replica in replica_list:
clone_replica = DatasetReplica(replica.dataset, replica.site)
clone_replica.copy(replica)
result.append(clone_replica)
for block_replica in replica.block_replicas:
clone_block_replica = BlockReplica(block_replica.block, block_replica.site, block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clone_replica.block_replicas.add(clone_block_replica)
return result
def schedule_deletions(self,
replica_list,
operation_id,
comments=''): #override
sites = set(r.site for r, b in replica_list)
if len(sites) != 1:
raise OperationalError(
'schedule_copies should be called with a list of replicas at a single site.'
)
site = list(sites)[0]
LOG.info(
'Scheduling deletion of %d replicas from %s using RLFSM (operation %d)',
len(replica_list), site.name, operation_id)
clones = []
for dataset_replica, block_replicas in replica_list:
if block_replicas is None:
to_delete = dataset_replica.block_replicas
else:
to_delete = block_replicas
for block_replica in to_delete:
for lfile in block_replica.files():
self.rlfsm.desubscribe_file(block_replica.site, lfile)
# No external dependency -> all operations are successful
clone_replica = DatasetReplica(dataset_replica.dataset,
dataset_replica.site)
clone_replica.copy(dataset_replica)
if block_replicas is None:
clones.append((clone_replica, None))
else:
clones.append((clone_replica, []))
for block_replica in block_replicas:
clone_block_replica = BlockReplica(block_replica.block,
block_replica.site)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clones[-1][1].append(clone_block_replica)
return clones
def schedule_copies(self, replica_list, operation_id, comments = ''): #override
sites = set(r.site for r in replica_list)
if len(sites) != 1:
raise OperationalError('schedule_copies should be called with a list of replicas at a single site.')
LOG.info('Scheduling copy of %d replicas to %s using RLFSM (operation %d)', len(replica_list), list(sites)[0], operation_id)
result = []
for replica in replica_list:
# Function spec is to return clones (so that if specific block fails to copy, we can return a dataset replica without the block)
clone_replica = DatasetReplica(replica.dataset, replica.site)
clone_replica.copy(replica)
result.append(clone_replica)
for block_replica in replica.block_replicas:
LOG.debug('Subscribing files for %s', str(block_replica))
if block_replica.file_ids is None:
LOG.debug('No file to subscribe for %s', str(block_replica))
return
all_files = block_replica.block.files
missing_files = all_files - block_replica.files()
for lfile in missing_files:
self.rlfsm.subscribe_file(block_replica.site, lfile)
clone_block_replica = BlockReplica(block_replica.block, block_replica.site, block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clone_replica.block_replicas.add(clone_block_replica)
if not self._read_only:
for clone_replica in result:
if clone_replica.growing:
self.mysql.query('INSERT INTO `phedex_transfer_reservations` (`operation_id`, `item`, `site`, `group`) VALUES (%s, %s, %s, %s)', operation_id, clone_replica.dataset.name, clone_replica.site.name, clone_replica.group.name)
else:
for block_replica in clone_replica.block_replicas:
self.mysql.query('INSERT INTO `phedex_transfer_reservations` (`operation_id`, `item`, `site`, `group`) VALUES (%s, %s, %s, %s)', operation_id, block_replica.block.full_name(), clone_replica.site.name, block_replica.group.name)
# no external dependency - everything is a success
return result
def maker_deletions(block, block_entry):
replicas = []
for deletion_entry in block_entry['deletion']:
block_replica = BlockReplica(block, Site(deletion_entry['node']),
Group.null_group)
replicas.append(block_replica)
return replicas
def maker_deletions(block, block_entry, site_check=None):
replicas = []
for deletion_entry in block_entry['deletion']:
if site_check and not site_check(deletion_entry['node']):
continue
block_replica = BlockReplica(block, Site(deletion_entry['node']),
Group.null_group)
replicas.append(block_replica)
return replicas
def schedule_deletions(self, replica_list, operation_id, comments = ''): #override
LOG.info('Ignoring deletion schedule of %d replicas (operation %d)', len(replica_list), operation_id)
result = []
for replica, block_replicas in replica_list:
clone_replica = DatasetReplica(replica.dataset, replica.site)
clone_replica.copy(replica)
if block_replicas is None:
result.append((clone_replica, None))
else:
clone_block_replicas = []
for block_replica in block_replicas:
clone_block_replica = BlockReplica(block_replica.block, block_replica.site, block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clone_block_replicas.append(clone_block_replica)
result.append((clone_replica, clone_block_replicas))
return result
def schedule_deletions(self, replica_list, operation_id, comments = ''): #override
sites = set(r.site for r, b in replica_list)
if len(sites) != 1:
raise OperationalError('schedule_copies should be called with a list of replicas at a single site.')
site = list(sites)[0]
LOG.info('Scheduling deletion of %d replicas from %s using RLFSM (operation %d)', len(replica_list), site.name, operation_id)
clones = []
for dataset_replica, block_replicas in replica_list:
if block_replicas is None:
to_delete = dataset_replica.block_replicas
else:
to_delete = block_replicas
for block_replica in to_delete:
for lfile in block_replica.files():
self.rlfsm.desubscribe_file(block_replica.site, lfile)
# No external dependency -> all operations are successful
clone_replica = DatasetReplica(dataset_replica.dataset, dataset_replica.site)
clone_replica.copy(dataset_replica)
if block_replicas is None:
clones.append((clone_replica, None))
else:
clones.append((clone_replica, []))
for block_replica in block_replicas:
clone_block_replica = BlockReplica(block_replica.block, block_replica.site)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clones[-1][1].append(clone_block_replica)
return clones
def maker_blockreplicas(block, block_entry):
replicas = []
for replica_entry in block_entry['replica']:
block_replica = BlockReplica(
block,
Site(replica_entry['node']),
Group(replica_entry['group']),
is_complete=(replica_entry['bytes'] == block.size),
is_custodial=(replica_entry['custodial'] == 'y'),
size=replica_entry['bytes'],
last_update=int(replica_entry['time_update']))
replicas.append(block_replica)
return replicas
def _build_partition(self, inventory):
"""Create a mini-inventory consisting only of replicas in the partition."""
partition_repository = ObjectRepository()
partition_repository._store = inventory._store
LOG.info('Identifying target sites.')
partition = inventory.partitions[self.policy.partition_name]
partition.embed_tree(partition_repository)
# Ask each site if deletion should be triggered.
target_sites = set() # target sites of this detox cycle
tape_is_target = False
for site in inventory.sites.itervalues():
# target_site_defs are SiteConditions, which take site_partition as the argument
site_partition = site.partitions[partition]
for targdef in self.policy.target_site_def:
if targdef.match(site_partition):
target_sites.add(site)
if site.storage_type == Site.TYPE_MSS:
tape_is_target = True
break
if len(target_sites) == 0:
LOG.info('No site matches the target definition.')
return partition_repository
# Safety measure - if there are empty (no block rep) tape replicas, create block replicas with size 0 and
# add them into the partition. We will not report back to the main process though (i.e. won't call inventory.update).
if tape_is_target:
for site in filter(lambda s: s.storage_type == Site.TYPE_MSS,
target_sites):
for replica in site.dataset_replicas():
if len(replica.block_replicas) != 0:
continue
for block in replica.dataset.blocks:
block_replica = BlockReplica(block,
site,
Group.null_group,
size=0)
replica.block_replicas.add(block_replica)
block.replicas.add(block_replica)
# Add to the site partition
site.partitions[partition].replicas[replica] = None
# Create a copy of the inventory, limiting to the current partition
# We will be stripping replicas off the image as we process the policy in iterations
LOG.info('Creating a partition image.')
for group in inventory.groups.itervalues():
group.embed_into(partition_repository)
# Now clone the sites, datasets, and replicas
# Basically a copy-paste of various embed_into() functions ommitting the checks
# make a map to avoid excessive lookups
block_to_clone = {}
for site in target_sites:
site_clone = site.embed_into(partition_repository)
site_partition = site.partitions[partition]
site_partition_clone = site_partition.embed_tree(
partition_repository)
for dataset_replica, block_replica_set in site_partition.replicas.iteritems(
):
dataset = dataset_replica.dataset
try:
dataset_clone = partition_repository.datasets[dataset.name]
except KeyError:
dataset_clone = dataset.embed_into(partition_repository)
for block in dataset.blocks:
block_clone = Block(block.name,
dataset_clone,
size=block.size,
num_files=block.num_files,
is_open=block.is_open,
last_update=block.last_update,
bid=block.id)
dataset_clone.blocks.add(block_clone)
block_to_clone[block] = block_clone
if dataset_replica.group is None:
group = None
else:
group = partition_repository.groups[
dataset_replica.group.name]
replica_clone = DatasetReplica(dataset_clone,
site_clone,
growing=dataset_replica.growing,
group=group)
dataset_clone.replicas.add(replica_clone)
site_clone.add_dataset_replica(replica_clone,
add_block_replicas=False)
if block_replica_set is None:
# all block reps in partition
block_replica_set = dataset_replica.block_replicas
full_replica = True
site_partition_clone.replicas[replica_clone] = None
else:
full_replica = False
block_replica_clone_set = site_partition_clone.replicas[
replica_clone] = set()
for block_replica in block_replica_set:
block_clone = block_to_clone[block_replica.block]
if block_replica.is_complete():
size = -1
else:
size = block_replica.size
block_replica_clone = BlockReplica(
block_clone,
site_clone,
partition_repository.groups[block_replica.group.name],
is_custodial=block_replica.is_custodial,
size=size,
last_update=block_replica.last_update,
file_ids=block_replica.file_ids)
replica_clone.block_replicas.add(block_replica_clone)
block_clone.replicas.add(block_replica_clone)
if not full_replica:
block_replica_clone_set.add(block_replica_clone)
return partition_repository
def _load_replicas(self, inventory, id_group_map, id_site_map,
id_dataset_map, id_block_maps):
sql = 'SELECT dr.`dataset_id`, dr.`site_id`,'
sql += ' br.`block_id`, br.`group_id`, br.`is_complete`, br.`is_custodial`, brs.`size`, UNIX_TIMESTAMP(br.`last_update`)'
sql += ' FROM `dataset_replicas` AS dr'
sql += ' INNER JOIN `blocks` AS b ON b.`dataset_id` = dr.`dataset_id`'
sql += ' LEFT JOIN `block_replicas` AS br ON (br.`block_id`, br.`site_id`) = (b.`id`, dr.`site_id`)'
sql += ' LEFT JOIN `block_replica_sizes` AS brs ON (brs.`block_id`, brs.`site_id`) = (b.`id`, dr.`site_id`)'
if self._mysql.table_exists('groups_load_tmp'):
sql += ' INNER JOIN `groups_load_tmp` AS gt ON gt.`id` = br.`group_id`'
if self._mysql.table_exists('sites_load_tmp'):
sql += ' INNER JOIN `sites_load_tmp` AS st ON st.`id` = dr.`site_id`'
if self._mysql.table_exists('datasets_load_tmp'):
sql += ' INNER JOIN `datasets_load_tmp` AS dt ON dt.`id` = dr.`dataset_id`'
sql += ' ORDER BY dr.`dataset_id`, dr.`site_id`'
# Blocks are left joined -> there will be (# sites) x (# blocks) entries per dataset
_dataset_id = 0
_site_id = 0
dataset_replica = None
for dataset_id, site_id, block_id, group_id, is_complete, b_is_custodial, b_size, b_last_update in self._mysql.xquery(
sql):
if dataset_id != _dataset_id:
_dataset_id = dataset_id
dataset = id_dataset_map[_dataset_id]
dataset.replicas.clear()
id_block_map = id_block_maps[dataset_id]
if site_id != _site_id:
_site_id = site_id
site = id_site_map[site_id]
if dataset_replica is None or dataset is not dataset_replica.dataset or site is not dataset_replica.site:
if dataset_replica is not None:
# this is the last dataset_replica
# add to dataset and site after filling all block replicas
# this does not matter for the dataset, but for the site there is some heavy
# computation needed when a replica is added
dataset_replica.dataset.replicas.add(dataset_replica)
dataset_replica.site.add_dataset_replica(
dataset_replica, add_block_replicas=True)
dataset_replica = DatasetReplica(dataset, site)
if block_id is None:
# this block replica does not exist
continue
block = id_block_map[block_id]
group = id_group_map[group_id]
block_replica = BlockReplica(
block,
site,
group=group,
is_complete=(is_complete == 1),
is_custodial=(b_is_custodial == 1),
size=block.size if b_size is None else b_size,
last_update=b_last_update)
dataset_replica.block_replicas.add(block_replica)
block.replicas.add(block_replica)
if dataset_replica is not None:
# one last bit
dataset_replica.dataset.replicas.add(dataset_replica)
dataset_replica.site.add_dataset_replica(dataset_replica,
add_block_replicas=True)
def _build_partition(self, inventory):
"""Create a mini-inventory consisting only of replicas in the partition."""
partition_repository = ObjectRepository()
partition_repository._store = inventory._store
LOG.info('Identifying target sites.')
partition = inventory.partitions[self.policy.partition_name]
partition.embed_tree(partition_repository)
# Ask each site if deletion should be triggered.
target_sites = set() # target sites of this detox cycle
tape_is_target = False
for site in inventory.sites.itervalues():
# target_site_defs are SiteConditions, which take site_partition as the argument
site_partition = site.partitions[partition]
for targdef in self.policy.target_site_def:
if targdef.match(site_partition):
target_sites.add(site)
if site.storage_type == Site.TYPE_MSS:
tape_is_target = True
break
if len(target_sites) == 0:
LOG.info('No site matches the target definition.')
return partition_repository
# Safety measure - if there are empty (no block rep) tape replicas, create block replicas with size 0 and
# add them into the partition. We will not report back to the main process though (i.e. won't call inventory.update).
if tape_is_target:
for site in filter(lambda s: s.storage_type == Site.TYPE_MSS, target_sites):
for replica in site.dataset_replicas():
if len(replica.block_replicas) != 0:
continue
for block in replica.dataset.blocks:
block_replica = BlockReplica(block, site, Group.null_group, size = 0)
replica.block_replicas.add(block_replica)
block.replicas.add(block_replica)
# Add to the site partition
site.partitions[partition].replicas[replica] = None
# Create a copy of the inventory, limiting to the current partition
# We will be stripping replicas off the image as we process the policy in iterations
LOG.info('Creating a partition image.')
for group in inventory.groups.itervalues():
group.embed_into(partition_repository)
# Now clone the sites, datasets, and replicas
# Basically a copy-paste of various embed_into() functions ommitting the checks
# make a map to avoid excessive lookups
block_to_clone = {}
for site in target_sites:
site_clone = site.embed_into(partition_repository)
site_partition = site.partitions[partition]
site_partition_clone = site_partition.embed_tree(partition_repository)
for dataset_replica, block_replica_set in site_partition.replicas.iteritems():
dataset = dataset_replica.dataset
try:
dataset_clone = partition_repository.datasets[dataset.name]
except KeyError:
dataset_clone = dataset.embed_into(partition_repository)
for block in dataset.blocks:
block_clone = Block(
block.name,
dataset_clone,
size = block.size,
num_files = block.num_files,
is_open = block.is_open,
last_update = block.last_update,
bid = block.id
)
dataset_clone.blocks.add(block_clone)
block_to_clone[block] = block_clone
if dataset_replica.group is None:
group = None
else:
group = partition_repository.groups[dataset_replica.group.name]
replica_clone = DatasetReplica(
dataset_clone,
site_clone,
growing = dataset_replica.growing,
group = group
)
dataset_clone.replicas.add(replica_clone)
site_clone.add_dataset_replica(replica_clone, add_block_replicas = False)
if block_replica_set is None:
# all block reps in partition
block_replica_set = dataset_replica.block_replicas
full_replica = True
site_partition_clone.replicas[replica_clone] = None
else:
full_replica = False
block_replica_clone_set = site_partition_clone.replicas[replica_clone] = set()
for block_replica in block_replica_set:
block_clone = block_to_clone[block_replica.block]
if block_replica.is_complete():
size = -1
else:
size = block_replica.size
block_replica_clone = BlockReplica(
block_clone,
site_clone,
partition_repository.groups[block_replica.group.name],
is_custodial = block_replica.is_custodial,
size = size,
last_update = block_replica.last_update,
file_ids = block_replica.file_ids
)
replica_clone.block_replicas.add(block_replica_clone)
block_clone.replicas.add(block_replica_clone)
if not full_replica:
block_replica_clone_set.add(block_replica_clone)
return partition_repository
def _run_subscription_request(self, operation_id, site, group, level,
subscription_list, comments):
# Make a subscription request for potentitally multiple datasets or blocks but to one site and one group
full_catalog = collections.defaultdict(list)
if level == 'dataset':
for dataset in subscription_list:
full_catalog[dataset] = []
elif level == 'block':
for block in subscription_list:
full_catalog[block.dataset].append(block)
history_sql = 'INSERT INTO `phedex_requests` (`id`, `operation_type`, `operation_id`, `approved`) VALUES (%s, \'copy\', %s, %s)'
success = []
# make requests in chunks
request_catalog = {}
chunk_size = 0
items = []
while len(full_catalog) != 0:
dataset, blocks = full_catalog.popitem()
request_catalog[dataset] = blocks
if level == 'dataset':
chunk_size += dataset.size
items.append(dataset)
elif level == 'block':
chunk_size += sum(b.size for b in blocks)
items.extend(blocks)
if chunk_size < self.subscription_chunk_size and len(
full_catalog) != 0:
continue
options = {
'node': site.name,
'data': self._phedex.form_catalog_xml(request_catalog),
'level': level,
'priority': 'low',
'move': 'n',
'static': 'n',
'custodial': 'n',
'group': group.name,
'request_only': 'n',
'no_mail': 'n',
'comments': comments
}
try:
if self._read_only:
result = [{'id': 0}]
else:
result = self._phedex.make_request('subscribe',
options,
method=POST)
except:
LOG.error('Copy %s failed.', str(options))
# we should probably do something here
else:
request_id = int(result[0]['id']) # return value is a string
LOG.warning('PhEDEx subscription request id: %d', request_id)
if not self._read_only:
self._history.db.query(history_sql, request_id,
operation_id, True)
for dataset, blocks in request_catalog.iteritems():
if level == 'dataset':
replica = DatasetReplica(dataset,
site,
growing=True,
group=group)
for block in dataset.blocks:
replica.block_replicas.add(
BlockReplica(block,
site,
group,
size=0,
last_update=int(time.time())))
else:
replica = DatasetReplica(dataset, site, growing=False)
for block in blocks:
replica.block_replicas.add(
BlockReplica(block,
site,
group,
size=0,
last_update=int(time.time())))
success.append(replica)
request_catalog = {}
chunk_size = 0
items = []
return success
def schedule_deletions(self,
replica_list,
operation_id,
comments=''): #override
sites = set(r.site for r, b in replica_list)
if len(sites) != 1:
raise OperationalError(
'schedule_copies should be called with a list of replicas at a single site.'
)
site = list(sites)[0]
if site.storage_type == Site.TYPE_MSS and not self.allow_tape_deletion:
LOG.warning('Deletion from MSS not allowed by configuration.')
return []
if self.allow_tape_deletion and self.auto_approval:
LOG.warning(
'You cannot have auto-approved tape deletions. Set auto-approval to False.'
)
return []
# execute the deletions in two steps: one for dataset-level and one for block-level
datasets = []
blocks = []
# maps used later for cloning
# getting ugly here.. should come up with a better way of making clones
replica_map = {}
block_replica_map = {}
for dataset_replica, block_replicas in replica_list:
if block_replicas is None:
datasets.append(dataset_replica.dataset)
else:
blocks.extend(br.block for br in block_replicas)
replica_map[dataset_replica.dataset] = dataset_replica
block_replica_map.update(
(br.block, br) for br in block_replicas)
success = []
deleted_datasets = self._run_deletion_request(operation_id, site,
'dataset', datasets,
comments)
for dataset in deleted_datasets:
replica = DatasetReplica(dataset,
site,
growing=False,
group=Group.null_group)
success.append((replica, None))
tmp_map = dict((dataset, []) for dataset in replica_map.iterkeys())
deleted_blocks = self._run_deletion_request(operation_id, site,
'block', blocks, comments)
for block in deleted_blocks:
tmp_map[block.dataset].append(block)
for dataset, blocks in tmp_map.iteritems():
replica = DatasetReplica(dataset, site)
replica.copy(replica_map[dataset])
success.append((replica, []))
for block in blocks:
block_replica = BlockReplica(block, site, Group.null_group)
block_replica.copy(block_replica_map[block])
block_replica.last_update = int(time.time())
success[-1][1].append(block_replica)
return success
def _determine_copies(self, partition, requests):
"""
@param partition Partition we copy into.
@param requests [(DealerRequest, plugin)]
@return {plugin: [new dataset replica]}
"""
# returned dict
copy_list = collections.defaultdict(list)
copy_volumes = dict(
(site, 0.) for site in self.policy.target_sites
) # keep track of how much we are assigning to each site
stats = {}
for plugin in self._plugin_priorities.keys():
stats[plugin.name] = {}
reject_stats = {
'Not a target site': 0,
'Replica exists': 0,
'Not allowed': 0,
'Destination is full': 0,
'Invalid request': 0,
'No destination available': 0,
'Source files missing': 0
}
# now go through all requests
for request, plugin in requests:
# make sure we have all blocks complete somewhere
if not self.policy.validate_source(request):
reject_stats['Source files missing'] += 1
continue
if request.destination is None:
# Randomly choose the destination site with probability proportional to free space
# request.destination will be set in the function
reject_reason = self.policy.find_destination_for(
request, partition)
else:
# Check the destination availability
reject_reason = self.policy.check_destination(
request, partition)
if reject_reason is not None:
reject_stats[reject_reason] += 1
continue
LOG.debug('Copying %s to %s requested by %s', request.item_name(),
request.destination.name, plugin.name)
try:
stat = stats[plugin.name][request.destination.name]
except KeyError:
stat = (0, 0)
stats[plugin.name][request.destination.name] = (
stat[0] + 1, stat[0] + request.item_size())
if request.block is not None:
blocks = [request.block]
growing = False
elif request.blocks is not None:
blocks = request.blocks
growing = False
else:
blocks = request.dataset.blocks
growing = True
new_replica = DatasetReplica(request.dataset,
request.destination,
growing=growing,
group=request.group)
for block in blocks:
new_replica.block_replicas.add(
BlockReplica(block,
request.destination,
request.group,
size=0))
copy_list[plugin].append(new_replica)
# New replicas may not be in the target partition, but we add the size up to be conservative
copy_volumes[request.destination] += request.item_size()
if not self.policy.is_target_site(
request.destination.partitions[partition],
copy_volumes[request.destination]):
LOG.info('%s is not a target site any more.',
request.destination.name)
self.policy.target_sites.remove(request.destination)
if sum(copy_volumes.itervalues()
) > self.policy.max_total_cycle_volume:
LOG.warning(
'Total copy volume has exceeded the limit. No more copies will be made.'
)
break
for plugin_name in sorted(stats.keys()):
plugin_stats = stats[plugin_name]
for destination_name in sorted(plugin_stats.keys()):
dest_stats = plugin_stats[destination_name]
LOG.info('Plugin %s requests %d items (%.1f TB) to %s',
plugin_name, dest_stats[0], dest_stats[1] * 1.e-12,
destination_name)
for reason in sorted(reject_stats.keys()):
LOG.info('%d items rejected for [%s]', reject_stats[reason],
reason)
return copy_list
def new_replica_from_dataset(dataset, site, group):
replica = DatasetReplica(dataset, site)
for block in dataset.blocks:
replica.block_replicas.add(BlockReplica(block, site, group, size = 0))
return replica
def maker_blockreplicas(block, block_entry, site_check=None):
"""Return a list of block replicas using blockreplicas data or a combination of blockreplicas and filereplicas calls."""
sites = {}
invalid_sites = set()
groups = {}
block_replicas = {}
for replica_entry in block_entry['replica']:
site_name = replica_entry['node']
try:
site = sites[site_name]
except KeyError:
if site_check:
if site_name in invalid_sites:
continue
if not site_check(site_name):
invalid_sites.add(site_name)
continue
site = sites[site_name] = Site(site_name)
group_name = replica_entry['group']
try:
group = groups[group_name]
except KeyError:
group = groups[group_name] = Group(group_name)
try:
time_update = int(replica_entry['time_update'])
except TypeError:
# time_update was None
time_update = 0
block_replica = BlockReplica(
block,
site,
group,
is_custodial=(replica_entry['custodial'] == 'y'),
last_update=time_update)
block_replicas[site_name] = block_replica
if replica_entry['complete'] == 'n':
# temporarily make this a list
block_replica.file_ids = []
block_replica.size = 0
LOG.info("Incomplete %s" % str(block_replica))
if 'file' in block_entry:
for file_entry in block_entry['file']:
for replica_entry in file_entry['replica']:
site_name = replica_entry['node']
try:
block_replica = block_replicas[site_name]
except KeyError:
continue
if block_replica.file_ids is None:
continue
# add LFN instead of file id
block_replica.file_ids.append(file_entry['name'])
file_size = file_entry['bytes']
if file_size is not None:
block_replica.size += file_size
try:
time_create = int(replica_entry['time_create'])
except TypeError:
pass
else:
if time_create > block_replica.last_update:
block_replica.last_update = time_create
for block_replica in block_replicas.itervalues():
if block_replica.file_ids is not None:
block_replica.file_ids = tuple(block_replica.file_ids)
return block_replicas.values()
def schedule_deletions(self,
replica_list,
operation_id,
comments=''): #override
sites = set(r.site for r, b in replica_list)
if len(sites) != 1:
raise OperationalError(
'schedule_deletions should be called with a list of replicas at a single site.'
)
site = list(sites)[0]
LOG.info(
'Scheduling deletion of %d replicas from %s using RLFSM (operation %d)',
len(replica_list), site.name, operation_id)
clones = []
for dataset_replica, block_replicas in replica_list:
if block_replicas is None:
to_delete = dataset_replica.block_replicas
else:
to_delete = block_replicas
for block_replica in to_delete:
for lfile in block_replica.files():
self.rlfsm.desubscribe_file(block_replica.site, lfile)
# No external dependency -> all operations are successful
clone_replica = DatasetReplica(dataset_replica.dataset,
dataset_replica.site)
clone_replica.copy(dataset_replica)
if block_replicas is None:
clones.append((clone_replica, None))
else:
clones.append((clone_replica, []))
for block_replica in block_replicas:
clone_block_replica = BlockReplica(block_replica.block,
block_replica.site,
block_replica.group)
clone_block_replica.copy(block_replica)
clone_block_replica.last_update = int(time.time())
clones[-1][1].append(clone_block_replica)
if not self._read_only:
for clone_replica, block_replicas in clones:
if block_replicas is None:
self.mysql.query(
'INSERT INTO `phedex_deletion_reservations` (`operation_id`, `item`, `site`) VALUES (%s, %s, %s)',
operation_id, clone_replica.dataset.name,
clone_replica.site.name)
else:
for block_replica in block_replicas:
self.mysql.query(
'INSERT INTO `phedex_deletion_reservations` (`operation_id`, `item`, `site`) VALUES (%s, %s, %s)',
operation_id, block_replica.block.full_name(),
clone_replica.site.name)
return clones
def new_replica_from_block(block, site, group):
return BlockReplica(block, site, group, size = 0)