def __init__(self, settings, cluster_spec: ClusterSpec,
test_config: TestConfig):
self.cluster_spec = cluster_spec
self.test_config = test_config
self.mc = MetadataClient(settings)
self.store = PerfStore(settings.cbmonitor_host)
self.workload_setting = PhaseSettings
self.interval = self.MAX_SAMPLING_INTERVAL
self.cluster = settings.cluster
self.clients = []
self.cb_host = self.cluster_spec.servers[int(self.test_config.nodes)]
self.sg_host = next(self.cluster_spec.masters)
src_client = new_client(host=self.cb_host,
bucket='bucket-1',
password='******',
timeout=self.TIMEOUT)
self.clients.append(('bucket-1', src_client))
self.new_docs = Document(1024)
class DurabilityLatency(ObserveIndexLatency, Latency):
COLLECTOR = "durability"
METRICS = "latency_replicate_to", "latency_persist_to"
DURABILITY_TIMEOUT = 120
def __init__(self, settings, workload):
super().__init__(settings)
self.new_docs = Document(workload.size)
self.pools = self.init_pool(settings)
@staticmethod
def gen_key() -> Key:
return Key(number=numpy.random.random_integers(0, 10 ** 9),
prefix='endure',
fmtr='hex')
def endure(self, pool, metric):
client = pool.get_client()
key = self.gen_key()
doc = self.new_docs.next(key)
t0 = time()
client.upsert(key.string, doc)
if metric == "latency_persist_to":
client.endure(key.string, persist_to=1, replicate_to=0, interval=0.010,
timeout=120)
else:
client.endure(key.string, persist_to=0, replicate_to=1, interval=0.001)
latency = 1000 * (time() - t0) # Latency in ms
sleep_time = max(0, self.MAX_POLLING_INTERVAL - latency)
client.delete(key.string)
pool.release_client(client)
return {metric: latency}, sleep_time
def sample(self):
while True:
for bucket, pool in self.pools:
for metric in self.METRICS:
try:
stats, sleep_time = self.endure(pool, metric)
self.store.append(stats,
cluster=self.cluster,
bucket=bucket,
collector=self.COLLECTOR)
sleep(sleep_time)
except Exception as e:
logger.warn(e)
def collect(self):
ObserveIndexLatency.collect(self)
def init_docs(self):
if not hasattr(self.ws, 'doc_gen') or self.ws.doc_gen == 'basic':
self.docs = Document(self.ws.size)
elif self.ws.doc_gen == 'string':
self.docs = String(self.ws.size)
elif self.ws.doc_gen == 'nested':
self.docs = NestedDocument(self.ws.size)
elif self.ws.doc_gen == 'reverse_lookup':
self.docs = ReverseLookupDocument(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'reverse_range_lookup':
self.docs = ReverseRangeLookupDocument(self.ws.size,
prefix='n1ql',
range_distance=self.ws.range_distance)
elif self.ws.doc_gen == 'ext_reverse_lookup':
self.docs = ExtReverseLookupDocument(self.ws.size,
self.ts.prefix,
self.ws.items)
elif self.ws.doc_gen == 'join':
self.docs = JoinedDocument(self.ws.size,
self.ts.prefix,
self.ws.items,
self.ws.num_categories,
self.ws.num_replies)
elif self.ws.doc_gen == 'ref':
self.docs = RefDocument(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'array_indexing':
self.docs = ArrayIndexingDocument(self.ws.size,
self.ts.prefix,
self.ws.array_size,
self.ws.items)
elif self.ws.doc_gen == 'profile':
self.docs = ProfileDocument(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'import_export_simple':
self.docs = ImportExportDocument(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'import_export_array':
self.docs = ImportExportDocumentArray(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'import_export_nested':
self.docs = ImportExportDocumentNested(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'large_subdoc':
self.docs = LargeDocument(self.ws.size)
elif self.ws.doc_gen == 'gsi_multiindex':
self.docs = GSIMultiIndexDocument(self.ws.size)
elif self.ws.doc_gen == 'small_plasma':
self.docs = SmallPlasmaDocument(self.ws.size)
elif self.ws.doc_gen == 'sequential_plasma':
self.docs = SequentialPlasmaDocument(self.ws.size)
elif self.ws.doc_gen == 'large_item_plasma':
self.docs = LargeItemPlasmaDocument(self.ws.size,
self.ws.item_size)
elif self.ws.doc_gen == 'varying_item_plasma':
self.docs = VaryingItemSizePlasmaDocument(self.ws.size,
self.ws.size_variation_min,
self.ws.size_variation_max)
def __init__(self, settings, workload):
super().__init__(settings)
self.interval = self.MAX_SAMPLING_INTERVAL
self.clients = []
for bucket in self.get_buckets():
src_client = new_client(host=settings.master_node,
bucket=bucket,
password=settings.bucket_password,
timeout=self.TIMEOUT)
dst_client = new_client(host=settings.dest_master_node,
bucket=bucket,
password=settings.bucket_password,
timeout=self.TIMEOUT)
self.clients.append((bucket, src_client, dst_client))
self.new_docs = Document(workload.size)
class DurabilityLatency(ObserveIndexLatency, Latency):
COLLECTOR = "durability"
METRICS = "latency_replicate_to", "latency_persist_to"
DURABILITY_TIMEOUT = 120
def __init__(self, settings, workload):
super().__init__(settings)
self.new_docs = Document(workload.size)
self.pools = self.init_pool(settings)
def endure(self, pool, metric):
client = pool.get_client()
key = uhex()
doc = self.new_docs.next(key)
t0 = time()
client.upsert(key, doc)
if metric == "latency_persist_to":
client.endure(key, persist_to=1, replicate_to=0, interval=0.010,
timeout=120)
else:
client.endure(key, persist_to=0, replicate_to=1, interval=0.001)
latency = 1000 * (time() - t0) # Latency in ms
sleep_time = max(0, self.MAX_POLLING_INTERVAL - latency)
client.delete(key)
pool.release_client(client)
return {metric: latency}, sleep_time
def sample(self):
while True:
for bucket, pool in self.pools:
for metric in self.METRICS:
try:
stats, sleep_time = self.endure(pool, metric)
self.store.append(stats,
cluster=self.cluster,
bucket=bucket,
collector=self.COLLECTOR)
sleep(sleep_time)
except Exception as e:
logger.warn(e)
def collect(self):
ObserveIndexLatency.collect(self)
def init_docs(self):
if not hasattr(self.ws, 'doc_gen') or self.ws.doc_gen == 'basic':
self.docs = Document(self.ws.size)
elif self.ws.doc_gen == 'string':
self.docs = String(self.ws.size)
elif self.ws.doc_gen == 'nested':
self.docs = NestedDocument(self.ws.size)
elif self.ws.doc_gen == 'reverse_lookup':
self.docs = ReverseLookupDocument(self.ws.size, self.ts.prefix)
elif self.ws.doc_gen == 'reverse_range_lookup':
self.docs = ReverseRangeLookupDocument(self.ws.size,
self.ts.prefix,
self.ws.range_distance)
elif self.ws.doc_gen == 'ext_reverse_lookup':
self.docs = ExtReverseLookupDocument(self.ws.size, self.ts.prefix,
self.ws.items)
elif self.ws.doc_gen == 'hash_join':
self.docs = HashJoinDocument(self.ws.size, self.ts.prefix,
self.ws.range_distance)
elif self.ws.doc_gen == 'join':
self.docs = JoinedDocument(self.ws.size, self.ts.prefix,
self.ws.items, self.ws.num_categories,
self.ws.num_replies)
elif self.ws.doc_gen == 'ref':
self.docs = RefDocument(self.ws.size, self.ts.prefix)
elif self.ws.doc_gen == 'array_indexing':
self.docs = ArrayIndexingDocument(self.ws.size, self.ts.prefix,
self.ws.array_size,
self.ws.items)
elif self.ws.doc_gen == 'array_indexing_unique':
self.docs = ArrayIndexingUniqueDocument(self.ws.size,
self.ts.prefix,
self.ws.array_size,
self.ws.items)
elif self.ws.doc_gen == 'array_indexing_range_scan':
self.docs = ArrayIndexingRangeScanDocument(self.ws.size,
self.ts.prefix,
self.ws.array_size,
self.ws.items)
elif self.ws.doc_gen == 'profile':
self.docs = ProfileDocument(self.ws.size, self.ts.prefix)
elif self.ws.doc_gen == 'import_export_simple':
self.docs = ImportExportDocument(self.ws.size, self.ts.prefix)
elif self.ws.doc_gen == 'import_export_array':
self.docs = ImportExportDocumentArray(self.ws.size, self.ts.prefix)
elif self.ws.doc_gen == 'import_export_nested':
self.docs = ImportExportDocumentNested(self.ws.size,
self.ts.prefix)
elif self.ws.doc_gen == 'large':
self.docs = LargeDocument(self.ws.size)
elif self.ws.doc_gen == 'gsi_multiindex':
self.docs = GSIMultiIndexDocument(self.ws.size)
elif self.ws.doc_gen == 'small_plasma':
self.docs = SmallPlasmaDocument(self.ws.size)
elif self.ws.doc_gen == 'sequential_plasma':
self.docs = SequentialPlasmaDocument(self.ws.size)
elif self.ws.doc_gen == 'large_item_plasma':
self.docs = LargeItemPlasmaDocument(self.ws.size,
self.ws.item_size)
elif self.ws.doc_gen == 'varying_item_plasma':
self.docs = VaryingItemSizePlasmaDocument(
self.ws.size, self.ws.size_variation_min,
self.ws.size_variation_max)
elif self.ws.doc_gen == 'eventing_small':
self.docs = EventingSmallDocument(self.ws.size)
elif self.ws.doc_gen == 'tpc_ds':
self.docs = TpcDsDocument()
elif self.ws.doc_gen == 'package':
self.docs = PackageDocument(self.ws.size)
elif self.ws.doc_gen == 'incompressible':
self.docs = IncompressibleString(self.ws.size)
elif self.ws.doc_gen == 'big_fun':
self.docs = BigFunDocument()
elif self.ws.doc_gen == 'multibucket':
self.docs = MultiBucketDocument(self.ws.size)
elif self.ws.doc_gen == 'advancedfilter':
self.docs = AdvFilterDocument(self.ws.size)
elif self.ws.doc_gen == 'advancedfilterxattr':
self.docs = AdvFilterXattrBody(self.ws.size)
def __init__(self, settings, workload):
super().__init__(settings)
self.new_docs = Document(workload.size)
self.pools = self.init_pool(settings)
class XdcrLag(Latency):
COLLECTOR = "xdcr_lag"
METRICS = "xdcr_lag",
INITIAL_POLLING_INTERVAL = 0.001 # 1 ms
TIMEOUT = 600 # 10 minutes
MAX_SAMPLING_INTERVAL = 0.25 # 250 ms
def __init__(self, settings, workload):
super().__init__(settings)
self.interval = self.MAX_SAMPLING_INTERVAL
self.clients = []
for bucket in self.get_buckets():
src_client = new_client(host=settings.master_node,
bucket=bucket,
password=settings.bucket_password,
timeout=self.TIMEOUT)
dst_client = new_client(host=settings.dest_master_node,
bucket=bucket,
password=settings.bucket_password,
timeout=self.TIMEOUT)
self.clients.append((bucket, src_client, dst_client))
self.new_docs = Document(workload.size)
@staticmethod
def gen_key() -> Key:
return Key(number=numpy.random.random_integers(0, 10 ** 9),
prefix='xdcr',
fmtr='hex')
def measure(self, src_client, dst_client):
key = self.gen_key()
doc = self.new_docs.next(key)
polling_interval = self.INITIAL_POLLING_INTERVAL
src_client.upsert(key.string, doc)
t0 = time()
while time() - t0 < self.TIMEOUT:
if dst_client.get(key.string, quiet=True).success:
break
sleep(polling_interval)
polling_interval *= 1.05 # increase interval by 5%
else:
logger.warn('XDCR sampling timed out after {} seconds'
.format(self.TIMEOUT))
t1 = time()
src_client.remove(key.string, quiet=True)
dst_client.remove(key.string, quiet=True)
return {'xdcr_lag': (t1 - t0) * 1000} # s -> ms
def sample(self):
for bucket, src_client, dst_client in self.clients:
lags = self.measure(src_client, dst_client)
self.store.append(lags,
cluster=self.cluster,
bucket=bucket,
collector=self.COLLECTOR)
class SGImport_latency(Collector):
COLLECTOR = "sgimport_latency"
METRICS = "sgimport_latency"
INITIAL_POLLING_INTERVAL = 0.001 # 1 ms
TIMEOUT = 3600 # 1hr minutes
MAX_SAMPLING_INTERVAL = 10 # 250 ms
def __init__(self, settings, cluster_spec: ClusterSpec,
test_config: TestConfig):
self.cluster_spec = cluster_spec
self.test_config = test_config
self.mc = MetadataClient(settings)
self.store = PerfStore(settings.cbmonitor_host)
self.workload_setting = PhaseSettings
self.interval = self.MAX_SAMPLING_INTERVAL
self.cluster = settings.cluster
self.clients = []
self.cb_host = self.cluster_spec.servers[int(self.test_config.nodes)]
self.sg_host = next(self.cluster_spec.masters)
src_client = new_client(host=self.cb_host,
bucket='bucket-1',
password='******',
timeout=self.TIMEOUT)
self.clients.append(('bucket-1', src_client))
self.new_docs = Document(1024)
def check_longpoll_changefeed(self, host: str, key: str,
last_sequence: str):
sg_db = 'db'
api = 'http://{}:4985/{}/_changes'.format(host, sg_db)
last_sequence_str = "{}".format(last_sequence)
data = {
'filter': 'sync_gateway/bychannel',
'feed': 'longpoll',
"channels": "123",
"since": last_sequence_str,
"heartbeat": 3600000
}
response = requests.post(url=api, data=json.dumps(data))
t1 = time()
record_found = 0
if response.status_code == 200:
for record in response.json()['results']:
if record['id'] == key:
record_found = 1
break
if record_found != 1:
self.check_longpoll_changefeed(host=host,
key=key,
last_sequence=last_sequence)
return t1
def insert_doc(self, src_client, key: str, doc):
src_client.upsert(key, doc)
return time()
def get_lastsequence(self, host: str):
sg_db = 'db'
api = 'http://{}:4985/{}/_changes'.format(host, sg_db)
data = {
'filter': 'sync_gateway/bychannel',
'feed': 'normal',
"channels": "123",
"since": "0"
}
response = requests.post(url=api, data=json.dumps(data))
last_sequence = response.json()['last_seq']
return last_sequence
def measure(self, src_client):
key = "sgimport_{}".format(uhex())
doc = self.new_docs.next(key)
last_sequence = self.get_lastsequence(host=self.sg_host)
executor = ThreadPoolExecutor(max_workers=2)
future1 = executor.submit(self.check_longpoll_changefeed,
host=self.sg_host,
key=key,
last_sequence=last_sequence)
future2 = executor.submit(self.insert_doc,
src_client=src_client,
key=key,
doc=doc)
t1, t0 = future1.result(), future2.result()
print('import latency t1, t0', t1, t0, (t1 - t0) * 1000)
return {'sgimport_latency': (t1 - t0) * 1000} # s -> ms
def sample(self):
for bucket, src_client in self.clients:
lags = self.measure(src_client)
self.store.append(lags,
cluster=self.cluster,
collector=self.COLLECTOR)
def update_metadata(self):
self.mc.add_cluster()
self.mc.add_metric(self.METRICS, collector=self.COLLECTOR)
def __init__(self, settings, workload):
super().__init__(settings)
self.new_docs = Document(workload.size)
self.pools = self.init_pool(settings)
def update(self, key, subdoc_fields, size):
newdoc = Document(size)
alphabet = newdoc._build_alphabet(key)
for field in subdoc_fields.split(','):
new_field_value = getattr(newdoc, '_build_' + field)(alphabet)
self.client.mutate_in(key, SD.upsert(field, new_field_value))