Files
opensearch-pyd/benchmarks/bench_sync.py
T
Daniel (dB.) Doubrovkine af263a060a Add micro benchmarks. (#537)
* Align pool_maxsize for different connection pool implementations.

Signed-off-by: dblock <dblock@amazon.com>

* Added benchmarks.

Signed-off-by: dblock <dblock@amazon.com>

* Multi-threaded vs. async benchmarks.

Signed-off-by: dblock <dblock@amazon.com>

* Set pool size to the number of threads.

Signed-off-by: dblock <dblock@amazon.com>

* Added sync/async benchmark.

Signed-off-by: dblock <dblock@amazon.com>

* Report client-side latency.

Signed-off-by: dblock <dblock@amazon.com>

* Various updates to benchmarks, demonstrating threading improves throughput.

Signed-off-by: dblock <dblock@amazon.com>

* Bench info.

Signed-off-by: dblock <dblock@amazon.com>

* Fixup format.

Signed-off-by: dblock <dblock@amazon.com>

* Undo async maxsize.

Signed-off-by: dblock <dblock@amazon.com>

* Moved benchmarks folder.

Signed-off-by: dblock <dblock@amazon.com>

* Updated documentation and project description.

Signed-off-by: dblock <dblock@amazon.com>

---------

Signed-off-by: dblock <dblock@amazon.com>
2023-10-13 11:29:54 -07:00

133 lines
3.1 KiB
Python

#!/usr/bin/env python
# SPDX-License-Identifier: Apache-2.0
#
# The OpenSearch Contributors require contributions made to
# this file be licensed under the Apache-2.0 license or a
# compatible open source license.
import json
import time
import uuid
from thread_with_return_value import ThreadWithReturnValue
from opensearchpy import OpenSearch, Urllib3HttpConnection
host = "localhost"
port = 9200
auth = ("admin", "admin")
index_name = "test-index-sync"
item_count = 1000
import logging
import sys
root = logging.getLogger()
# root.setLevel(logging.DEBUG)
# logging.getLogger("urllib3.connectionpool").setLevel(logging.DEBUG)
handler = logging.StreamHandler(sys.stdout)
handler.setLevel(logging.DEBUG)
formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
handler.setFormatter(formatter)
root.addHandler(handler)
def index_records(client, item_count):
tt = 0
for n in range(10):
data = []
for i in range(item_count):
data.append(
json.dumps({"index": {"_index": index_name, "_id": str(uuid.uuid4())}})
)
data.append(json.dumps({"value": i}))
data = "\n".join(data)
start = time.time() * 1000
rc = client.bulk(data)
if rc["errors"]:
raise Exception(rc["errors"])
server_time = rc["took"]
total_time = time.time() * 1000 - start
if total_time < server_time:
raise Exception(f"total={total_time} < server={server_time}")
tt += total_time - server_time
return tt
def test(thread_count=1, item_count=1, client_count=1):
clients = []
for i in range(client_count):
clients.append(
OpenSearch(
hosts=[{"host": host, "port": port}],
http_auth=auth,
use_ssl=True,
verify_certs=False,
ssl_show_warn=False,
pool_maxsize=thread_count,
connection_class=Urllib3HttpConnection,
)
)
if clients[0].indices.exists(index_name):
clients[0].indices.delete(index_name)
clients[0].indices.create(
index=index_name,
body={
"mappings": {
"properties": {
"value": {"type": "float"},
}
}
},
)
threads = []
for thread_id in range(thread_count):
thread = ThreadWithReturnValue(
target=index_records, args=[clients[thread_id % len(clients)], item_count]
)
threads.append(thread)
thread.start()
latency = 0
for t in threads:
latency += t.join()
clients[0].indices.refresh(index=index_name)
count = clients[0].count(index=index_name)
clients[0].indices.delete(index_name)
print(f"{count}, latency={latency}")
def test_1():
test(1, 32 * item_count, 1)
def test_2():
test(2, 16 * item_count, 2)
def test_4():
test(4, 8 * item_count, 3)
def test_8():
test(8, 4 * item_count, 8)
def test_32():
test(32, item_count, 32)
__benchmarks__ = [(test_1, test_32, "1 thread vs. 32 threads (sync)")]