Question

我们的一名DBA使用相同的Python代码（如下）在AWS EC2上将Cassandra对Oracle进行了基准测试以提高INSERT性能（1M条记录），并获得了以下令人惊讶的结果：

Oracle 12.2 ，单节点，64cores / 256GB，EC2 EBS存储， 38秒

Cassandra 5.1.13（DDAC），单节点，2核/ 4GB，EC2 EBS存储， 464秒

Cassandra 3.11.4，四个节点，16cores / 64GB（每个节点），EC2 EBS存储， 486秒

SO-我们在做什么错了？
Cassandra的表现如何如此缓慢？
*没有足够的节点？（为什么四个节点比单个节点要慢？）
*配置问题？
*还有吗？

谢谢！

以下是Python代码：

import logging
import time
from cassandra import ConsistencyLevel
from cassandra.cluster import Cluster, BatchStatement
from cassandra.query import SimpleStatement
from cassandra.auth import PlainTextAuthProvider

class PythonCassandraExample:

    def __init__(self):
        self.cluster = None
        self.session = None
        self.keyspace = None
        self.log = None

    def __del__(self):
        self.cluster.shutdown()

    def createsession(self):
        auth_provider = PlainTextAuthProvider(username='cassandra', password='cassandra')
        self.cluster = Cluster(['10.220.151.138'],auth_provider = auth_provider)
        self.session = self.cluster.connect(self.keyspace)

    def getsession(self):
        return self.session

    # How about Adding some log info to see what went wrong
    def setlogger(self):
        log = logging.getLogger()
        log.setLevel('INFO')
        handler = logging.StreamHandler()
        handler.setFormatter(logging.Formatter("%(asctime)s [%(levelname)s] %(name)s: %(message)s"))
        log.addHandler(handler)
        self.log = log

    # Create Keyspace based on Given Name
    def createkeyspace(self, keyspace):
        """
        :param keyspace:  The Name of Keyspace to be created
        :return:
        """
        # Before we create new lets check if exiting keyspace; we will drop that and create new
        rows = self.session.execute("SELECT keyspace_name FROM system_schema.keyspaces")
        if keyspace in [row[0] for row in rows]:
            self.log.info("dropping existing keyspace...")
            self.session.execute("DROP KEYSPACE " + keyspace)

        self.log.info("creating keyspace...")
        self.session.execute("""
                CREATE KEYSPACE %s
                WITH replication = { 'class': 'SimpleStrategy', 'replication_factor': '2' }
                """ % keyspace)

        self.log.info("setting keyspace...")
        self.session.set_keyspace(keyspace)

    def create_table(self):
        c_sql = """
                CREATE TABLE IF NOT EXISTS employee (emp_id int PRIMARY KEY,
                                              ename varchar,
                                              sal double,
                                              city varchar);
                 """
        self.session.execute(c_sql)
        self.log.info("Employee Table Created !!!")

    # lets do some batch insert
    def insert_data(self):
        i = 1
        while i < 1000000:
          insert_sql = self.session.prepare("INSERT INTO  employee (emp_id, ename , sal,city) VALUES (?,?,?,?)")
          batch = BatchStatement()
          batch.add(insert_sql, (i, 'Danny', 2555, 'De-vito'))
          self.session.execute(batch)
          # self.log.info('Batch Insert Completed for ' + str(i))
          i += 1

    # def select_data(self):
    #    rows = self.session.execute('select count(*) from perftest.employee limit 5;')
    #    for row in rows:
    #        print(row.ename, row.sal)

    def update_data(self):
        pass

    def delete_data(self):
        pass


if __name__ == '__main__':
    example1 = PythonCassandraExample()
    example1.createsession()
    example1.setlogger()
    example1.createkeyspace('perftest')
    example1.create_table()

    # Populate perftest.employee table
    start = time.time()
    example1.insert_data()
    end = time.time()
    print ('Duration: ' + str(end-start) + ' sec.')

    # example1.select_data()

Answer 1

这里有多个问题：

对于第二个测试，您没有为DDAC分配足够的内存和内核，因此Cassandra仅获得1Gb堆-默认情况下，Cassandra占用所有可用内存的1/4。第三项测试也是如此-它将仅获得16Gb RAM用于堆，您可能需要将其提高到更高的值，例如24Gb甚至更高。
不清楚每个测试中有多少个IOP-EBS的吞吐量取决于卷的大小及其类型
您正在使用同步API执行命令-基本上，您在确认已插入前一个项目之后才插入下一个项目。最好的吞吐量可以通过using asynchronous API;
您正在每次迭代中准备语句-导致每次将CQL字符串发送到服务器，因此这会减慢所有操作-只需将insert_sql = self.session.prepare(行移出循环即可即可。
（不完全相关），您正在使用批处理语句写入数据-它是anti-pattern in Cassandra，因为数据仅发送到一个节点，然后应将数据分发到真正拥有该数据的节点。这解释了为什么4个节点的群集比1个节点的群集差。

P.S。实际的负载测试非常困难。为此，有专门的工具，例如，您可以在this blog post中找到更多信息。

Answer 2

下面的更新代码将每100条记录批处理：

"""
Python  by Techfossguru
Copyright (C) 2017  Satish Prasad

"""
import logging
import warnings
import time
from cassandra import ConsistencyLevel
from cassandra.cluster import Cluster, BatchStatement
from cassandra.query import SimpleStatement
from cassandra.auth import PlainTextAuthProvider

class PythonCassandraExample:

    def __init__(self):
        self.cluster = None
        self.session = None
        self.keyspace = None
        self.log = None

    def __del__(self):
        self.cluster.shutdown()

    def createsession(self):
        auth_provider = PlainTextAuthProvider(username='cassandra', password='cassandra')
        self.cluster = Cluster(['10.220.151.138'],auth_provider = auth_provider)
        self.session = self.cluster.connect(self.keyspace)

    def getsession(self):
        return self.session

    # How about Adding some log info to see what went wrong
    def setlogger(self):
        log = logging.getLogger()
        log.setLevel('INFO')
        handler = logging.StreamHandler()
        handler.setFormatter(logging.Formatter("%(asctime)s [%(levelname)s] %(name)s: %(message)s"))
        log.addHandler(handler)
        self.log = log

    # Create Keyspace based on Given Name
    def createkeyspace(self, keyspace):
        """
        :param keyspace:  The Name of Keyspace to be created
        :return:
        """
        # Before we create new lets check if exiting keyspace; we will drop that and create new
        rows = self.session.execute("SELECT keyspace_name FROM system_schema.keyspaces")
        if keyspace in [row[0] for row in rows]:
            self.log.info("dropping existing keyspace...")
            self.session.execute("DROP KEYSPACE " + keyspace)

        self.log.info("creating keyspace...")
        self.session.execute("""
                CREATE KEYSPACE %s
                WITH replication = { 'class': 'SimpleStrategy', 'replication_factor': '2' }
                """ % keyspace)

        self.log.info("setting keyspace...")
        self.session.set_keyspace(keyspace)

    def create_table(self):
        c_sql = """
                CREATE TABLE IF NOT EXISTS employee (emp_id int PRIMARY KEY,
                                              ename varchar,
                                              sal double,
                                              city varchar);
                 """
        self.session.execute(c_sql)
        self.log.info("Employee Table Created !!!")

    # lets do some batch insert
    def insert_data(self):
        i = 1
        insert_sql = self.session.prepare("INSERT INTO  employee (emp_id, ename , sal,city) VALUES (?,?,?,?)")
        batch = BatchStatement()
        warnings.filterwarnings("ignore", category=FutureWarning)

        while i < 1000001:
          # insert_sql = self.session.prepare("INSERT INTO  employee (emp_id, ename , sal,city) VALUES (?,?,?,?)")
          # batch = BatchStatement()
          batch.add(insert_sql, (i, 'Danny', 2555, 'De-vito'))

          # Commit every 100 records
          if (i % 100 == 0):
             self.session.execute(batch)
             batch = BatchStatement()
             # self.log.info('Batch Insert Completed for ' + str(i))
          i += 1
        self.session.execute(batch)

    # def select_data(self):
    #    rows = self.session.execute('select count(*) from actimize.employee limit 5;')
    #    for row in rows:
    #        print(row.ename, row.sal)

    def update_data(self):
        pass

    def delete_data(self):
        pass


if __name__ == '__main__':
    example1 = PythonCassandraExample()
    example1.createsession()
    example1.setlogger()
    example1.createkeyspace('actimize')
    example1.create_table()

    # Populate actimize.employee table
    start = time.time()
    example1.insert_data()
    end = time.time()
    print ('Duration: ' + str(end-start) + ' sec.')

    # example1.select_data()

Cassandra在AWS上的性能下降

2 个答案: