Kafka 缓冲区错误:本地:Python中的队列已满

ldioqlga  于 2023-04-29  发布在  Apache
关注(0)|答案(1)|浏览(304)
import logging
from confluent_kafka import Producer
import os

logger = logging.getLogger("main")

BOOTSTRAP_SERVERS = os.environ['BOOTSTRAP_SERVERS']
APPLICATION_ID = os.getenv('APPLICATION_ID', default = "nke-data-source")
RECONNECT_BACKOFF_MS = os.getenv('RECONNECT_BACKOFF_MS', default = 1000)
REQUEST_TIMEOUT_MS = os.getenv('REQUEST_TIMEOUT_MS', default = 40000)
ACKS = os.getenv('ACKS', default = "all")
RETRIES = os.getenv('RETRIES', default = 15)
RETRY_BACK_OFF = os.getenv('RETRY_BACK_OFF', default = 1000)
MAX_IN_FLIGHT_REQUESTS = os.getenv('MAX_IN_FLIGHT_REQUESTS', default = 1)
topic = os.getenv('OUTBOUND_TOPIC', default = "tti-nke-raw")

p = Producer({'bootstrap.servers': BOOTSTRAP_SERVERS, 
    'client.id': APPLICATION_ID, 
    'reconnect.backoff.ms': RECONNECT_BACKOFF_MS,
    'request.timeout.ms': REQUEST_TIMEOUT_MS,
    'acks': ACKS,
    'retries': RETRIES,
    'retry.backoff.ms': RETRY_BACK_OFF,
    'max.in.flight.requests.per.connection': MAX_IN_FLIGHT_REQUESTS,
    'compression.type': "lz4"})

def send(key, event):
    try:
        logger.info("Sending key: [{0}] value: [{1}]".format(key, event))
        p.produce(topic=topic, value=event.encode('utf-8'), key=key)
    except Exception:
        logger.error("error sending events to kafka", exc_info=True)

错误:-

Traceback (most recent call last):
BufferError: Local: Queue full
File "/app/sender.py", line 30, in send
p.produce(topic=topic, value=event.encode('utf-8'), key=key)

有人能帮助我在这方面,因为我是新的Python

wa7juj8i

wa7juj8i1#

这个Queue是在librdkafka库中实现的(confluent_kafka绑定到这个库)
produce有一个内部Queue,它接收production交付报告并等待production处理这些报告(大多数情况下什么都不做),但是您需要触发这种遍历队列的机制,可以通过调用poll来执行
你应该在每次调用produce之后调用producer.poll(0),所以改变:

p.produce(topic=topic, value=event.encode('utf-8'), key=key)

致:

p.produce(topic=topic, value=event.encode('utf-8'), key=key)
p.poll(0)

这将触发队列清理,不要担心性能,因为这是一个非常简单的函数,并没有像librdkafka的作者所写的那样做太多:
poll()的调用很便宜,它不会对性能产生影响,所以请将它添加到您的生产者循环中。
基本上它是这样做的:
定期调用poll()来服务生产者的交付报告回调。
考虑在此Issue中阅读有关此内容

相关问题