Stephen Chu created KAFKA-2614:
----------------------------------
Summary: No more clients can connect after
`TooManyConnectionsException` threshold (max.connections.per.ip) is reached
Key: KAFKA-2614
URL: https://issues.apache.org/jira/browse/KAFKA-2614
Project: Kafka
Issue Type: Bug
Affects Versions: 0.8.2.1
Environment: Debian Jessie
Reporter: Stephen Chu
Priority: Critical
It seems no more clients can connect to Kafka after `max.connections.per.ip` is
reached, even if previous clients were already disconnected.
Using 0.8.3 (9c936b18), upon starting a fresh Kafka server that is configured
with (max.connections.per.ip = 24), I noticed that I can cause the server to
hit the error case of {{INFO Rejected connection from /0:0:0:0:0:0:0:1, address
already has the configured maximum of 24 connections.}} very quickly, by simply
looping through a bunch of simple clients against the server:
{noformat}
#! /bin/bash
for i in {1..30}; do
# either:
nc -vz 127.0.0.1 9092;
# or:
( telnet 127.0.0.1 9092; ) &
done
# if using telnet, kill all connected jobs now via:
kill %{2..31}
{noformat}
The problem seems to be that the counter for such short-lived client
connections aren't properly decrementing when using the
`max.connections.per.ip` feature.
Turning on DEBUG logs, I cannot see the log lines "Closing connection from xxx"
on [this
line|https://github.com/apache/kafka/blob/0.8.2/core/src/main/scala/kafka/network/SocketServer.scala#L164]
from the first few still-under-threshold short-lived connections, but starts
showing *after* I hit the limit per that config.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
