Kafka cluster performance degradation (Kafka 0.8.2.1)

[Clelio De Souza <cl...@gmail.com>]

Hi all,


We are using Kafka in production and we have been facing some performance
degradation of the cluster, apparently after the cluster is a bit "old".


We have our production cluster which is up and running since 31/12/2015 and
performance tests on our application measuring a full round trip of TCP
packets and Kafka producing/consumption of data (3 hops in total for every
single TCP packet being sent, persisted and consumed in the other end). The
results for the production cluster shows a latency of ~ 130ms to 200ms.


In our Test environment we have the very same software and specification in
AWS instances, i.e. Test environment as being a mirror of Prod. The Kafka
cluster has been running in Test since 18/12/2015 and the same performance
tests (as described above) shows a increase of latency to ~ 800ms to 1000ms.


We have just recently setup a new fresh Kafka cluster (on 18/01/2016)
trying to get to the bottom of this performance degradation problem and in
the new Kafka cluster deployed in Test in replacement of the original Test
Kafka cluster we found a very small latency of ~ 10ms to 15ms.


We are using Kafka 0.8.2.1 version for all those environment mentioned
above. And the same cluster configuration has been setup on all of them, as
3 brokers as m3.xlarge AWS instances. The amount of data and Kafka topics
are roughly the same among those environments, therefore the performance
degradation seems to be not directly related to the amount of data in the
cluster. We suspect that something running inside of the Kafka cluster,
such as repartitioning or log compaction (even though our topics are to
setup to last for ~ 2 years.


The Kafka broker config can be found as below. If anyone could shed some
lights on what it may be causing the performance degradation for our Kafka
cluster, it would be great and very much appreciate it.


Thanks,
Leo

--------------------


# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements.  See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License.  You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# see kafka.server.KafkaConfig for additional details and defaults

############################# Server Basics #############################

# The id of the broker. This must be set to a unique integer for each
broker.
broker.id=<broker_num>

############################# Socket Server Settings
#############################

# The port the socket server listens on
port=9092

# Hostname the broker will bind to. If not set, the server will bind to all
interfaces
#host.name=localhost

# Hostname the broker will advertise to producers and consumers. If not
set, it uses the
# value for "host.name" if configured.  Otherwise, it will use the value
returned from
# java.net.InetAddress.getCanonicalHostName().
#advertised.host.name=<hostname routable by clients>

# The port to publish to ZooKeeper for clients to use. If this is not set,
# it will publish the same port that the broker binds to.
#advertised.port=<port accessible by clients>

# The number of threads handling network requests
num.network.threads=3

# The number of threads doing disk I/O
num.io.threads=8

# The send buffer (SO_SNDBUF) used by the socket server
socket.send.buffer.bytes=102400

# The receive buffer (SO_RCVBUF) used by the socket server
socket.receive.buffer.bytes=102400

# The maximum size of a request that the socket server will accept
(protection against OOM)
socket.request.max.bytes=104857600

############################# Log Basics #############################

# A comma seperated list of directories under which to store log files
log.dirs=/data/kafka/logs

# The default number of log partitions per topic. More partitions allow
greater
# parallelism for consumption, but this will also result in more files
across
# the brokers.
num.partitions=8

# The number of threads per data directory to be used for log recovery at
startup and flushing at shutdown.
# This value is recommended to be increased for installations with data
dirs located in RAID array.
num.recovery.threads.per.data.dir=1

############################# Log Flush Policy #############################

# Messages are immediately written to the filesystem but by default we only
fsync() to sync
# the OS cache lazily. The following configurations control the flush of
data to disk.
# There are a few important trade-offs here:
#    1. Durability: Unflushed data may be lost if you are not using
replication.
#    2. Latency: Very large flush intervals may lead to latency spikes when
the flush does occur as there will be a lot of data to flush.
#    3. Throughput: The flush is generally the most expensive operation,
and a small flush interval may lead to exceessive seeks.
# The settings below allow one to configure the flush policy to flush data
after a period of time or
# every N messages (or both). This can be done globally and overridden on a
per-topic basis.

# The number of messages to accept before forcing a flush of data to disk
#log.flush.interval.messages=10000

# The maximum amount of time a message can sit in a log before we force a
flush
#log.flush.interval.ms=1000

############################# Log Retention Policy
#############################

# The following configurations control the disposal of log segments. The
policy can
# be set to delete segments after a period of time, or after a given size
has accumulated.
# A segment will be deleted whenever *either* of these criteria are met.
Deletion always happens
# from the end of the log.

# The minimum age of a log file to be eligible for deletion
# Failsafe is we don't lose any messages for 20+ years, topics should
# be configured individually
log.retention.hours=200000

# A size-based retention policy for logs. Segments are pruned from the log
as long as the remaining
# segments don't drop below log.retention.bytes.
#log.retention.bytes=1073741824

# The maximum size of a log segment file. When this size is reached a new
log segment will be created.
log.segment.bytes=1073741824

# The interval at which log segments are checked to see if they can be
deleted according
# to the retention policies
log.retention.check.interval.ms=300000

# By default the log cleaner is disabled and the log retention policy will
default to just delete segments after their retention expires.
# If log.cleaner.enable=true is set the cleaner will be enabled and
individual logs can then be marked for log compaction.
log.cleaner.enable=false

default.replication.factor=3

auto.create.topics.enable=true

controlled.shutdown.enable=true

delete.topic.enable=true

############################# Zookeeper #############################

# Zookeeper connection string (see zookeeper docs for details).
# This is a comma separated host:port pairs, each corresponding to a zk
# server. e.g. "127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002".
# You can also append an optional chroot string to the urls to specify the
# root directory for all kafka znodes.
zookeeper.connect=<zk1-address>:2181,<zk2-address>:2181,<zk3-address>:2181

# Timeout in ms for connecting to zookeeper
zookeeper.connection.timeout.ms=6000