1. 101* ways to configure Kafka
- badly
Hva FINN.no har lært av å kjøre Kafka
Audun Fauchald Strand
Lead Developer Infrastructure
@audunstrand
bio: gof, mq, ejb,
wli, bpel eda, soa
esb, ddd, k8s
Henning Spjelkavik
Architect
@spjelkavik
bio: Skiinfo (Vail Resorts),
FINN.no
enjoys reading jstacks
2. agenda
Introduction to kafka
kafka @ finn.no
101* mistakes
questions
“From a certain point onward
there is no longer any turning
back. That is the point that must
be reached.”
― Franz Kafka, The Trial
4. why use kafka
#notAnESB
what is a log
terminology
components
giant leap
“A First Sign of the Beginning of
Understanding is the Wish to
Die.”
― Franz Kafka
https://commons.wikimedia.org/wiki/File:Kafka.jpg
5. Why use Kafka?
“Apache Kafka is publish-subscribe messaging
rethought as a distributed commit log.”
Fast
Scalable
Durable
Distributed by design
Sweet spot: High volume, low latency
Quora:
“Use Kafka if you have a fire hose of events (100k+/sec)
you need delivered in partitioned order 'at least once' with
a mix of online and batch consumers, you want to be able
to re-read messages”
“Use Rabbit if you have messages (20k+/sec) that need to
be routed in complex ways to consumers, you want per-
message delivery guarantees, you don't care about ordered
delivery”
6. What is a (data) log / journal?
A log is perhaps the simplest possible storage abstraction.
It is an append-only, totally-ordered sequence of records ordered by time.
Appended to the end of the log, reads proceed left-to-right.
Each entry is assigned a unique sequential log entry number.
The ordering of records defines a notion of "time" since entries to the left are
defined to be older then entries to the right.
This is a data log, not an application log (i.e not log4j)
7. Changelog 101: Tables and Events are Dual
Duality: a log of changes and a table.
Accounting
log: credit and debit (events pr key)
table: all current balances (i.e state pr key)
In a sense the log is the more fundamental data structure: in addition to creating the
original table you can also transform it to create all kinds of derived tables.
8. producers writes to brokers
consumers reads from brokers
everything is distributed
data is stored in topics
topics are split into partitions
which are replicated
kafka cluster
consumer
producerproducer
producer producer
consumer
consumer
consumer
consumer
consumer
producer
producer
terminology
11. Giant leap?
In fact, persistent replicated messaging is such a giant leap in messaging architecture that it may be worthwhile to point out a few side
effects:
a. Per-message acknowledgments have disappeared
b. ordered delivery
c. The problem of mismatched consumer speed has disappeared. A slow consumer can peacefully co-exist with a fast
consumer now
d. Need for difficult messaging semantics like delayed delivery, re-delivery etc. has disappeared. Now it is all up to the
consumer to read whatever message whenever - onus has shifted from broker to consumer
e. The holy grail of message delivery guarantee: at-least-once is the new reality - both Kafka and Azure Event Hub
provides this guarantee. You still have to make your consumers and downstream systems idempotent so that recovering
from a failure and processing the same message twice does not upset it too much, but hey - that has always been the
case
http://blogs.msdn.com/b/opensourcemsft/archive/2015/08/08/choose-between-azure-event-hub-and-kafka-_2d00_-what-
you-need-to-know.aspx
13. Top 5
1. no consideration of data on the
inside vs outside
2. schema not externally defined
3. same config for every
client/topic
4. 128 partitions as default config
5. running on 8 overloaded nodes
19. #javazone @spjelkavik @audunstrand
in the beginning ...
Architecture governance board decided to use RabbitMQ as message queue.
Kafka was installed for a proof of concept, after developers spotted it januar 2013.
20. #javazone @spjelkavik @audunstrand
2013 - POC
“High” volume
Stream of classified ads
Ad matching
Ad indexed
mod05
zk
kafka
mod07
zk
kafka
mod01
zk
kafka
mod03
zk
kafka
mod06
zk
kafka
mod08
zk
kafka
mod02
zk
kafka
mod04
zk
kafka
dc 1
dc 2
Version 0.8.1
4 partitions
common client
java library
thrift
21. #javazone @spjelkavik @audunstrand
2014 - Adoption and
complaining
low volume/ high
reliability
Ad Insert
Product Orchestration
Payment
Build Pipeline
click streams
mod05
zk
kafka
mod07
zk
kafka
mod01
zk
kafka
mod03
zk
kafka
mod06
zk
kafka
mod08
zk
kafka
mod02
zk
kafka
mod04
zk
kafka
dc 1
dc 2
Version 0.8.1
4 partitions
experimenting
with
configuration
common java
library
27. Pattern
Language
why is it a mistake
what is the consequence
what is the correct solution
what has finn.no done
28. Top 5
1. no consideration of data on the
inside vs outside
2. schema not externally defined
3. same config for every
client/topic
4. 128 partitions as default config
5. running on 8 overloaded nodes
31. #javazone @spjelkavik @audunstrand
what is the consequence
direct reads across services/domains is quite normal in legacy and/or enterprise
systems
coupling makes it hard to make changes
unknown and unwanted coupling has a cost
Kafka had no security per topic - you must add that yourself
32. #javazone @spjelkavik @audunstrand
what is the correct solution
Consider what is data on the inside, versus data on the outside
Convention for what is private data and what is public data
If you want to change your internal representation often, map it before publishing it
publicly (Anti corruption layer)
33. #javazone @spjelkavik @audunstrand
what has finn.no done
Decided on a naming convention (i.e Public.xyzzy) for public topics
Communicates the intention (contract)
35. #javazone @spjelkavik @audunstrand
why is it a mistake
data and code needs separate versioning strategies
version should be part of the data
defining schema in a java library makes it more difficult to access data from non-
jvm languages
very little discoverability of data, people chose other means to get their data
difficult to create tools
36. #javazone @spjelkavik @audunstrand
what is the consequence
development speed outside jvm has been slow
change of data needs coordinated deployment
no process for data versioning, like backwards compatibility checks
difficult to create tooling that needs to know data format, like data lake
and database sinks
37. #javazone @spjelkavik @audunstrand
what is the correct solution
confluent.io platform has a separate schema registry
apache avro
multiple compatibility settings and evolutions strategies
connect
more automatic tooling
38. #javazone @spjelkavik @audunstrand
what has finn.no done
still using java library, with schemas in builders
confluent platform 2.0 is planned for the next step, not (just) kafka 0.9
40. #javazone @spjelkavik @audunstrand
why is it a mistake
Historically - One Big Database with Expensive License
Database world - OLTP and OLAP
Changed with Open Source software and Cloud
Tried to simplify the developer's day with a single config
Kafka supports very high throughput and highly reliable
41. #javazone @spjelkavik @audunstrand
what is the consequence
Trade off between throughput and degree of reliability
With a single configuration - the last commit wins
Either high throughput, and risk of loss - or potentially too slow
42. #javazone @spjelkavik @audunstrand
what is the correct solution
Understand your use cases and their needs!
Use proper pr topic configuration
Consider splitting / isolation
43. #javazone @spjelkavik @audunstrand
Defaults that are quite reliable
Exposing configuration variables in the client
Ask the questions;
at least once delivery
ordering - if you partition, what must have strict ordering
99% delivery - is that good enough?
what level of throughput is needed
what has finn.no done
44. #javazone @spjelkavik @audunstrand
Configuration
Configuration for production
Partitions
Replicas (default.replication.factor)
Minimum ISR (min.insync.replicas)
Wait for acknowledge when producing messages (request.required.acks, block.on.buffer.full)
Retries
Leader election
Configuration for consumer
Number of threads
47. #javazone @spjelkavik @audunstrand
why is it a mistake
partitions are kafkas way of scaling consumers, 128 partitions can handle 128
consumer processes
in 0.8; clusters could not reduce the number of partitions without deleting data
highest number of consumers today is 20
48. #javazone @spjelkavik @audunstrand
what is the consequence
our 0.8 cluster was configured with 128 partitions as default, for all topics.
many partitions and many topics creates many datapoints that must be coordinated
zookeeper must coordinate all this
rebalance must balance all clients on all partitions
zookeeper and kafka went down (may 2015)
(500 topics * 128 partitions)
49. #javazone @spjelkavik @audunstrand
what is the correct solution
small number of partitions as default
increase number of partitions for selected topics
understand your use case
reduce length of transactions on consumer side
partitions per topic: max(t/p, t/c)
Max partitions on a broker 100 x brokers x replication factor => 1500 in our case
http://www.confluent.io/blog/how-to-choose-the-number-of-topicspartitions-in-a-kafka-cluster/
52. #javazone @spjelkavik @audunstrand
why is it a mistake
Kafka was set up by Ops for a proof of concept - not for hardened production use
By coincidence we had 8 nodes for kafka, the same 8 nodes for zookeeper
Zookeeper is dependent on a majority quorum, low latency between nodes
The 8 nodes were NOT dedicated - in fact - they were overloaded already
53. #javazone @spjelkavik @audunstrand
what is the consequence
Zookeeper recommends 3 nodes for normal usage, 5 for high, and any more is
questionable
More nodes leads to longer time for finding consensus, more communication
If we get a split between data centers, there will be 4 in each
You should not run Zk between data centers, due to latency and outage
possibilities
54. #javazone @spjelkavik @audunstrand
what is the correct solution
Have an odd number of Zookeeper nodes - preferrably 3, at most 5
Don’t cross data centers
Check the documentation before deploying serious production load
Don’t run a sensitive service (Zookeeper) on a server with 50 jvm-based services,
300% over committed on RAM
Watch GC times
55. #javazone @spjelkavik @audunstrand
what has finn.no done
dc 1
dc 2
broker05
zk
kafka
broker01
zk
kafka
broker03
zk
kafka
broker04
zk
kafka
broker02
zk
kafka
Version 0.8.2
5-20 partitions
multiple
configurations
58. “It's only because of
their stupidity that
they're able to be so
sure of themselves.”
― Franz Kafka, The
Trial
Audun Fauchald Strand
@audunstrand
Henning Spjelkavik
@spjelkavik
Q?
https://www.finn.no/apply-here
https://tech.finn.no
https://twitter.com/@FINN_tech
https://github.com/finn.no
http://www.schibsted.com/en/Career/
59. #javazone @spjelkavik @audunstrand
Runner up
Using pre-1.0 software
Have control of topic creation
Kafka is storage - treat it like one also ops-wise
Client side rebalancing
Commiting on all consumer threads, believing that you only commited on one
60. #javazone @spjelkavik @audunstrand
References / Further reading
Designing data intensive systems, Martin Kleppmann
Data on the inside - data on the outside, Pat Helland
The Confluent Blog, http://confluent.io/
Kafka - The definitive guide
This presentation, in English: http://www.confluent.io/blog/the-top-sessions-from-
kafka-summit-2016
www.finn.no/apply-here
jobs.schibsted.com/
tech.finn.no
twitter.com/@FINN_tech
github.com/finn.no