scripting around it, it can also run unattended when upscaling.
You can find the newborn tool here:
https://github.com/dba-git/kafka_smart_rebalancer/
Feedback and contributions are welcome!
Regards,
Fabio Pardi
Agileos Consulting LTD
https://www.agileosconsulting.com/
a lot of data movement.
Is there a tool around which generates a new partitions distribution minimizing
the number of changes?
Something written in bash would be the best of life, suitable to run on vanilla
K8s Kafka images too.
regards,
fabio pardi
Is it possible to have a separate Kraft controller from the brokers? Is there
any documentation on setting this up?
Hi,
you should be able to achieve such setup running kafka in 'controller' mode.
https://kafka.apache.org/documentation/#brokerconfigs_process.roles
regards,
fabio
replicated
kafka-topics.sh --describe
I think it might be a good starting point to understand what is going on.
To blindly reassign partitions is in my experience not an ideal solution,
because you will have data shuffling around unnecessarily.
regards,
fabio pardi
metadata is stored next to each message. you can see it running
kafka-console-consumer.sh and '--property'.
eg, to see the the timestamp of a message:
kafka-console-consumer --bootstrap-server your_kafka:9092 --from-beginning
--topic topic_name --property print.timestamp=true
regards,
fabio pardi
hi,
for the ones of you working with medical data, here is how FHIR STU3 data
responds to different compression options in kafka:
https://portavita.github.io/2021-01-25-Why_Kafka_compression_might_save_you_thousands_of_dollars/
ideas and comments are welcome
regards,
fabio pardi
are of course welcome.
Regards,
fabio pardi
s.
>
> If the produce request fails, what does the partition leader do with the
> records it has written to the local log. Are they deleted, or will the
> producer's retry cause duplication?
>
Hi,
the record is not committed to the filesystem and an error is returned to the
producer.
regards,
fabio pardi
hi Mathieu,
the best approach in my opinion is to try to understand where your bottleneck
is, analyzing the graphs produced during history reprocessing.
my best bet are the disks, but indeed it might be anywhere.
regards,
fabio pardi
On 23/10/2020 20:07, Mathieu D wrote:
> He
Solved.
For the future us: the reason why offsets are 2 times the messages is to be
found in how (our) producer works.
The producer commits the message and the transaction, thus the offset is
incremented by 2 for each sent message.
regards,
fabio pardi
On 08/06/2020 13:45, Fabio Pardi wrote
suspicious besides the offset and the number of messages not
being identical, is that the former is exactly 2 times the latter.
regards,
fabio pardi
On 08/06/2020 12:26, Liam Clarke-Hutchinson wrote:
> Hi Fabio,
>
> -1 is shorthand for latest when passed as --time to GetOffsetShel
5.4.1-ccs (Commit:fd1e543386b47352)
kafka-run-class -version
openjdk version "1.8.0_212"
OpenJDK Runtime Environment (Zulu 8.38.0.13-CA-linux64) (build 1.8.0_212-b04)
OpenJDK 64-Bit Server VM (Zulu 8.38.0.13-CA-linux64) (build 25.212-b04, mixed
mode)
regards,
fabio pardi
12 matches
Mail list logo