• Uncategorized

About linux : kafka-jbod-disks-why-kafka-service-not-re-create-a-new-replica-when-one-disk-is-full-or-disk-fauilre

Question Detail

we have 3 Kafka brokers machines with version 2.7

each Kafka have 8 disks as the following

/dev/sdh                    1.7T  1.2T  511G  70% /var/data/topic-logs8
/dev/sde                    1.7T  891G  786G  54% /var/data/topic-logs5
/dev/sdc                    1.7T  1.1T  551G  68% /var/data/topic-data3
/dev/sdf                    1.7T  1.5T  161G  91% /var/data/topic-data6
/dev/sdg                    1.7T 1017G  660G  61% /var/data/topic-data7
/dev/sdd                    1.7T  1.7T  480K 100% /var/data/topic-data4
/dev/sda                    1.9T  1.6T  311G  84% /var/data/topic-data1

as we can see above one of the disks became 100%

since one of the disk is 100% , then all replica data on that disk isn’t reached

therefore we get the following behavior:

kafka-topics.sh  --zookeeper zoo_server:2181 --describe

Topic: mnior_mnol_frontend     Partition: 85   Leader: none    Replicas: 1012,1010,1011        Isr: 1011
Topic: mnior_mnol_frontend     Partition: 86   Leader: 1010    Replicas: 1010,1011,1012        Isr: 1010,1011,1012
Topic: mnior_mnol_frontend     Partition: 87   Leader: 1011    Replicas: 1011,1010,1012        Isr: 1010,1011,1012
Topic: mnior_mnol_frontend     Partition: 88   Leader: 1012    Replicas: 1012,1011,1010        Isr: 1010,1012
Topic: mnior_mnol_frontend     Partition: 89   Leader: 1011    Replicas: 1010,1012,1011        Isr: 1011
Topic: mnior_mnol_frontend     Partition: 90   Leader: 1011    Replicas: 1011,1012,1010        Isr: 1010,1011,1012
Topic: mnior_mnol_frontend     Partition: 91   Leader: 1012    Replicas: 1012,1010,1011        Isr: 1010,1011,1012
Topic: mnior_mnol_frontend     Partition: 92   Leader: 1010    Replicas: 1010,1011,1012        Isr: 1010,1011,1012
Topic: mnior_mnol_frontend     Partition: 93   Leader: 1011    Replicas: 1011,1010,1012        Isr: 1011

topic name – mnior_mnol_frontend
kafka brokers id’s – 1010,1011,1012
Isr – we have replica 3 , and we can see that Isr is missing due disk faulty

as we can see above and due disk failure , we have here missing replica and also some none leaders

actually , we was thinking that Kafka itself should create the missing replica on the other good ( that are less then 100% ) disks , but this inst happens

so I want to understand – if this is limitation of Kafka?

or maybe we can resolve the issue by Kafka settings ( for example by Kafka server.properties )

Question Answer

No answer for now.

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.