[
https://issues.apache.org/jira/browse/HDDS-14424?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Andika updated HDDS-14424:
-------------------------------
Description:
This is an epic is to implement features that allow client to read from OM
followers.
The purpose of reading from OM followers are
* Increase resource utilization : increase the overall read throughput to
overcome the read bottleneck of a single OM leader and increase overall read
throughput.
* Improve write throughput: Leader write throughput might also increase since
it does need to content with reads
* Reduce read latency (if possible): This can be achieved by reading from the
nearest follower and avoiding inter DC traffic
** If the client and OM leader is in different DC, we can read from OM follower
*** The read payloads for list and get might be big so we can reduce the
traffic cost
** This will be enhanced if we can support follower reads that do not need to
talk to the leader for each read request (e.g. through ReadIndex) by
sacrificing some consistency
This is not limited to linearizable reads. In the future, we can support weaker
consistency such as bounded staleness.
was:
This is epic is to implement features that allow client to read from OM
followers.
The purpose of reading from OM followers are
* Increase resource utilization : increase the overall read throughput to
overcome the read bottleneck of a single OM leader and increase overall read
throughput.
* Improve write throughput: Leader write throughput might also increase since
it does need to content with reads
* Reduce read latency (if possible): This can be achieved by reading from the
nearest follower and avoiding inter DC traffic
** If the client and OM leader is in different DC, we can read from OM follower
*** The read payloads for list and get might be big so we can reduce the
traffic cost
** This will be enhanced if we can support follower reads that do not need to
talk to the leader for each read request (e.g. through ReadIndex) by
sacrificing some consistency
This is not limited to linearizable reads. In the future, we can support weaker
consistency such as bounded staleness.
> OM HA: Support Follower Read
> ----------------------------
>
> Key: HDDS-14424
> URL: https://issues.apache.org/jira/browse/HDDS-14424
> Project: Apache Ozone
> Issue Type: Epic
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
>
> This is an epic is to implement features that allow client to read from OM
> followers.
> The purpose of reading from OM followers are
> * Increase resource utilization : increase the overall read throughput to
> overcome the read bottleneck of a single OM leader and increase overall read
> throughput.
> * Improve write throughput: Leader write throughput might also increase
> since it does need to content with reads
> * Reduce read latency (if possible): This can be achieved by reading from
> the nearest follower and avoiding inter DC traffic
> ** If the client and OM leader is in different DC, we can read from OM
> follower
> *** The read payloads for list and get might be big so we can reduce the
> traffic cost
> ** This will be enhanced if we can support follower reads that do not need
> to talk to the leader for each read request (e.g. through ReadIndex) by
> sacrificing some consistency
> This is not limited to linearizable reads. In the future, we can support
> weaker consistency such as bounded staleness.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
