Bug 1813976 - [RFE] Add support for overriding the read-from-replica policy
Summary: [RFE] Add support for overriding the read-from-replica policy
Keywords:
Status: CLOSED ERRATA
Alias: None
Product: Red Hat Ceph Storage
Classification: Red Hat Storage
Component: RBD
Version: 5.0
Hardware: Unspecified
OS: Unspecified
unspecified
medium
Target Milestone: ---
: 5.0
Assignee: Ilya Dryomov
QA Contact: Gopi
Amrita
URL:
Whiteboard:
Depends On:
Blocks: 1929671 1959686
TreeView+ depends on / blocked
 
Reported: 2020-03-16 15:47 UTC by Jason Dillaman
Modified: 2021-08-30 08:24 UTC (History)
10 users (show)

Fixed In Version: ceph-16.0.0-8633.el8cp
Doc Type: Enhancement
Doc Text:
.Overriding read-from-replica policy in librbd clients is supported Previously there was no way to limit the inter-DC/AZ network traffic, as when a cluster is stretched across data centers, the primary OSD may be on a higher latency and cost link in comparison with other OSDs in the PG. With this release, the `rbd_read_from_replica_policy` configuration option is now available and can be used to send reads to a random OSD or to the closest OSD in the PG, as defined by the CRUSH map and the client location in the CRUSH hierarchy. This can be done per-image, per-pool or globally. See the link:{block-dev-guide}#block-device-input-output-options_block[_Block device input and output options_] section in the _{storage-product} Block Device Guide_ for more information.
Clone Of:
Environment:
Last Closed: 2021-08-30 08:23:52 UTC
Embargoed:

Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Red Hat Issue Tracker RHCEPH-1139 0 None None None 2021-08-30 00:12:54 UTC
Red Hat Product Errata RHBA-2021:3294 0 None None None 2021-08-30 08:24:33 UTC

Description Jason Dillaman 2020-03-16 15:47:08 UTC 
librbd-based clients can now set the 'rbd_read_from_replica_policy' configuration option to "default" (i.e. read from the PG's primary OSD), "balance" (send the read to a random OSD), or "localize" (send to the closest OSD as defined by the CRUSH map and the librbd client's "crush_location" config option). The RBD configuration option can be set globally, per-pool, or per-image. The "crush_location" option should be set via "ceph.conf" on a per-node basis.

This feature is useful for stretch clusters where the PG's primary OSD might be across a higher-cost link as compared to other OSDs in the PG.
Comment 1 RHEL Program Management 2020-03-16 15:47:13 UTC 
Please specify the severity of this bug. Severity is defined here:
https://bugzilla.redhat.com/page.cgi?id=fields.html#bug_severity.
Comment 2 Ben England 2020-12-02 21:36:58 UTC 
This feature is also very useful for OCS in public clouds such as AWS, where by default OCS PGs are spread across "availability zones" (AZs) with higher latency between AZs than within them.
Comment 5 Gopi 2021-03-05 05:05:09 UTC 
Feature is working as expected, hence moving this bug to verified state.
Comment 9 Amrita 2021-06-09 06:51:37 UTC 
LGTM Ilya , 
I just added the `previously...` before `With this rel` to follow our doc standards.
Comment 11 errata-xmlrpc 2021-08-30 08:23:52 UTC 
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (Red Hat Ceph Storage 5.0 bug fix and enhancement), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2021:3294
Note You need to log in before you can comment on or make changes to this bug.