Feature #63645: rbd: scalability issue on Windows due to TCP session count - rbd - Ceph

Actions

Copy link

Feature #63645

open

rbd: scalability issue on Windows due to TCP session count

Added by Lucian Petrut 5 months ago. Updated 2 months ago.

Status:

Pending Backport

Priority:

Normal

Assignee:

Lucian Petrut

Target version:

% Done:

Source:

Tags:

backport_processed

Backport:

squid

Reviewed:

Affected Versions:

Pull request ID:

52540

Description

rbd-wnbd doesn't currently reuse OSD connections across disk mappings, which can lead to an excessive amount of TCP sessions, exceeding Windows limits and eventually failing.

An user reported the following scenario [1]: 60 connected rbd volumes per Windows host, 13 ceph nodes with 24 OSDs each, resulting approximately 18720 sessions. This exceeds the Windows default limit (16384).

We can avoid this by having a single rbd-wnbd daemon per host and sharing OSD connections across mappings.

[1] https://ask.cloudbase.it/question/3598/ceph-for-windows-tcp-session-count/

Related issues 1 (0 open — 1 closed)