Bug #39038: beast: multiple v4 and v6 endpoints with the same port will cause failure - rgw - Ceph

Actions

Copy link

Bug #39038

closed

beast: multiple v4 and v6 endpoints with the same port will cause failure

Added by Abhishek Lekshmanan about 5 years ago. Updated almost 5 years ago.

Status:

Resolved

Priority:

Normal

Assignee:

Abhishek Lekshmanan

Target version:

% Done:

Source:

Tags:

Backport:

nautilus

Regression:

Severity:

3 - minor

Reviewed:

Affected Versions:

ceph-qa-suite:

Pull request ID:

27270

Crash signature (v1):

Crash signature (v2):

Description

currently the following combinations are known to fail:
rgw frontends = beast endpoint=[::]:8000 endpoint=0.0.0.0:8000
rgw frontends = beast port=8000 endpoint=[::]:8000
with something like:

2019-03-29 17:22:29.728 7f15db9e9900  4 frontend listening on 0.0.0.0:8000
2019-03-29 17:22:29.728 7f15db9e9900 -1 failed to bind address [::]:8000: Address already in use

As the systemflag for bindv6only (cat /proc/sys/net/ipv6/bindv6only to check) is usually true for many Linux systems, specifying both v4 and v6 addresses will mean that configuration has to be aware of the flag and set only v6 for eg. A better behaviour here would be to be explicit here and only bind to v4 or v6 addresses as configured so that configuration remains explicit.

Related issues 2 (0 open — 2 closed)

Actions

Copy link

Updated by Matt Benjamin about 5 years ago

@abhishek, is there a strong reason not to backport as far as luminous?

Matt

Actions

Copy link

Updated by Abhishek Lekshmanan about 5 years ago

Backport changed from nautilus to nautilus, mimic

for luminous the only thing was beast was an experimental feature, otherwise I can mark it for l as well, given that people upgrading from luminous -> nautilus then can get by without needing to change the configuration.

Actions

Copy link