Bug #57563: haproxy requires the `net.ipv4.ip_nonlocal_bind` sysctl setting - Orchestrator - Ceph

Actions

Copy link

Bug #57563

closed

haproxy requires the `net.ipv4.ip_nonlocal_bind` sysctl setting

Added by Michael Fritch over 1 year ago. Updated about 1 year ago.

Status:

Resolved

Priority:

Normal

Assignee:

Michael Fritch

Category:

cephadm (binary)

Target version:

% Done:

Source:

Tags:

backport_processed

Backport:

quincy, pacific

Regression:

Severity:

3 - minor

Reviewed:

Affected Versions:

ceph-qa-suite:

Pull request ID:

48120

Crash signature (v1):

Crash signature (v2):

Description

haproxy requires the `net.ipv4.ip_nonlocal_bind` sysctl setting otherwise it cannot bind to the configured ingress VIP:

Sep 14 16:17:27 host3 systemd[1]: Started Ceph haproxy.nfs.ha.host3.lrxbea for a2e6fbda-e4eb-4558-8c3c-f5bc91dea147.
Sep 14 16:17:27 host3 conmon[29241]: [NOTICE] 256/221727 (7) : haproxy version is 2.3.17-d1c9119
Sep 14 16:17:27 host3 conmon[29241]: [NOTICE] 256/221727 (7) : path to executable is /usr/local/sbin/haproxy
Sep 14 16:17:27 host3 conmon[29241]: [ALERT] 256/221727 (7) : Starting frontend stats: cannot bind socket (Cannot assign requested address) [192.168.20.77:9049]
Sep 14 16:17:27 host3 conmon[29241]: [ALERT] 256/221727 (7) : Starting frontend frontend: cannot bind socket (Cannot assign requested address) [192.168.20.77:2049]
Sep 14 16:17:27 host3 conmon[29241]: [ALERT] 256/221727 (7) : [haproxy.main()] Some protocols failed to start their listeners! Exiting.

This happens to work in most instances where typically both haproxy and keepalived are colocated and this setting is set by the keepalived deployment.

Related issues 2 (0 open — 2 closed)