1203776 - Reconfigure bonding mode on git1.dmz.scl3 to mode=0 balance-rr

Status: RESOLVED WONTFIX

(Infrastructure & Operations Graveyard :: NetOps, task)

Description

[Gregory Szorc [:gps]]

bond0 on git1.dmz.scl3 (set via /etc/modprobe.d/local.conf) is currently set to mode=1, which is "active backup" mode. In this mode, the secondary network interface is only used if the primary has issues.

git1.dmz.scl3 semi-frequently is network bound and saturates its eth0 1 Gbps interface. Since we have 2 network interfaces already bonded, I was thinking we could help alleviate the local network interface as the choke point by switching the bonding mode to actively use both interfaces, giving us 2 Gbps.

I'm filing this bug against Netops because I figure someone will want to verify that whatever network gear eth0 and eth1 are connected to won't be overloaded if we start sending >1Gbps to them.

git1 overload has been responsible for a number of Firefox tree closures lately. So having a quick resolution to this bug (whatever the outcome) would be appreciated.

Comment 1

[Van Le [:van]]

:gps, this host is a blade and the chassis wont be overloaded. we'll need to configure port channeling[1] on the cisco switch side as well, so it'll cause some downtime. can you open a CAB for this?

[1]http://www.cisco.com/c/en/us/td/docs/ios/12_2sb/feature/guide/sbcelacp.html

Comment 2

[Gregory Szorc [:gps]]

It's been 7 months and I don't really have an intention of following through with this. The fact that git.mo hasn't blown up in the past few months reinforces my laziness.

Should we resolve this "meh?"

Comment 3

[Kendall Libby [:fubar] (he/him)]

between adding gitweb[123] and the de-prioritization of b2g, I'm ok with that.