Great question—this is a common point of confusion because the role of the currently active slave depends entirely on the bonding mode you're using. Let's break this down clearly:

1. Active-Backup Mode (Mode 1) #

This is the scenario where you’re seeing the "only for failover" behavior. In this mode:

• The currently active slave is the single network interface handling all inbound and outbound traffic for the bond.
• All other slave interfaces are in a standby state—they don’t forward any traffic at all, and their only job is to wait for the active slave to fail.
• If the active slave loses link or stops responding, the bonding driver immediately switches one of the standby slaves to become the new currently active slave, and all traffic shifts to it. This is pure redundancy, no load balancing involved.

2. Load Balancing Modes (e.g., Mode 6: Adaptive Load Balancing) #

This is the mode you were expecting to see dynamic ARP manipulation and distributed traffic. Here, the currently active slave has a more nuanced role:

• First, all slave interfaces are active and participate in load balancing: Outbound traffic is split across slaves based on their current load (e.g., queue length, throughput), and inbound traffic is distributed by having the bonding driver send different slave MAC addresses to different clients via ARP responses (this is the "ARP manipulation" you thought of).
• The currently active slave acts as the bond’s "default representative": It handles initial ARP requests for the bond’s IP address, and sends out ARP announcements when the bond’s configuration changes. It’s not the only interface handling traffic—just the one responsible for ARP coordination.
• If a slave fails, the currently active slave (or the driver) will update ARP entries to redirect that slave’s traffic to other active slaves, and the load balancing logic will adjust to compensate for the lost interface. So it still provides failover, but alongside active load balancing.

Key Clarification #

Your initial understanding of adaptive load balancing using ARP to distribute traffic is correct—but that only applies to load-balancing modes. If you were reading documentation about active-backup mode, that’s why you got the "only failover" impression. The currently active slave’s role shifts entirely based on which mode you’ve configured.

内容的提问来源于stack exchange，提问作者iAdjunct