Bridge dropping 3-5% of packets on vmbr0 despite clean physical NIC - kernel 6.14.11

[bucker00]

System Info:

• Proxmox VE 9.0.11
• Kernel: 6.14.11-3-pve and 6.14.11-4-pve (both affected)
• Physical NIC: Intel e1000e (I219-V)
• Hardware: 12 CPU cores, 64GB RAM
• Workload: 2 LXC containers and 3 VMs

Problem:The bridge (vmbr0) is consistently dropping 3-5% of received packets, causing intermittent connection failures for services like MySQL and cloudflared. The drops occur with light network load (~40Kb/s) and low CPU usage (70%+ idle).


Key Findings:

• Physical NIC (eno1) shows zero packet drops
• Bridge (vmbr0) drops 3-5% of packets consistently
• Drops occur with multiple different NICs (tested e1000e and USB 2.5Gb)
• Very light traffic and low system load when drops occur
• Issue appears to be in bridge/kernel processing, not hardware

Diagnostics performed:

# Physical NIC - clean
ethtool -S eno1 | grep -v ": 0$"
# Only shows packet counters, zero errors/drops

# Bridge - dropping packets
ip -s link show vmbr0
# RX dropped: ~3-5% of total packets

# Traffic load - minimal
iftop -i vmbr0 -t -s 5
# Only ~40Kb/s total traffic when drops occur

Attempted fixes (none worked):

• Increased NIC ring buffers (ethtool -G eno1 rx 4096 tx 4096)
• Increased system network buffers (net.core.rmem_max/wmem_max to 16MB)
• Added fq_codel qdisc to bridge and veth interfaces
• Enabled RPS/RFS for packet steering
• Disabled hardware offloads on e1000e
• Tested with different physical NIC (USB 2.5Gb) - same issue
• Tested older kernel (6.14.11-3 vs 6.14.11-4) - same issue

Questions:

1. Is this a known issue with kernel 6.14.11 bridge implementation?
2. Has anyone else experienced bridge packet drops despite clean physical NICs?
3. Would downgrading to 6.8 kernel series be recommended?

The physical hardware appears fine - this seems to be a software issue in the bridge forwarding path.
Any guidance would be appreciated!