RDS/IB Re-connection And Stalling for More Than 1 Min (Doc ID 2611709.1)

Last updated on NOVEMBER 29, 2019

Applies to:

Symptoms

RDS/IB connection issue causes IO timeout on the voting disk.

During the time of the issue, the following patterns of the error events can be seen in the Linux system log.
- stalling for more than 1 min
- dropped due to 'incoming REQ in CONN_UP state'
- dropped due to 'DISCONNECTED event'

# cat messages
Jan 1 00:00:40 hostname kernel: [12751633.853935] RDS/IB: connection <::ffff:192.168.1.143,::ffff:192.168.0.7,0> dropped due to 'DISCONNECTED event'
Jan 1 00:00:45 hostname kernel: [12751698.035675] RDS/IB: connection <::ffff:192.168.0.3,::ffff:192.168.1.144,0> dropped due to 'incoming REQ in CONN_UP state'
Jan 1 00:00:45 hostname kernel: [12751698.063546] RDS/IB: Passive conn ffff77ffffffffff i_cm_id ffff88ffffffffff, frag 16KB, connected <::ffff:192.168.0.3,::ffff:192.168.1.144,0> version 4.1
Jan 1 00:00:48 hostname kernel: [12751701.033437] RDS/IB: connection <::ffff:192.168.0.3,::ffff:192.168.1.142,0> dropped due to 'incoming REQ in CONN_UP state'
Jan 1 00:00:48 hostname kernel: [12751701.033771] RDS/IB: re-connect <::ffff:192.168.1.142,::ffff:192.168.0.3,0> stalling for more than 1 min...(drops=1 err=0)
Jan 1 00:00:48 hostname kernel: [12751701.035083] RDS/IB: re-connect <::ffff:192.168.1.142,::ffff:192.168.0.1,0> stalling for more than 1 min...(drops=1 err=0)

Check the model and the version of the Infiniband interface card below.

# lspci
03:00.0 InfiniBand: Mellanox Technologies MT27500 Family [ConnectX-3]

# dcli -l root -g dbs_group ibstat |grep Firm
uxdxdb01: Firmware version: 2.35.5532

Cause

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.