LAGs are not updated correctly after teamd restart

[marian-pritsak]

Reproduced on:

build_version: HEAD.211-5585221
debian_version: 8.7
kernel_version: 3.16.0-4-amd64
asic_type: mellanox
build_date: Wed Apr 12 05:07:40 UTC 2017
built_by: johnar@jenkins-worker-1

Steps to reproduce:

$ systemctl restart teamd

After I restart teamd, some LAGs are not working anymore. From SDK dump I can see that part of them have no members on hardware, some have only one, and some have two.

[marian-pritsak]

127.0.0.1:6379> keys LAG_TABLE*
 1) "LAG_TABLE:PortChannel56:Ethernet60"
 2) "LAG_TABLE:PortChannel16:Ethernet16"
 3) "LAG_TABLE:PortChannel0"
 4) "LAG_TABLE:PortChannel24:Ethernet28"
 5) "LAG_TABLE:PortChannel24"
 6) "LAG_TABLE:PortChannel48:Ethernet48"
 7) "LAG_TABLE:PortChannel24:Ethernet24"
 8) "LAG_TABLE:PortChannel32:Ethernet32"
 9) "LAG_TABLE:PortChannel0:Ethernet0"
10) "LAG_TABLE:PortChannel40:Ethernet40"
11) "LAG_TABLE:PortChannel8:Ethernet12"
12) "LAG_TABLE:PortChannel0:Ethernet4"
13) "LAG_TABLE:PortChannel8:Ethernet8"
14) "LAG_TABLE:PortChannel40:Ethernet44"
15) "LAG_TABLE:PortChannel48"
16) "LAG_TABLE:PortChannel32:Ethernet36"
17) "LAG_TABLE:PortChannel8"
18) "LAG_TABLE:PortChannel16:Ethernet20"
19) "LAG_TABLE:PortChannel56:Ethernet56"
20) "LAG_TABLE:PortChannel32"
21) "LAG_TABLE:PortChannel16"
22) "LAG_TABLE:PortChannel40"
23) "LAG_TABLE:PortChannel48:Ethernet52"
24) "LAG_TABLE:PortChannel56"

[marian-pritsak]

127.0.0.1:6379> hget LAG_TABLE:PortChannel56:Ethernet0 status
(nil)
127.0.0.1:6379> hget LAG_TABLE:PortChannel56:Ethernet8 status
(nil)
127.0.0.1:6379> hget LAG_TABLE:PortChannel56:Ethernet12 status
(nil)

[stcheng]

Hello @marian-pritsak , i cannot reproduce this issue:

127.0.0.1:6379> hgetall LAG_MEMBER_TABLE:PortChannel04:Ethernet124
1) "status"                                                       
2) "enabled"                                                      

this is get after docker-teamd restart.

[stcheng]

i think on your case, it is because the docker-teamd restart causes the LAG fails to be brought up. And that is the reason (nil) appears in the redis database.