Network connectivity issues on AWS

General eFa discussion
Post Reply
ramiroblanco
Posts: 2
Joined: 25 Feb 2020 07:25

Network connectivity issues on AWS

Post by ramiroblanco »

Hello all,

I am trying to set up EFA on a Centos 7 instance on AWS. The instance works perfectly until I install EFA, after a few minutes the instance loses connectivity. Everything works perfectly till after I install EFA (tried creating the instance multiple times, left it running for days before installing EFA and there were no issues).

Can't find anything on the logs other that the connections dropping (eg an autossh proccess I use to connect to a remote box).

So I was wondering if anyone knows if EFA executes a network restart during one of the cronjobs or if there is anything else that might be messing with the network.

Thank you in advance.
henk
Posts: 517
Joined: 14 Dec 2015 22:16
Location: Netherlands
Contact:

Re: Network connectivity issues on AWS

Post by henk »

No clue about AWS, but you can check a few network things. When dhcp is enabled, before the efa-configure, check your ip settings.
If you configure efa, most likely you set a fixed IP. The efa configure option 4 will do this for you, assuming you enter the correct settings..
The network interface boot protocol will change from dhcp to static/none (check the /etc/sysconfig/network-scripts.bak for the previous dhcp version)
When setting a static IP, DNS and GATEWAY entries are added to your primary interface script.

/etc/sysconfig/network-scripts/ifcfg-eth1

Code: Select all

# Generated by parse-kickstart
DEVICE="eth1"
IPV6INIT="no"
BOOTPROTO="none"
UUID="blablabla"
ONBOOT="yes"
IPADDR="172.16.xxx.xxx"
NETMASK="255.255.0.0"
GATEWAY="172.16.xxx.xxx"
DNS1="127.0.0.1"
DNS2="::1"
The next thing you should check is

Code: Select all

/etc/resolv.conf
When using recursion,

Code: Select all

# Generated by NetworkManager
search yourdomain.xxx
nameserver 127.0.0.1
and make sure Unbound is running

Code: Select all

 systemctl status unbound
NetworkManager will update /etc/resolv.conf to reflect the nameservers provided by currently active connections
To prevent Network Manager to overwrite your resolv.conf changes, remove the DNS1, DNS2 or set NM_CONTROLLED="no" to the specific interface script. or set dns=none in NetworkManager.conf see man 5 NetworkManager.conf

You can check/reset the bootproto for the primary to dhcp in the efa configure menu to see the initial -working-settings
efaf-configure menu -> Option 4 _> Bootproto-> select dhcp

Code: Select all

You seem to have multiple network interfaces
Please select the interface you want to configure
The interface names on your machine are:
 - eth0
 - eth1
 - eth2
 - eth3

Enter the primary interface name:

[eFa] : eth1

Current IP settings for eth1 are:
 1) BOOTPROTO            :  none
 2) IPV6_AUTOCONF        :
 3) IP                   :  172.16.xxx.xxx
 4) Netmask              :  255.255.0.0
 5) Gateway              :  172.16.xxx.xxx
 6) Use IPv6 DNS         :  no
 7) IPv6 IP              :
 8) IPv6 Prefix          :
 9) IPv6 Gateway         :
10) DNS Recursion        :  ENABLED
11) Primary DNS          :  172.16.xx.xx
12) Secondary DNS        :
13) Hostname             :  efa4
14) Domain Name          :  yourdomain.com
Setting 1 will set bootproto to DHCP
“We are stuck with technology when what we really want is just stuff that works.” -Douglas Adams
Post Reply