Task #7180
Updated by Nico Schottelius about 5 years ago
* router1-new is up and running
h2. Steps
* Verify / update cdist configuration
* Checkout IP configuration
** Add VRRP IPs to loopback interface (will be used instead of keepalived)
* Verify sysctl configuration
* Rerun cdist, exclude announcing of routes
* Adjust radvd configuration
** lower interval
** lower life time
h2. cdist configuration
Was written for Devuan/keepalived. Need to check step-by-step.
* __ungleich_bgp_router: *TEMPFIX*
** contains static route announcements
** can only be used once the upstream networking is verified
* __dcl_node_exporter: still makes sense
** Is debian/devuan specific, needs changes
* __ungleich_ipv6_radvd --config router-$dc
** 80% done
** restart needs to be implemented
* __ungleich_nftables --config router-$dc
* --__dcl_router_resolvconf --search-path "$d"--: done
** no changes required
* __dcl_tftp_http_pxe_bootserver
** ok, needs to be tested for alpine
** needs to be updated for ipv6 only boot
** could be moved to APUs
* -require="__dcl_router_network" __dcl_router_keepalived --master-: done
** Needs to be ported to loopback + bgp
** -commented out statement for place6 routers-: done
* __dcl_router_network --ipsuffix 5
** Need to add keepalived IPs
* # __dcl_nat64 --mtu 9000
h2. Networking
* Update interfaces for alpine
** -place6-
** place5
h2. Implementing virtual IPs with BGP
* Announce the IPv6/128 IPv4/32 address
** Or network alone is enough?
* -Add the addresses to the loopback interface-: done
** all done locally on router1-new
** Needs to be put back into cdist
* Add addresses for place5
h2. Firewalling
h2. Change router advertisements
We will have 2 active routers now with both advertising with radvd at the same time.
This is part of replacing keepalived.
* -Change router advertisement lifetime to 10 seconds-
* -Change advertisement interval to 3..5-
* Modify cdist type to restart/reload radvd on changes
<pre>
MinRtrAdvInterval 3;
MaxRtrAdvInterval 5;
AdvDefaultLifetime 10;
</pre>
* Good documentation: https://packetlife.net/blog/2011/apr/18/ipv6-neighbor-discovery-high-availability/
h2. IPv4 NAT session table
* With two active routers, the masquarading session information is only stored on one router
* If reply is received by other router, no session table match is found
* This only effects client devices
** VMs are using direct public IPv4
h2. Switch configuration
* Switches *might* need ipv4 bgp peering for providing virtual IPv4 address support
* Don't see traffic on router1-new from upstream
** Need to verify trunk configuration on switches
<pre>
switch5-place6(config)#vlan 100
switch5-place6(config-vlan-100)#name netstream
switch5-place6(config-vlan-100)#write
Copy completed successfully.
switch6-place6#configure
switch6-place6(config)#vlan 100
switch6-place6(config-vlan-100)#name netstream
switch6-place6(config-vlan-100)#write
Copy completed successfully.
</pre>
h2. Building jool
Requirements:
<pre>
apk upgrade
apk add alpine-sdk
apk add libnl3-dev
apk add iptables-dev
apk add linux-vanilla-dev
apk add argp-standalone
reboot
</pre>
Build:
<pre>
wget https://github.com/NICMx/Jool/releases/download/v4.0.1/jool_4.0.1.tar.gz
tar xvfz jool_4.0.1.tar.gz
cd jool-4.0.1/
./configure && make && make install
</pre>
h2. Missing routes from iBGP
* new router1.place6 does not receive full routing table from peers in place5
* Error "Invalid NEXT_HOP attribute" appears in the config file
** Good documentation:
*** https://blog.cybozu.io/entry/bird (use translate...)
** It seems bird1.6 uses the peer's ip address for the route while
** bird2 uses the "correct" next hop address inside the route
* Solution for bird2: "next hop self ebgp;"
h2. Consul
* Missing: correct init script: not compatible w/ alpine
* Started manually -> works
<pre>
[15:21] line:~% echo __dcl_consul_agent --datacenter place6 --server black1.place6.ungleich.ch --server black2.place6.ungleich.ch --server black3.place6.ungleich.ch | cdist config -i - -vv -c ~/vcs/ungleich-dot-cdist/ -j8 -p8 router1.place6.ungleich.ch
</pre>