When to use BGP address-family?

[jwbensley]

I commonly see BGP configured on a provider (P/PE) router like this (this is a real router snippet just anonymised);

router bgp 12345

neighbour 1.1.1.1 remote-as 12345     ! iBGP peer

address-family ipv4

neighbour 1.1.1.1 remote-as 12345     ! iBGP peer

address-family vpnv4

neighbour 1.1.1.1 remote-as 12345     ! iBGP peer

address-family ipv4 vrf Customer1

neighbour 10.0.0.1 remote-as 67890    ! eBGP peer

• I understand the CE peering defined under "address-family ipv4 vrf Customer1" because we are placing their routes into that VRF
• I understand that "address-family vpnv4" is used to define provider iBGP peers to which other CE's are going to be attached, carrying this customer VRF
• Finally, I understand that a provider router is an iBGP peer so it is defined directly under "router bgp 12345"

What is the need for "address-family ipv4" (without vrf) here, or ever?

What does it provies that "normal" "neighbour" definitions (those directly under "router bgp 12345") and "vpnv4 neighbour" definitions, don't provide?

[rookie R]

@Peter Paluch Thanks for your brilliant answer, I have some questions about BGP. Would you please answer those for me?
1. How do you tell if a BGP process is an ordinary BGP that doesn't support multiprotocol or an MP-BGP that supports multiprotocol? Does it mean that if there is another address-family other than ipv4 in the configuration is MP-BGP? If there is only address-family ipv4, then it's an ordinary BGP that doesn't support multiprotocol?

2. You said the 'no bgp default-ipv4 unicast' command prevents BGP from automatically assigning new neighbor into 'address-family ipv4' section, but why does this document said 

Use the no bgp default ipv4-unicast command if you use this neighbor for Multiprotocol Label Switching (MPLS) routes only. Here is the link to the document:

https://www.cisco.com/c/en/us/td/docs/ios-xml/ios/mp_l3_vpns/configuration/xe-3s/asr903/17-1-1/b-mpls-l3-vpns-xe-17-1-asr900/m_mp-bgp-mpls-vpn.html#GUID-BE5E2865-5D2B-45AE-830D-65CACC35320A

3. The output of the command show IP bgp all has several address families, the reason why the routes in the vpnv4 section is because ipv4 plus RD is vpnv4 routes, right?

4. You said, 'These routes that are placed into VRFs and can be exchanged with other PEs, along with additional attributes in BGP that allow these routes to be uniquely distinguished and imported into proper VRFs, including MPLS labels, are called "VPNv4 routes" in short for IPv4, and "VPNv6 routes" for IPv6'. But from what I know, the ipv4 prefix plus RD is the vpnv4 route. Does that mean the MPLS labels are on layer 2.5 to let the routers transmit the packet?