-
Notifications
You must be signed in to change notification settings - Fork 1.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged EVPN VxLAN MH HLD from Cisco and BCM #1702
base: master
Are you sure you want to change the base?
Conversation
@adyeung please add me to the reviewers list? |
#### 1.2.1.2 L2 Unicast forwarding in EVPN MH | ||
The control-plane and forwarding-plane handling of MAC and local adjacency (ARP/ND) learnt on the Multihomed Ethernet-Segments significantly differs from the single homed Ethernet-Segments in the EVPN VxLAN network. | ||
|
||
Each VTEP advertises EVPN Type-1 (per ES/EAD) route per local Ethernet-Segment. These VTEPs also advertises EVPN Type-2 (MAC/IP) route per locally learn adjacency (ARP/ND). Remote VTEPs perform EVPN route resolution based on matching ESI value carried as part of these routes. For each Ethernet-Segment, an unique L2 Next-hop group (NHG) is formed that contains the participating VTEPs. All processed EVPN Type-2 routes are pointing to that L2 NHG on matching ESI. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could you please add a section to capture new BGP EVPN routes(Type-1 and Type-4) being supported?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Route-type 1 and 4 works as described in RFC7432. No need to add them here explicitely.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hello @pbrisset
I represent a team working on the EVPN Multihoming implementation and we found some comments to this PR.
Can you, please, check the comments? We think they are crucial to have the implementation.
Thank you
**EVPN_ETHERNET_SEGMENT** | ||
|
||
``` | ||
;New table | ||
;Specifies EVPN Ethernet Segment interface | ||
; | ||
; Status: stable | ||
key = EVPN_ETHERNET_SEGMENT|"ifname"if_id | ||
; field = value | ||
esi = "AUTO" or es_id | ||
; es_id is 10 byte colon (:) separated string when type = "TYPE_0_OPERATOR_CONFIGURED". | ||
; Otherwise, esi value should be "AUTO" falling back on ESI type 1 or 3. | ||
type = esi_type | ||
; esi_type should be string with one of the below values: | ||
; "TYPE_0_OPERATOR_CONFIGURED" for Type-0 ESI | ||
; "TYPE_1_LACP_BASED" for Type-1 ESI | ||
; "TYPE_3_MAC_BASED" for Type-3 ESI | ||
ifname = "ifname"if_id | ||
; if_id is the interface identifier. Same value as in the key. | ||
; e.g., PortChannel1, where the ifname is PortChannel and the if_id is 1. | ||
df_pref = 1*5DIGIT | ||
; Designated-Forwarder election preference for this router in (1..65535) range. | ||
; Default=32767. | ||
``` |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
EVPN_ETHERNET_SEGMENT table does not map on existing FRR commands and is missing required functionality: mac and id for type-3 ESI. Those 2 items are both mandatory in FRR for type-3 ES to work. Also to our knowledge there is no type-1 ESI support in FRR right now or in any PR. So even if everything else is ready to resolve ESI through LACP, FRR is not ready at all.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We dont care about ESI value in SONIC. FRR using the Ethernet Segment info programs needed DB in SONIC such as SHL, DF/NDF, etc.
curl -X PATCH "https://SWICH_IP:9090/restconf/data/openconfig-network-instance:network-instances/network-instance=default/evpn/evpn-mh/config" -H "accept: */*" -H "Content-Type: application/yang-data+json" -d "{\"openconfig-network-instance:config\":{\"startup-delay\":150,\"mac-holdtime\":200,\"neigh-holdtime\":250}}" | ||
``` | ||
<a id="4-Flow-Diagrams"></a> | ||
# 4 Flow Diagrams |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The Flow Diagrams section contains multiple distinctly different ways of to naming/implementing feature comparing to the rest of the document. We think these diagrams are outdated artifacts from #1638 requirements and contradict the new requirements.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
hummm. Can you be explicit?
### 3.3.8 L2nhgorch | ||
- New Orchagent class being introduced. | ||
- Subscription to different tables | ||
- APP DB - L2_NEXTHOP_GROUP, L2_NEXTHOP_GROUP_MEMBER | ||
- CONFIG DB - EVPN_ES_INTERFACE. | ||
- Subscription to L2_NEXTHOP_GROUP | ||
- Creates/deletes the L2_ECMP_GROUP ASIC DB table. | ||
- Interacts with PortsOrch to create a bridge-port object of type L2_ECMP_GROUP. | ||
- Subscription to L2_NEXTHOP_GROUP_MEMBER | ||
- Creates/deletes the L2_ECMP_GROUP_MEMBER ASIC DB tables entries on the L2_NEXTHOP_GROUP APP DB table updates. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The "DB Changes" and "Impact to existing modules" section conflicts with the L2_NEXTHOP_GROUP schema in APP_DB (the latter mentions a separate table for members). We propose that the one in "DB Changes" section is to be chosen as the correct approach and "L2nhgorch" subsection is to be rewritten accordingly.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Text was updated. You may want to re-review this section.
### 2.2.4 Linux Kernel Support | ||
EVPN Multihoming feature requires L2 next-hop group (NHG) support in the Linux kernel that is available in v5.10. | ||
EVPN Multihoming feature will not be supported in SONiC releases running 4.x kernel versions. | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
According to RFC 8365, Section 8.3.3
VXLAN-GPE encapsulation needs to be used between these PEs and the ingress PE needs to set the BUM Traffic Bit (B bit) to indicate that this is an ingress-replicated BUM traffic.
It might need an additional patch to support this in the v5.10??
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@AntonButenkoGL @pbrisset Does the kernel v5.10 need patch for this?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
SONiC is currently at Linux kernel v6.1.38.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is the BUM Traffic Bit already supported by Linux Kernel v6.1.38?
In the current implementation, can all the functionalities already work on the VS (virtual machine) platform? |
Yes, it was tested with VS platform. And there is a work in progress to align the ESI types logic with the current requirements. |
Updated kernel, SAI, config, and design sections
Update EVPN_VxLAN_Multihoming.md
;Specifies EVPN Ethernet Segment interface | ||
; | ||
; Status: stable | ||
key = EVPN_ETHERNET_SEGMENT|"ifname"if_id |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"ifname"if_id looks weird, can we just have an ifname as key and in the description, it's already mentioned about if_id. Same suggestion for other cases like "Vlan"vlan_id as well.
The Isolation group SAI object used for achieving Local Bias and Split Horizon Filtering in the MCLAG will be reused for EVPN MH. In EVPN MH an Isolation group for every Peer Multihoming VTEP is created as opposed to a single group for MCLAG. The Isolation group members are populated based on the EVPN_SPLIT_HORIZON_TABLE APP-DB entry. Traffic received from the Tunnel towards the peer multihoming VTEP will not be forwarded to the members of the Isolation group. | ||
|
||
<a id="224-Linux-Kernel-Support"></a> | ||
### 2.2.4 Linux Kernel Support |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can SONiC-VM work for MH feature any without limitations/restriction?
Configuration handling of MCLAG should now be aware of EVPN Ethernet-Segment configurations and throw appropriate errors in order to achieve mutual-exclusion between EVPN Multihoming and MCLAG features. | ||
OpenConfig evpn yang is imported into SONiC and ethernet-segment URIs are implemented. | ||
OpenConfig LACP yang already has system-id-mac configuration and it will be implemented. | ||
OpenConfig evpn yang will be extended for global non-standard parameter configurations (e.g. MAC/ARP hold-timer, etc.) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We dont have openconfig support for EVPN feature as such in the SONiC community, are we planning to bring that support via mgmt-framework as part of EVPN MH effort?
sonic community review recording on 7/30 https://zoom.us/rec/share/V2Xv_D2UOXdjLKJHazb1AV11JQxEsXS6275dtFiGA_QyGPgoVvWmL0lh9A5DJBfE.l-gvQEzam4TFRpXT |
##### 1.2.1.1.1 Local-bias and split-horizon filtering | ||
In multi-homed scenario, if the BUM traffic originates from a device attached to an EVPN Ethernet-Segment, local-bias procedure is applied and BUM traffic is flooded on all locally attached multi-homed and single-homed devices. The BUM traffic is also replicated to the remote VTEPs, and split-horizon filtering is applied such that traffic is not replicated to any of the devices that are attached to the shared EVPN Ethernet-Segment. | ||
|
||
Local-bias procedure is described in EVPN VxLAN Overlay [RFC 8365](https://datatracker.ietf.org/doc/html/rfc8365). The behavior of local-bias is explained with the help of below diagram. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
- What is the VXLAN encapsulation type used for BUM traffic indication ?
- Is it going to be VXLAN-GPE tunnel type ?
- If yes, is there any extension planned for sai_tunnel_type_t ?
This is the result of merging Cisco and BCM HLDs.
Cisco HLD
BCM HLD