Skip to content
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.

Sign up for GitHub

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

Jump to bottom

EVPN VxLAN update for platforms using P2MP tunnel based L2 forwarding #806

Merged
merged 18 commits into from
Nov 10, 2021
Merged

EVPN VxLAN update for platforms using P2MP tunnel based L2 forwarding #806

Show file tree
Hide file tree
Changes from 14 commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
3220491
Add files via upload
dgsudharsan May 27, 2021
2470cea
Update EVPN_VXLAN_HLD.md
dgsudharsan May 27, 2021
ebd5c97
Update EVPN_VXLAN_HLD.md
dgsudharsan May 27, 2021
8611ebf
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 5, 2021
e33233f
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 5, 2021
9e11684
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 5, 2021
6739f66
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 5, 2021
3d35f88
Updating VLAN flow
dgsudharsan Jun 15, 2021
dd2be56
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 15, 2021
ab13816
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 16, 2021
eeba0d4
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 18, 2021
9b177f5
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 21, 2021
1715455
Update EVPN_VXLAN_HLD.md
dgsudharsan Jun 29, 2021
54ab573
Update EVPN_VXLAN_HLD.md
dgsudharsan Jul 10, 2021
cc0e8c8
Update EVPN_VXLAN_HLD.md
dgsudharsan Aug 3, 2021
7107e00
Update EVPN_VXLAN_HLD.md
dgsudharsan Aug 4, 2021
17583e3
Update EVPN_VXLAN_HLD.md
dgsudharsan Aug 4, 2021
21d2ae4
Update EVPN_VXLAN_HLD.md
dgsudharsan Nov 10, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
59 changes: 50 additions & 9 deletions doc/vxlan/EVPN/EVPN_VXLAN_HLD.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@

# EVPN VXLAN HLD

#### Rev 0.9
#### Rev 1.0

# Table of Contents

Expand All @@ -28,7 +28,11 @@
- [COUNTER_DB](#counter_db-changes)
- [4.3 Modules Design and Flows](#43-modules-design-and-flows)
- [4.3.1 Tunnel Creation](#431-tunnel-auto-discovery-and-creation)
- [4.3.1.1 P2P Tunnel Creation](#4311-p2p-tunnel-creation)
- [4.3.1.2 P2MP Tunnel Creation](#4312-p2mp-tunnel-creation)
- [4.3.2 Tunnel Deletion](#432-tunnel-deletion)
- [4.3.2.1 P2P Tunnel Deletion](#4321-p2p-tunnel-deletion)
- [4.3.2.2 P2MP Tunnel Deletion](#4322-p2mp-tunnel-deletion)
- [4.3.3 Mapper Handling](#433-per-tunnel-mapper-handling)
- [4.3.4 VXLAN State DB Changes](#434-vxlan-state-db-changes)
- [4.3.5 Tunnel ECMP](#435-support-for-tunnel-ecmp)
Expand Down Expand Up @@ -69,6 +73,7 @@
| 0.7 | | Rajesh Sankaran | Click and SONiC CLI added |
| 0.8 | | Hasan Naqvi | Linux kernel section and fdbsyncd testcases added |
| 0.9 | | Nikhil Kelhapure | Warm Reboot Section added |
| 1.0 | | Sudharsan D.G | Using P2MP Tunnel for Layer2 functionality |

# Definition/Abbreviation

dgsudharsan marked this conversation as resolved.
Show resolved Hide resolved
Expand Down Expand Up @@ -623,6 +628,9 @@ In the current implementation, Tunnel Creation handling in the VxlanMgr and Vxla
The VTEP is represented by a VxlanTunnel Object created as above with the DIP as 0.0.0.0 and
SAI object type as TUNNEL. This SAI object is P2MP.

Some vendors support P2P Tunnels to handle Layer2 extension and fdb learning while some vendors support using existing P2MP for Layer2 scenarios. In order to differentiate the different requirements evpn_remote_vni orch which currently handles remote VNI is split into two types - evpn_remote_vni_p2p to handle the flow involving the P2P tunnel creation and evpn_remote_vni_p2mp to handle the flow for using the existing P2MP tunnel. The decision to chose which orch to use is dependent on the SAI enum query capability for the attribute SAI_TUNNEL_ATTR_PEER_MODE. If the vendors have SAI_TUNNEL_PEER_MODE_P2P listed, then evpn_remote_vni_p2p orch will be used, else evpn_remote_vni_p2mp will be used.
Copy link
Collaborator

@venkatmahalingam venkatmahalingam Jul 20, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Add some details on P2P vs P2MP here.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done


#### 4.3.1.1 P2P Tunnel creation
In this feature enhancement, the following events result in remote VTEP discovery and trigger tunnel creation. These tunnels are referred to as dynamic tunnels and are P2P.

- IMET route rx
Expand All @@ -643,10 +651,15 @@ For every dynamic tunnel discovered, the following processing occurs.
The creation sequence assuming only IMET rx is depicted in the diagram below.

![Tunnel Creation](images/tunnelcreate.PNG "Figure : Tunnel Creation")
__Figure 5: EVPN Tunnel Creation__
__Figure 5.1: EVPN P2P Tunnel Creation__

### 4.3.2 Tunnel Deletion
#### 4.3.1.2 P2MP Tunnel Creation
In the current implementation P2MP tunnel creation flow exist with the exception of a bridgeport not created for P2MP tunnel. To support using P2MP tunnel for L2 purposes a bridge port is created for the P2MP tunnel object.
![P2MP Tunnel Creation](images/p2mptunnelcreate.jpg "Figure : P2MP Tunnel Creation")
__Figure 5.2: EVPN P2MP Tunnel Creation__

### 4.3.2 Tunnel Deletion
#### 4.3.2.1 P2P Tunnel Deletion
EVPN Tunnel Deletion happens when the refcnt goes down to zero. So depending on the last route being deleted (IMET, MAC or IP prefix) the tunnel is deleted.

sai_tunnel_api remove calls are incompletely handled in the current implementation.
Expand All @@ -656,6 +669,9 @@ The following will be added as part of tunnel deletion.
- sai_tunnel_remove_map, sai_tunnel_remove_tunnel_termination, sai_tunnel_remove_tunnel when the tunnel is to be removed on account of the last entry being removed.
- VxlanTunnel object will be deleted.

#### 4.3.2.2 P2MP Tunnel Deletion
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. The tunnel object can be deleted only after all the IMR, MAC, IP Prefix routes are removed by BGP.
    The bridge port object can be deleted only after the fdb count referencing the bridgeport goes down to 0.

  2. Once the map count becomes 0, the vxlanmgr holds off handling further Map creations till the number of P2P
    tunnels goes down to 0. This is to handle quick delete and readd of mapping entries.
    VxlanMgr gets to know of P2P tunnel count by looking at the number of state table entries.
    However for P2MP the state table entries are not populated..
    How is this scenario handled ?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There are no changes to P2MP tunnel logic currently except int this case it will always have 0 dip tunnels. Rest of the logic is the same as in existing code. I did have a question on FDB reference. In current orchagent code i don't not see tunnel delete skipped based on fdb reference count. Only bridge port removal is skipped while the actual tunnel might be deleted even when FDB reference count is non zero. Is that by design? If that's the case isn't it an issue with bridge port still referencing the tunnel port? https://github.com/Azure/sonic-swss/blob/4f1d726d4cbf8a283b22cd5f612cf03ca21a27b3/orchagent/vxlanorch.cpp#L1499

Copy link
Contributor Author

@dgsudharsan dgsudharsan Jul 14, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Few more questions and clarifications.

  1. I don't see TUNNEL_USER_MAC being used anywhere. Is it same as fdb count?
  2. I believe IP prefix routes use the P2MP tunnel rather than P2P tunnel but their references are stored in P2P tunnels. In this scenario since there are no P2P tunnels, I believe i should have route references on P2MP tunnel itself. Am I right here?
  3. When IMR is received the tunnel's bridge port will be added to L2MC group member. So maintaining a reference here and deleting the bridge port and tunnel only when IMR count reaches zero. This will be identical to DIP tunnel reference used. Is this fine?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. m_fdb_count is not the same as the evpn mac count. However currently the evpn mac ref cnt is not being used.
  2. The refcnts were maintained for P2P tunnels to create and delete the P2P tunnels. The DIP tunnel count was used for deciding when to delete the P2MP tunnel.
  3. For the P2MP only case the refcnts could be reused to decide when to delete the P2MP tunnel object. Please check for del_tnl_hw_pending and getDipTunnelCnt and deletePendingSIPTunnel. On the last map entry deletion the P2MP tunnel object will be deleted once the total of refcnts across all the remote VTEP IP becomes zero.
  4. For bridgeport deletion.. the Port object and Associated bridgeport object (P2P) is deleted only when the total fdb count goes down to 0. It is after the bridge port deletion do we delete the Tunnel Object as well.
  5. My comment was that these are applicable to P2MP scenarios also and needs to be mentioned in the HLD. Code wise we may have to do some more adjustments to reuse the existing implementation for P2MP as well.
  6. Regarding the L2MC group member yes using the existing IMR refcnt to decide when to add/remove the L2MC group members would be good.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Regarding your query https://github.com/Azure/SONiC/pull/806/files#r669059056 , No the bridgeport is deleted first and then the tunnel object which is being referenced by the bridgeport.
As part of deleteDynamicDIPTunnel there is a call to getTunnelPort. If this exists then the Tunnel Object is not deleted.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@dgsudharsan @srj102 Currently when tunnel nexthops are created by orchagent, it doesn't check for reachability of remote destination. Are we following the same approach or add the reachability condition while creating the the nexthops?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think we are enhancing that here. The existing approach still holds true.

In case of P2MP tunnels, the flow is same as the existing flow where the tunnel is deleted after last vxlan-vni map or vrf-vni map is deleted. Additionally before the tunnel deletion, the bridge port created is deleted.

### 4.3.3 Per Tunnel Mapper handling

The SAI Tunnel interface requires encap and decap mapper id to be specified along with every sai tunnel create call.
Expand Down Expand Up @@ -698,6 +714,7 @@ It is proposed to handle these variances in the SAI implementation.

### 4.3.6 IMET route handling
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are there any changes required for MAC route and MAC Move handling?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In case of P2MP tunnel there will be only one bridge port and thus there are some minor changes in code where instead of checking bridgeport alone fur duplicate, endpoint IP is also checked.


#### 4.3.6.1 P2P Tunnel Vlan extension
The IMET route is used in EVPN to specify how BUM traffic is to be handled. This feature enhancement supports only ingress replication as the method to originate BUM traffic.

The VLAN, Remote IP and VNI to be used is encoded in the IMET route.
Expand All @@ -707,7 +724,15 @@ The VLAN, Remote IP and VNI to be used is encoded in the IMET route.
The IMET rx processing sequence is depicted in the diagram below.

![Vlan extension](images/vlanextend.PNG "Figure : VLAN Extension")
__Figure 6: IMET route processing VLAN extension__
__Figure 6.1: IMET route processing P2P Tunnel VLAN extension__

#### 4.3.6.2 P2MP Tunnel Vlan extension

Similar to P2P tunnel scenario, the feature supports only the ingress replication. However the remote end points are added to VLAN as follows. In SONiC VLAN is created currently using SAI_VLAN_FLOOD_CONTROL_TYPE_ALL(default). To support flooding in P2MP based tunnels, the VLAN's flood control type is set to SAI_VLAN_FLOOD_CONTROL_TYPE_COMBINED which would support flooding to local ports as well as an additional multicast group. When type 2 prefixs are received, the remote end points are added to VLAN by creating a L2MC group and setting it to VLAN created in combined mode, and adding one L2MC group member per remote end point as shown in the flow below

![P2MP Vlan extension](images/p2mpvlanextension.jpg "Figure : P2MP VLAN Extension")
__Figure 6.2: IMET route processing P2MP TunnelVLAN extension__


##### FRR processing
When remote IMET route is received, fdbsyncd will install entry in REMOTE_VNI_TABLE in APP_DB:
Expand Down Expand Up @@ -1082,6 +1107,7 @@ Linux kernel version 4.9.x used in SONiC requires backport of a few patches to s
4. show vxlan tunnel
- lists all the discovered tunnels.
- SIP, DIP, Creation Source, OperStatus are the columns.
- This command is not supported when the platform supports only P2MP tunnels.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why is this command not supported for P2MP case? this is important command to show the tunnel status.

Copy link
Contributor Author

@dgsudharsan dgsudharsan Aug 3, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@venkatmahalingam I don't think the command's output signifies the actual tunnel status, meaning we don't perform any kind of tunnel status /monitoring or checking reachability at both ends. These tunnels are logically created within the device and status signifies just receiving the remote end point. I believe With P2MP model we have only one tunnel. What i can do is consider introducing an enhancement using which end points are printed when displaying the P2MP tunnel.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ok, maybe, you can remove the lines you have added for P2MP, as nothing is going to be changed in the output for P2MP case.

Do we have to display P2P/P2MP in any cmd output?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Venkat. I believe the HLD is not inline with the actual implementation. This command's output maps to the implementation of "show vxlan remotevtep" https://github.com/Azure/sonic-utilities/blob/c2ac2d2327acc90a35b33ddead070995034d457a/show/vxlan.py#L176
This indication of remote tunnels will not be support in P2MP based scenarios since internally we create only one tunnel and not multiple tunnels. I will remove the line here and will try to update the command "show vxlan remotevtep" for P2MP scenario.


+---------+---------+-------------------+--------------+
| SIP | DIP | Creation Source | OperStatus |
Expand Down Expand Up @@ -1147,7 +1173,14 @@ Linux kernel version 4.9.x used in SONiC requires backport of a few patches to s
+---------+--------------+-------+
Total count : 1

7. show vxlan counters
+--------+------------+----------+------------+----------+
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We need to add more details on how counters are designed for P2P (in the future ?) and P2MP case.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

| Tunnel | RX_PACKETS | RX_BYTES | TX_PACKETS | TX_BYTES |
+========+============+==========+============+==========+
| Vtep1 | 1234 | 1512034 | 2213 | 2235235 |
+--------+------------+----------+------------+----------+


```

### 5.2 KLISH CLI
Expand Down Expand Up @@ -1385,18 +1418,26 @@ To support warm boot, all the sai_objects must be uniquely identifiable based on
- Verify that there is a SAI_OBJECT_TYPE_BRIDGE_PORT pointing to the above created P2P tunnel.
- Verify that there is a SAI_OBJECT_TYPE_VLAN_MEMBER entry for the vlan corresponding to the VNI created and pointing to the above bridge port.
7. Add more REMOTE_VNI table entries to different Remote IP.
- Verify that additional SAI_OBJECT_TYPE_TUNNEL, BRIDGEPORT and VLAN_MEMBER objects are created.
- Verify that additional SAI_OBJECT_TYPE_TUNNEL, BRIDGEPORT and VLAN_MEMBER objects are created in case of platforms that create dynamic P2P tunnels on type 3 routes.
- Verify that vlan flood type is set to SAI_VLAN_FLOOD_CONTROL_TYPE_COMBINED. Verify that L2MC group is created and SAI_OBJECT_TYPE_L2MC_GROUP_MEMBER with end point IP and P2MP bridge port is created and set in vlan's unknown unicast and broadcast flood group in case of platforms that use P2MP tunnel on type 3 routes.
8. Add more REMOTE_VNI table entries to the same Remote IP.
- Verify that additional SAI_OBJECT_TYPE_VLAN_MEMBER entries are created pointing to the already created BRIDGEPORT object per remote ip.
9. Remove the additional entries created above and verify that the created VLAN_MEMBER entries are deleted.
10. Remove the last REMOTE_VNI entry for a DIP and verify that the created VLAN_MEMBER, TUNNEL, BRIDGEPORT ports are deleted.
- Verify that additional SAI_OBJECT_TYPE_VLAN_MEMBER entries are created pointing to the already created BRIDGEPORT object per remote ip in case of platforms that create dynamic P2P tunnels on type 3 routes.
- Verify that additional SAI_OBJECT_TYPE_L2MC_GROUP_MEMBER entries are created per remote ip with P2MP bridge port in case of platforms that use P2MP tunnel on type 3 routes.
9. Remove the additional entries created above
- Verify that the created VLAN_MEMBER entries are deleted in case of platforms that create VLAN_MEMBER.
- Verify that L2MC_GROUP_MEMBER entries are deleted in case of platforms creating SAI_OBJECT_TYPE_L2MC_GROUP_MEMBER per end point IP.
10. Remove the last REMOTE_VNI entry for a DIP
- Verify that the created VLAN_MEMBER, TUNNEL, BRIDGEPORT ports are deleted for platforms that use P2P Tunnels.
- Verify that L2MC_GROUP_MEMBERS are removed, L2MC_GROUP is deleted and vlan's flood group are set to null object as well as vlan's flood type is updated to SAI_VLAN_FLOOD_CONTROL_TYPE_ALL in case of platforms that use P2MP tunnel.

### 8.2 FdbOrch

1. Create a VXLAN_REMOTE_VNI entry to a remote destination IP.
2. Add VXLAN_REMOTE_MAC entry to the above remote IP and VLAN.

- Verify ASIC DB table fdb entry is created with remote_ip and bridgeport information.
- Verify ASIC DB table fdb entry is created with remote_ip and bridgeport information.
- In case of platforms that use P2P tunnel, verify that P2P tunnel's bridgeport is used.
- In case of platforms that use P2MP tunnel, verify that P2MP tunnel's bridge port is used.
3. Remove the above MAC entry and verify that the corresponding ASIC DB entry is removed.
4. Repeat above steps for remote static MACs.
5. Add MAC in the ASIC DB and verify that the STATE_DB MAC_TABLE is updated.
Expand Down
Binary file added doc/vxlan/EVPN/images/p2mptunnelcreate.jpg
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added doc/vxlan/EVPN/images/p2mpvlanextension.jpg
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.