[Mellanox] Update dscp remapping cases for Nvidia platforms

[congh-nvidia]

Description of PR

Summary:
This PR contains the fix for two test cases in DSCP remapping test: test_tunnel_decap_dscp_to_pg_mapping and test_xoff_for_pcbb.
The reason that we need this fix:

1. Nvidia's implementation for DSCP remapping is different from Community. We need some additional mapping and the tunnel mode is pipe. For details please check the PR: [Mellanox] Support DSCP remapping in dual ToR topo on T0 switch sonic-buildimage#12605.
2. Nvidia's cell size is 144.
3. Need to stop packet aging in buffer.
4. Need to add qos parameters for pcbb xoff test case.

Type of change

• Bug fix
• Testbed and Framework(new/improvement)
• Test case(new/improvement)

Back port request

• 201911
• 202012
• 202205

Approach

What is the motivation for this PR?

Support dscp remapping cases test_tunnel_decap_dscp_to_pg_mapping and test_xoff_for_pcbb on Nvidia platforms.

How did you do it?

Changes:
For test case test_tunnel_decap_dscp_to_pg_mapping:

1. Add a new mapping file tunnel_qos_map_nvidia.json and use it when testing Nvidia platforms.
2. Use cell size 144 for Nvidia platforms.
3. Use outer DSCP instead of inner DSCP.
4. Add some logs for easier debugging.

For test_xoff_for_pcbb:

1. Add test data in qos.yml file for 4600C. (data for 2700 will be added in another PR)
2. Stop packet aging in buffer before test.
3. Skip filling the leak on Nvidia platforms for it is not needed.
4. Use cell size 144 and packet length 300, small packets will cause descriptor exhaustion.

How did you verify/test it?

By automation, all cases passed on 4600C dualtor setup.

Any platform specific information?

Supported testbed topology if it's a new test case?

Documentation

[roy-sror]

better to avoid the if-else statements for Nvidia, and do it once at the beginning.

if nvidia:
param1 = A
param2= B
else:
param1 = C
param2 = D

func_a(param1, param2)

[congh-nvidia]

Hi @roy-sror, can't do it here.
As you can see the params are named according to the mapping names defined in SONiC, and for Nvidia and other platforms, we are using different maps:

if is_nvidia_platform:
for k, v in maps['DSCP_TO_TC_MAP'][UPLINK_MAP_NAME].items():
ret['inner_dscp_to_pg_map'][int(k)] = int(
maps['TC_TO_PRIORITY_GROUP_MAP'][MAP_NAME][v])
else:
for k, v in maps['DSCP_TO_TC_MAP'][TUNNEL_MAP_NAME].items():
ret['inner_dscp_to_pg_map'][int(k)] = int(
maps['TC_TO_PRIORITY_GROUP_MAP'][TUNNEL_MAP_NAME][v])

It's very hard to unify the map names, and even if we can do that, the names will be long and the code will be hard to understand.

[bingwang-ms]

@neethajohn Could you help review the change? Thanks!

[liat-grozovik]

/azp run Azure.sonic-mgmt

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[liat-grozovik]

@congh-nvidia is it needed for other branches?

[congh-nvidia]

@congh-nvidia is it needed for other branches?

Yes, it's needed for 202205, but there will be conflict, I'll open a new PR for it. Thanks.

[congh-nvidia]

@liat-grozovik @bingwang-ms @neethajohn , this is the manual cherry-pick for 202205 #8713, please review, thanks.