AmsTcpIpRouter delays when using windows containers and NAT network driver

[peer-dellis]

This is the setup I am using

The use case for this setup is to be able to preform a hot swap of our software. Container 1 would be running version 1.1 of our software and container 2 startup running version 1.2. Once a handoff is completed container 2 would take over and container 1 would shutdown.

After many trial and error and support from [email protected] the above setup is working with one quirk, there is a 4000ms delay from when a client sends a request to when the router forwards the request to the PLC.

log example

//AdsClient.ReadAny called
2024-02-13 09:01:38.906 -05:00 [DBG] ReadRequest: Addr:20.0.21.162.1.1:851, InvokeId:14, IG:f005, IO:19000010, Len:2)
2024-02-13 09:01:38.906 -05:00 [VRB] RequestSync: Addr:20.0.21.162.1.1:851, InvokeId:14, serviceId:"Read", Len:0, header:ReadReqHeader(IG:f005,IO:19000010,Len:2, data:
2024-02-13 09:01:38.906 -05:00 [DBG] RequestSync: Addr:20.0.21.162.1.1:851, InvokeId:14, serviceId:"Read", Len:0, header:ReadReqHeader(IG:f005,IO:19000010,Len:2, data: succeeded!

//4 seconds later router sends read request 
2024-02-13 09:01:42.933 dbug: AdsRouter[0] Send: --> Sending AmsSend Command: 20.0.25.68.1.1:32768 --> 20.0.21.162.1.1:851, Cmd:Read REQ, State:MaskAdsRequest, Err:0, User:14, Bytes:12
2024-02-13 09:01:42.933 info: AdsRouter[0] Send: --> Send AmsSend Command: 20.0.25.68.1.1:32768 --> 20.0.21.162.1.1:851, Cmd:Read REQ, State:MaskAdsRequest, Err:0, User:14, Bytes:12

//router recieves response
2024-02-13 09:01:42.936 dbug: AdsRouter[0] Received: 20.0.25.68.1.1:32768 <-- 20.0.21.162.1.1:851, Cmd: Read RES, State:MaskAdsResponse, Err:0, User:14, Bytes:10 (Sender20.0.21.162.1.1:851 (Endpoint: [::ffff:20.0.21.162]:48898), CNT: 1)
2024-02-13 09:01:42.936 info: AdsRouter[0] Received: 20.0.25.68.1.1:32768 <-- 20.0.21.162.1.1:851, Cmd: Read RES, State:MaskAdsResponse, Err:0, User:14, Bytes:10 (Sender:20.0.21.162.1.1:851 (Endpoint: [::ffff:20.0.21.162]:48898), Receiver:20.0.25.68.1.1:32768 (Endpoint: 172.16.238.17:49160), Count:1

//AdsClient receives response
2024-02-13 09:01:42.941 -05:00 [DBG] onConfirmationReadAsync: Addr:20.0.21.162.1.1:851, InvokeId:14, Result:"NoError", Len:2, Data:CE-00
2024-02-13 09:01:42.941 -05:00 [DBG] RequestAndReceiveSync: InvokeID:'14' succeeded!

The router and client console apps are based on the example code Beckhoff provides, I've attached the program.cs file.
client and router.zip

I've also attached the docker images
docker images.zip

I've tested the same setup with linux containers using the same container host (Win 11).
There is no delay using linux, so I assume the problem is related to using the nat driver as opposed to the bridge driver.

[RalfHeitmann]

Unfortunately I am not an expert in Docker usage so that I cannot answer why windows containers should behave this way.

For reproducing the issue I miss some relevant information in the attached files or there are some inconsistencies to your description.

• I see a router instance not only in the attached router-program.cs, but also in the client-program.cs. Was that the intention? I cannot match this situation to your overview picture.
• What Version of ADS packages is used in your tests?
• About the delay, is this appearing with every read or only in the 'takeover' situation?
• If its only in the hot swap, how is the hot swap realized? (Cannot see it in the code files)

[peer-dellis]

Sorry for the delay.

• A router instance is required even for the client containers. The client router instance connects to the router container which routes traffic to the PLC. I've found that there must be a one to one route configured between the PLC and Container Host PC. By adding the router container we can establish the one to one route with the 2 clients connecting to the router contianer.
• 6.1.197
• Occurs with every read even if only one client container is connected.

To reproduce you only need two containers running, one router and one client. They delay is seen without the 2nd client running. However to reproduce with 2 clients I created a second container with the client-program and used the command line args to give it a different net id.