Windows SHM: Processes running in different Session IDs cannot communiate (eg when running from a windows service)

[YangHuaidong]

Problem Description

When I make my service into a Windows service, I expect that the main service with system privileges will start a process to publish a topic, and then I use a normal privileged process to subscribe to it and get messages. However, I find that publishing the topic and sending messages are all fine, but after subscribing, I can't get messages. Why is this? I hope to seek help

How to reproduce

As described above

How did you get eCAL?

Download from Release Page

Environment

eCAl： 5.12.4 windwos 10

Summary by eCAL Devs:

• The SHM Layer fails when Processes run with different Session IDs
• This Stackoverflow page tells how to open "global" objects: https://stackoverflow.com/questions/47021915/c-issues-using-named-shared-memory
• I think it should be possible to use psexec to run with any program uner a different session ID (the -i parameter)
  psexec.exe -accepteula -nobanner -i 1 -s C:\eCAL\bin\ecal_sys_client.exe --auto-accept-security-risk
  Session ID 1 is usually the ID of the session you are looking at when you sit in front of the PC
• The session ID of a windows task can be obtained from the Taskmanager or Process hacker by enabling the corresponding column ("Sitzungskennung" in German Windows)

[FlorianReimold]

Interesting discussion, even though the eCAL Version you are using has reached it's end of life and is not supported anymore. You can upgrade to the binary-compatible eCAL 5.12.7 though.

I believe the issue is related to the Shared Memory handling in Windows. Are you able to see the topic in eCAL Monitor list?

[YangHuaidong]

Yes, I can see the topic, but I can't receive messages after subscribing. I found that the message is in a different session ID and the UDP carousel is blocked. Can this problem be solved?

[FlorianReimold]

I found that the message is in a different session ID and the UDP carousel is blocked

Can you elaborate on that? I cannot follow your findings, sorry. But If you already dug into the topics, you may have valuable insights.

If you can see the topic, then UDP Multicast is working fine. That's what I expected. In the past I ran into the same issue as you when using psexec -i 1 to launch an eCAL Application on a remote Windows machine without the user being logged in. After login, I even saw the console window of the eCAL Application, but the shared memory seemed to not work.

For future reference, this was the entire command line of psexec:
psexec.exe -accepteula -nobanner -i 1 -s C:\eCAL\bin\ecal_sys_client.exe --auto-accept-security-risk
The comand was executed with ansible. Note sure about the technical details, but I think Ansible somehow uses a remote PowerShell to execute these commands.

[YangHuaidong]

At present, I can confirm that the problem is that different session IDs in Windows will prevent the upd carousel. After I set use_tcp in publisher to 1 in ecal.ini, I can receive messages normally.

[YangHuaidong]

I have another question I'd like to ask for your help. If use_inproc = 0, use_shm = 2, use_tcp = 1, and use_udp_mc = 0, will shm or tcp be used first?

[FlorianReimold]

At present, I can confirm that the problem is that different session IDs in Windows will prevent the upd carousel.

I am still convinced that UDP is working as expected. The Shared Memory data transport is causing the issues here. I don't know what you mean by "carousel" though, I don't recognize that term in this context.

After I set use_tcp in publisher to 1 in ecal.ini, I can receive messages normally.

Yes, use_tcp or use_udp_mc to 1 should solve your issues, as both the UDP and the TCP Layer should work as normal (but by default they are not used when communicating between tasks running on the same PC). I would however also recommend to disable shared memory by setting use_shm = 0. And you should only use one of the network layers, as otherwise the data will be sent multiple times.

If use_inproc = 0, use_shm = 2, use_tcp = 1, and use_udp_mc = 0, will shm or tcp be used first?

The data will be sent on both data transport layers (SHM and TCP) if you communicate within the same host. That is inefficient for obvious reasons, but if the data is small enough you will not notice it. You can check the data transport layers in the eCAL Monitor (You need to enable the Layer column by right-clicking the header