David Matson

I see - thanks for the context. > For various other ways to write to stdout/err, you have to use the experimental capture implementation. Any pointers on how to use...

Sounds good. On Windows, the standard approach is to use anonymous pipes, which avoids the need to maintain a temp file: https://docs.microsoft.com/en-us/windows/win32/procthread/creating-a-child-process-with-redirected-input-and-output (that's for a child process, but the same...

Sounds good. I did a little more digging, and anonymous pipes exist on POSIXy platforms as well: https://linux.die.net/man/7/pipe If we switched both Windows and POSIX over to use anonymous pipes,...

> But the danger with pipes is that they're not infinitely deep, and you have to keep them flushed or the writing end can end up blocking. Yep, that's absolutely...

@manojampalam: > Since we need to support downlevel platforms, we cannot completely replace the existing named-pipes based alternative yet. Can UNIX domain sockets be used on platforms that have them,...

Or, even better than a version check, look for whether there's an afunix driver present (API equivalent of checking sc.exe query afunix [State == RUNNING]).

I would just caution against using a OS version check. For example, some internal/test versions do not have afunix present even with a current OS version number.

This issue/feature request is specifically about the idea that if for any reason CNI stops working on a node, the node isn't marked Not Ready, so new pods still try...

This issue isn't specific to v1.0.18, but yes I am still seeing problems there. This issue is about the fact that multiple different CNI bugs can cause the node to...

/remove-stale