You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Count RAM usage of pipe generated by pipe() and fcntl(F_GETPIPE_SZ) #308
Hello, indeed that would be interesting to track. We have a few improvements to the memory instrument planned. We'll see if there's a generic way to retrieve this memory allocation through sys call monitoring, else we may just add a declarative way for users to signal ram usage that we fail to detect automatically, although I'd like to avoid this.
fcntl returns pipe size (which might larger than requested size).
In my thought, this is same with requesting to count size of files on /tmp if /tmp is tmpfs.
Activity
GuillaumeLagrange commentedon Apr 21, 2026
Hello, indeed that would be interesting to track. We have a few improvements to the memory instrument planned. We'll see if there's a generic way to retrieve this memory allocation through sys call monitoring, else we may just add a declarative way for users to signal ram usage that we fail to detect automatically, although I'd like to avoid this.
oech3 commentedon Apr 21, 2026
fcntl returns pipe size (which might larger than requested size).
In my thought, this is same with requesting to count size of files on
/tmpif/tmpis tmpfs.oech3 commentedon Apr 22, 2026
Same issue for memfd_create?
oech3 commentedon Jun 6, 2026
In context of zero-copy, pipe size is not an actuall RAM usage, but I think we can detect it as page cash's generated by splice call.