bazel: fix numactl

[rockwotj]

Incorporate libnuma into the Bazel build. We force dynamic linking because it's LGPL. The dynamic linking forcing is really annoying, but honestly less annoying than the new way of doing things with cc_shared_library. However I ran into issues using the cc_shared_library approach, so I just fallback to using the "old school" stable ways of making shared libraries.

Backports Required

• none - not a bug fix
• none - this is a backport
• none - issue does not exist in previous branches
• none - papercut/not impactful enough to backport
• v24.3.x
• v24.2.x
• v24.1.x

Release Notes

• none

[dotnwat]

but honestly less annoying than the new way of doing things with cc_shared_library

i'm not saying we should do this, but from what I can tell seastar's libnuma dependency is a single call to mbind in memory.cc to set a policy on some memory regions.

looking at this:

https://github.com/lrita/numa/blob/29f88905963797c57362861a2dd8adb14a6f152a/numa_linux.go#L161

i do wonder if freeing ourselves from the libnuma dependency would be as simple as writing our own version of this

func MBind(addr unsafe.Pointer, length, mode, flags int, nodemask Bitmask) (err error) {
	var mask, maxnode uintptr
	if maxnode = uintptr(nodemask.Len()); maxnode != 0 {
		mask = uintptr(unsafe.Pointer(&nodemask[0]))
	}
	_, _, errno := syscall.Syscall6(syscall.SYS_MBIND, uintptr(addr),
		uintptr(length), uintptr(mode), mask, maxnode, uintptr(flags))
	if errno != 0 {
		err = errno
	}
	return
}

[dotnwat]

nice. i had gone with fake.rs

[rockwotj]

but honestly less annoying than the new way of doing things with cc_shared_library

i'm not saying we should do this, but from what I can tell seastar's libnuma dependency is a single call to mbind in memory.cc to set a policy on some memory regions.

looking at this:

https://github.com/lrita/numa/blob/29f88905963797c57362861a2dd8adb14a6f152a/numa_linux.go#L161

i do wonder if freeing ourselves from the libnuma dependency would be as simple as writing our own version of this

func MBind(addr unsafe.Pointer, length, mode, flags int, nodemask Bitmask) (err error) {

	var mask, maxnode uintptr

	if maxnode = uintptr(nodemask.Len()); maxnode != 0 {

		mask = uintptr(unsafe.Pointer(&nodemask[0]))

	}

	_, _, errno := syscall.Syscall6(syscall.SYS_MBIND, uintptr(addr),

		uintptr(length), uintptr(mode), mask, maxnode, uintptr(flags))

	if errno != 0 {

		err = errno

	}

	return

}

ooh maybe I will send a patch to seastar and see what drama unfolds...