[feature] Inherit resource limits from cgroups (#1336)

When GTS is running in a container runtime which has configured CPU or
memory limits or under an init system that uses cgroups to impose CPU
and memory limits the values the Go runtime sees for GOMAXPROCS and
GOMEMLIMIT are still based on the host resources, not the cgroup.

At least for the throttling middlewares which use GOMAXPROCS to
configure their queue size, this can result in GTS running with values
too big compared to the resources that will actuall be available to it.

This introduces 2 dependencies which can pick up resource contraints
from the current cgroup and tune the Go runtime accordingly. This should
result in the different queues being appropriately sized and in general
more predictable performance. These dependencies are a no-op on
non-Linux systems or if running in a cgroup that doesn't set a limit on
CPU or memory.

The automatic tuning of GOMEMLIMIT can be disabled by either explicitly
setting GOMEMLIMIT yourself or by setting AUTOMEMLIMIT=off. The
automatic tuning of GOMAXPROCS can similarly be counteracted by setting
GOMAXPROCS yourself.

vendor/github.com/cilium/ebpf/link/cgroup.go

@@ -0,0 +1,169 @@
+package link
+
+import (
+	"errors"
+	"fmt"
+	"os"
+
+	"github.com/cilium/ebpf"
+)
+
+type cgroupAttachFlags uint32
+
+// cgroup attach flags
+const (
+	flagAllowOverride cgroupAttachFlags = 1 << iota
+	flagAllowMulti
+	flagReplace
+)
+
+type CgroupOptions struct {
+	// Path to a cgroupv2 folder.
+	Path string
+	// One of the AttachCgroup* constants
+	Attach ebpf.AttachType
+	// Program must be of type CGroup*, and the attach type must match Attach.
+	Program *ebpf.Program
+}
+
+// AttachCgroup links a BPF program to a cgroup.
+func AttachCgroup(opts CgroupOptions) (Link, error) {
+	cgroup, err := os.Open(opts.Path)
+	if err != nil {
+		return nil, fmt.Errorf("can't open cgroup: %s", err)
+	}
+
+	clone, err := opts.Program.Clone()
+	if err != nil {
+		cgroup.Close()
+		return nil, err
+	}
+
+	var cg Link
+	cg, err = newLinkCgroup(cgroup, opts.Attach, clone)
+	if errors.Is(err, ErrNotSupported) {
+		cg, err = newProgAttachCgroup(cgroup, opts.Attach, clone, flagAllowMulti)
+	}
+	if errors.Is(err, ErrNotSupported) {
+		cg, err = newProgAttachCgroup(cgroup, opts.Attach, clone, flagAllowOverride)
+	}
+	if err != nil {
+		cgroup.Close()
+		clone.Close()
+		return nil, err
+	}
+
+	return cg, nil
+}
+
+// LoadPinnedCgroup loads a pinned cgroup from a bpffs.
+func LoadPinnedCgroup(fileName string) (Link, error) {
+	link, err := LoadPinnedRawLink(fileName)
+	if err != nil {
+		return nil, err
+	}
+
+	return &linkCgroup{link}, nil
+}
+
+type progAttachCgroup struct {
+	cgroup     *os.File
+	current    *ebpf.Program
+	attachType ebpf.AttachType
+	flags      cgroupAttachFlags
+}
+
+var _ Link = (*progAttachCgroup)(nil)
+
+func (cg *progAttachCgroup) isLink() {}
+
+func newProgAttachCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program, flags cgroupAttachFlags) (*progAttachCgroup, error) {
+	if flags&flagAllowMulti > 0 {
+		if err := haveProgAttachReplace(); err != nil {
+			return nil, fmt.Errorf("can't support multiple programs: %w", err)
+		}
+	}
+
+	err := RawAttachProgram(RawAttachProgramOptions{
+		Target:  int(cgroup.Fd()),
+		Program: prog,
+		Flags:   uint32(flags),
+		Attach:  attach,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("cgroup: %w", err)
+	}
+
+	return &progAttachCgroup{cgroup, prog, attach, flags}, nil
+}
+
+func (cg *progAttachCgroup) Close() error {
+	defer cg.cgroup.Close()
+	defer cg.current.Close()
+
+	err := RawDetachProgram(RawDetachProgramOptions{
+		Target:  int(cg.cgroup.Fd()),
+		Program: cg.current,
+		Attach:  cg.attachType,
+	})
+	if err != nil {
+		return fmt.Errorf("close cgroup: %s", err)
+	}
+	return nil
+}
+
+func (cg *progAttachCgroup) Update(prog *ebpf.Program) error {
+	new, err := prog.Clone()
+	if err != nil {
+		return err
+	}
+
+	args := RawAttachProgramOptions{
+		Target:  int(cg.cgroup.Fd()),
+		Program: prog,
+		Attach:  cg.attachType,
+		Flags:   uint32(cg.flags),
+	}
+
+	if cg.flags&flagAllowMulti > 0 {
+		// Atomically replacing multiple programs requires at least
+		// 5.5 (commit 7dd68b3279f17921 "bpf: Support replacing cgroup-bpf
+		// program in MULTI mode")
+		args.Flags |= uint32(flagReplace)
+		args.Replace = cg.current
+	}
+
+	if err := RawAttachProgram(args); err != nil {
+		new.Close()
+		return fmt.Errorf("can't update cgroup: %s", err)
+	}
+
+	cg.current.Close()
+	cg.current = new
+	return nil
+}
+
+func (cg *progAttachCgroup) Pin(string) error {
+	return fmt.Errorf("can't pin cgroup: %w", ErrNotSupported)
+}
+
+type linkCgroup struct {
+	*RawLink
+}
+
+var _ Link = (*linkCgroup)(nil)
+
+func (cg *linkCgroup) isLink() {}
+
+func newLinkCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program) (*linkCgroup, error) {
+	link, err := AttachRawLink(RawLinkOptions{
+		Target:  int(cgroup.Fd()),
+		Program: prog,
+		Attach:  attach,
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	return &linkCgroup{link}, err
+}