Documentation
¶
Overview ¶
Package containers resolves the cgroup identity (path + inode) of a pod's containers. The cgroup inode is the key both BPF engines use to attribute kernel events back to a workload, so nothing in this package may panic on a pod object, a mountinfo line, or a procfs file.
Index ¶
Constants ¶
const DefaultProcRoot = "/proc"
DefaultProcRoot is the procfs root used by the exported helpers that do not take one explicitly. Tests always pass their own fake tree.
Variables ¶
This section is empty.
Functions ¶
func ExtractCgPaths ¶
func ExtractCgPaths(cgInfos []*ContainerCgroupInfo) []string
func ExtractCgids ¶
func ExtractCgids(cgInfos []*ContainerCgroupInfo) []uint64
Types ¶
type ContainerCgroupInfo ¶
type ContainerCgroupInfo struct {
ID uint64
Path string
Name string // container name, needed by attribution
}
ContainerCgroupInfo is the resolved cgroup identity of a single container. It is used as a map key by the managers, so every field must stay comparable.
func ResolveCgInfos ¶
func ResolveCgInfos(pod *corev1.Pod) ([]*ContainerCgroupInfo, error)
ResolveCgInfos resolves the cgroup identity of every running container of pod. It never panics and never lets a single unresolvable container discard its healthy siblings: containers whose ID is missing (not started yet) or whose cgroup path cannot be found are skipped, and the infos that could be resolved are returned together with a joined error describing the rest, so callers may retry later.
func ResolveCgroupByPID ¶
func ResolveCgroupByPID(procRoot string, pid uint32) (*ContainerCgroupInfo, error)
ResolveCgroupByPID is the authoritative fallback: it asks the kernel which cgroup a pid belongs to instead of guessing a layout. The daemon runs with hostPID and mounts /, so this works for every runtime, cgroup driver and cgroup version. procRoot is parameterized for tests.
Source Files