Documentation
¶
Overview ¶
Manager of cAdvisor-monitored containers.
Index ¶
Constants ¶
const ( DockerNamespace = "docker" PodmanNamespace = "podman" )
The namespace under which aliases are unique.
Variables ¶
var ( // PerfManagerFactory builds the perf_event stats manager. Set by the root // binary to perf.NewManager. PerfManagerFactory func(configFile string, topology []info.Node) (stats.Manager, error) // ResctrlManagerFactory builds the resctrl stats manager. Set by the root // binary to resctrl.NewManager. ResctrlManagerFactory func(interval time.Duration, vendorID string, inHostNamespace bool) (stats.ResctrlManager, error) )
Collector-manager injection seam.
The kubelet-only library does not import the perf_event / resctrl implementations (or their dependencies). Instead, the full cAdvisor binary registers factories here at init time; when a factory is nil (the kubelet case) the manager falls back to a Noop manager, so no stats are collected for that controller. This keeps github.com/google/cadvisor/lib lean while letting the binary remain fully functional.
var CollectorManagerFactory func(handler container.ContainerHandler, readFile func(string) ([]byte, error)) (CollectorManager, error)
CollectorManagerFactory builds a per-container collector manager. It receives the container handler (to discover collector configs from its labels) and a readFile func (to read those config files from inside the container). Set by the root binary.
var CpuLoadReaderFactory func() (CpuLoadReader, error)
CpuLoadReaderFactory builds a CpuLoadReader. Set by the root binary when --enable_load_reader is on; nil for the kubelet.
var HousekeepingConfigFlags = HousekeepingConfig{ flag.Duration("max_housekeeping_interval", 60*time.Second, "Largest interval to allow between container housekeepings"), flag.Bool("allow_dynamic_housekeeping", true, "Whether to allow the housekeeping interval to be dynamic"), }
var HousekeepingInterval = flag.Duration("housekeeping_interval", 1*time.Second, "Interval between container housekeepings")
var InitialSplayFactor = flag.Float64("initial_splay_factor", jitterDefault, "Factor for the initial splay b/w the container housekeepings, default is 1.0. If negative value is passed, the value will be reset to default")
var JitterFactor = flag.Float64("jitter_factor", jitterDefault, "Factor for the jitters after the initial splay b/w the container housekeepings, default is 1.0. If negative value is passed, the value will be reset to default")
var ProcessListProvider func(containerName string, isRoot bool, cadvisorContainer string, inHostNamespace bool) ([]info.ProcessInfo, error)
ProcessListProvider lists the processes running in a container (the v2 /ps endpoint). containerName is the container's cgroup name; isRoot reports whether it is the root container. Set by the root binary to a ps-based implementation; nil for the kubelet (GetProcessList returns empty).
var SummaryReaderFactory func(spec info.ContainerSpec) (SummaryReader, error)
SummaryReaderFactory builds a per-container summary reader from its spec. Set by the root binary (wrapping summary.New).
Functions ¶
func ApplicationMetricsCountLimit ¶
func ApplicationMetricsCountLimit() int
ApplicationMetricsCountLimit exposes the application_metrics_count_limit flag value so the root binary can build its collectors without re-registering the flag.
func EventStorageAgeLimit ¶
func EventStorageAgeLimit() string
EventStorageAgeLimit and EventStorageEventLimit expose the values of the event_storage_* flags this package registers (for kubelet flag-compatibility) so the root binary can build its events storage policy without re-registering — and double-registering — the same flags.
func EventStorageEventLimit ¶
func EventStorageEventLimit() string
Types ¶
type CollectorManager ¶
type CollectorManager interface {
Collect() (time.Time, map[string][]info.MetricVal, error)
GetSpec() ([]info.MetricSpec, error)
}
CollectorManager runs application-metrics collectors for one container and aggregates their custom metrics. The full binary injects a manager backed by the collector package; the kubelet leaves CollectorManagerFactory nil (no custom metrics).
type CpuLoadReader ¶
type CpuLoadReader interface {
Start() error
Stop()
GetCpuLoad(name string, path string) (info.LoadStats, error)
}
CpuLoadReader reads per-container CPU load over netlink (runnable/uninterruptible task counts). The implementation lives in the root binary's utils/cpuload (CAP_NET_ADMIN, a persistent netlink connection); the kubelet leaves CpuLoadReaderFactory nil so no reader is created and LoadAverage stays zero.
type EventSink ¶
EventSink receives the container lifecycle (creation/deletion) and OOM events the manager emits. The full cAdvisor binary injects its events.EventManager here (it satisfies this interface) and serves them over the /events REST API; the kubelet leaves it nil, so the manager emits nothing and the library carries no event-storage machinery. This is the same injection pattern as the perf/resctrl factories — see plugins.go.
type HousekeepingConfig ¶
Housekeeping configuration for the manager
type Manager ¶
type Manager interface {
// Start the manager. Calling other manager methods before this returns
// may produce undefined behavior.
Start() error
// Stops the manager.
Stop() error
// Get V2 information about a container.
// Recursive (subcontainer) requests are best-effort, and may return a partial result alongside an
// error in the partial failure case.
GetContainerInfoV2(containerName string, options info.RequestOptions) (map[string]info.ContainerInfo, error)
// Get info for all requested containers based on the request options.
GetRequestedContainersInfo(containerName string, options info.RequestOptions) (map[string]*info.ContainerInfo, error)
// Get information about the machine.
GetMachineInfo() (*info.MachineInfo, error)
// Get version information about different components we depend on.
GetVersionInfo() (*info.VersionInfo, error)
// Get filesystem information for the filesystem that contains the given directory
GetDirFsInfo(dir string) (info.FsInfo, error)
// Get filesystem information for a given label.
// Returns information for all global filesystems if label is empty.
GetFsInfo(label string) ([]info.FsInfo, error)
// DebugInfo returns debug information about the manager's tracked containers.
DebugInfo() map[string][]string
// Get information about a container.
GetContainerInfo(containerName string, query *info.ContainerInfoRequest) (*info.ContainerInfo, error)
// Get information about all subcontainers of the specified container (includes self).
SubcontainersInfo(containerName string, query *info.ContainerInfoRequest) ([]*info.ContainerInfo, error)
// Get information about the docker containers (by container name) of an instance.
AllDockerContainers(query *info.ContainerInfoRequest) (map[string]info.ContainerInfo, error)
// Get information about the docker container with the specified name.
DockerContainer(dockerName string, query *info.ContainerInfoRequest) (info.ContainerInfo, error)
// Get information about the podman containers (by container name) of an instance.
AllPodmanContainers(query *info.ContainerInfoRequest) (map[string]info.ContainerInfo, error)
// Get information about the podman container with the specified name.
PodmanContainer(containerName string, query *info.ContainerInfoRequest) (info.ContainerInfo, error)
// Get the specs for a container, possibly with subcontainers.
GetContainerSpec(containerName string, options info.RequestOptions) (map[string]info.ContainerSpec, error)
// Get filesystem information for the filesystem identified by the given UUID.
GetFsInfoByFsUUID(uuid string) (info.FsInfo, error)
// Returns true if the named container exists.
Exists(containerName string) bool
// Get the derived (rolling-window percentile) stats for a container.
GetDerivedStats(containerName string, options info.RequestOptions) (map[string]info.DerivedStats, error)
// Get the list of processes running in a container.
GetProcessList(containerName string, options info.RequestOptions) ([]info.ProcessInfo, error)
// SetEventSink wires the sink that receives container lifecycle and OOM
// events. The full binary injects its events manager; the kubelet does not
// call this (no events are emitted). See events.go.
SetEventSink(sink EventSink)
}
The Manager interface defines operations for starting a manager and getting container and machine information.
func New ¶
func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, HousekeepingConfig HousekeepingConfig, includedMetricsSet container.MetricSet, rawContainerCgroupPathPrefixWhiteList, containerEnvMetadataWhiteList []string, perfEventsFile string, resctrlInterval time.Duration) (Manager, error)
New takes a memory storage and returns a new manager.
type SummaryReader ¶
type SummaryReader interface {
AddSample(stat info.ContainerStats) error
DerivedStats() (info.DerivedStats, error)
}
SummaryReader computes rolling-window derived/percentile usage stats for a single container. The full binary injects a reader backed by the summary package; the kubelet leaves SummaryReaderFactory nil and no derived stats are produced (GetDerivedStats reports "not enabled").
Source Files