Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func CiliumAvailable ¶
Types ¶
type Client ¶
type Client interface {
ListEntries(filter EntriesFilter) ([]*Entry, error)
Close() error
}
Client defines the interface for conntrack clients (Netfilter or Cilium)
func NewCiliumClient ¶
func NewCiliumClient(log logr.Logger, clockSource ClockSource) (Client, error)
type ClockSource ¶
type ClockSource string
const ( ClockSourceKtime ClockSource = "ktime" ClockSourceJiffies ClockSource = "jiffies" )
type Config ¶
type Config struct {
ReadInterval time.Duration
CleanupInterval time.Duration
FlushInterval time.Duration
NodeName string
OperatorVersion string
OperatorCommit string
}
Config holds configuration for the monitor
type EgressPathResolver ¶ added in v0.1.19
type EgressPathResolver struct {
// contains filtered or unexported fields
}
EgressPathResolver reads the node's instance metadata once and caches the answer for the lifetime of the process. The facts it reports -- provider, region, zone, VPC, subnet, whether the primary interface has a public address -- do not change while a node is running; a node that is re-addressed is a new node with a new pod.
func NewEgressPathResolver ¶ added in v0.1.19
func NewEgressPathResolver(log logr.Logger) *EgressPathResolver
NewEgressPathResolver builds a resolver against the real metadata endpoints.
func (*EgressPathResolver) Cached ¶ added in v0.1.19
func (r *EgressPathResolver) Cached() *gen.NodeEgressPath
Cached returns the best answer so far, or nil if no probe has succeeded yet. The answer may be incomplete (no egress path); the control plane records that as unknown. It never performs IO and never blocks on the network, so it is safe to call from the flush path while a collection lock is held.
func (*EgressPathResolver) Prime ¶ added in v0.1.19
func (r *EgressPathResolver) Prime(ctx context.Context)
Prime probes the metadata service unless a COMPLETE answer is already held, one is already running, or a recent attempt is still inside its retry interval. It performs network IO and so must never be called while holding a collection lock -- a black-holed link-local route costs several metadataTimeout waits.
Safe to call repeatedly; the monitor calls it once at startup and again from the flush loop for as long as the answer is still incomplete.
func (*EgressPathResolver) Resolve ¶ added in v0.1.19
func (r *EgressPathResolver) Resolve(ctx context.Context) *gen.NodeEgressPath
Resolve primes if the answer is not yet complete and returns whatever is known. It can block on the network; prefer Cached plus a background Prime on hot paths.
type EnrichedDNSLookup ¶
type EntriesFilter ¶
func FilterBySrcIP ¶
func FilterBySrcIP(ips map[netaddr.IP]struct{}) EntriesFilter
type Entry ¶
type Entry struct {
Src netaddr.IPPort
// Dst is the destination the client dialed, taken from conntrack's ORIGINAL
// tuple. For a ClusterIP connection that is the service address, not the pod
// that answered.
Dst netaddr.IPPort
// DstTranslated is the post-DNAT destination, taken from the source of
// conntrack's REPLY tuple: the backend that actually served the connection.
//
// It is the zero IPPort when the kernel did not translate the destination,
// which is the common case for pod-to-pod and egress traffic. Use
// HasTranslatedDst to test it rather than comparing against Dst, since a
// non-DNAT'd flow leaves this zero rather than echoing Dst.
DstTranslated netaddr.IPPort
TxBytes uint64
TxPackets uint64
RxBytes uint64
RxPackets uint64
Lifetime time.Time
Proto uint8
}
Entry represents a single conntrack entry
func (*Entry) HasTranslatedDst ¶ added in v0.1.19
HasTranslatedDst reports whether the kernel DNAT'd this connection, i.e. whether the address that answered differs from the address the client dialed.
type MetricsResponse ¶
type MetricsResponse struct {
NodeName string `json:"node_name"`
Items []*NetworkFlow `json:"items"`
DNSLookups []EnrichedDNSLookup `json:"dns_lookups"`
Ip2Domain map[string]string `json:"ip2domain,omitempty"`
}
type Monitor ¶
type Monitor struct {
// contains filtered or unexported fields
}
Monitor collects network flows using conntrack and aggregates them
func NewMonitor ¶
func NewMonitor(cfg Config, log logr.Logger, podCache *PodCache, ct Client, tracer *ebpf.Tracer, dnsCollector dns.DNSCollector, dakrClient transport.DakrClient, healthManager *health.HealthManager) (*Monitor, error)
NewMonitor creates a new Monitor instance
func (*Monitor) EnableContainerRedMetrics ¶ added in v0.1.12
func (m *Monitor) EnableContainerRedMetrics(agg *redmetrics.Aggregator)
EnableContainerRedMetrics attaches a container RED metrics aggregator, whose closed minute buckets are drained on every flush.
It is a setter rather than a NewMonitor parameter because the feature is off by default: a nil aggregator has to remain an ordinary, well-tested state, and threading a nil through the constructor invites the opposite.
func (*Monitor) GetMetricsHandler ¶
func (m *Monitor) GetMetricsHandler(w http.ResponseWriter, r *http.Request)
GetMetricsHandler serves the collected metrics NOTE: This currently resets the counters after read (Delta mode), mimicking egressd's behavior
type NetfilterClient ¶
type NetfilterClient struct {
// contains filtered or unexported fields
}
NetfilterClient uses the kernel's conntrack table via Netlink
func (*NetfilterClient) Close ¶
func (c *NetfilterClient) Close() error
func (*NetfilterClient) ListEntries ¶
func (c *NetfilterClient) ListEntries(filter EntriesFilter) ([]*Entry, error)
type NetworkFlow ¶
type NetworkFlow struct {
SrcIP string `json:"src_ip"`
DstIP string `json:"dst_ip"`
SrcDomain string `json:"src_domain,omitempty"`
DstDomain string `json:"dst_domain,omitempty"`
// Post-DNAT destination, empty when the kernel did not translate one.
// DstIP stays the address the client dialed (a ClusterIP, typically); these
// name the backend that actually answered. See Entry.DstTranslated.
DstIPTranslated string `json:"dst_ip_translated,omitempty"`
DstPortTranslated uint16 `json:"dst_port_translated,omitempty"`
// Pod Metadata
SrcPodName string `json:"src_pod_name,omitempty"`
SrcPodNamespace string `json:"src_pod_namespace,omitempty"`
DstPodName string `json:"dst_pod_name,omitempty"`
DstPodNamespace string `json:"dst_pod_namespace,omitempty"`
Protocol uint8 `json:"protocol"`
DstPort uint16 `json:"dst_port"`
TxBytes uint64 `json:"tx_bytes"`
RxBytes uint64 `json:"rx_bytes"`
TxPackets uint64 `json:"tx_packets"`
RxPackets uint64 `json:"rx_packets"`
Timestamp time.Time `json:"timestamp"`
// contains filtered or unexported fields
}
NetworkFlow represents a single aggregated network flow This matches the conceptual "RawNetworkMetric" from egressd
type PodCache ¶
type PodCache struct {
// contains filtered or unexported fields
}
PodCache watches for pods on the local node and maintains an IP lookup table
func NewPodCache ¶
func NewPodCache(informer cache.SharedIndexInformer) *PodCache
NewPodCache creates a new PodCache using the provided informer. If informer is nil, it returns a cache that relies only on manual updates (standalone mode).
func (*PodCache) GetLocalPodIPs ¶
GetLocalPodIPs returns a map of all local pod IPs
Source Files
¶
Directories
¶
| Path | Synopsis |
|---|---|
|
Package redmetrics turns socket observations into per-container, per-minute RED measurements for the containers running on this node.
|
Package redmetrics turns socket observations into per-container, per-minute RED measurements for the containers running on this node. |
|
socketprobe
Package socketprobe attaches the socket probes that feed container RED metric aggregation and turns their ring buffer records into redmetrics.SocketEvent values.
|
Package socketprobe attaches the socket probes that feed container RED metric aggregation and turns their ring buffer records into redmetrics.SocketEvent values. |