Documentation
¶
Overview ¶
Package hostmetrics reads one machine's own counters: CPU, memory, disk, network. It is a leaf on purpose. Both the panel's sampler and the node agent take this reading, and the agent must not import infra/metrics for it: metrics imports managedtiles, and that edge closed the cycle that kept managedtiles from importing the node-aware runtime (docs/plans/35-cluster.md).
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ProcNetPath ¶
ProcNetPath resolves a /proc/net file in the HOST network namespace. /proc/net symlinks to /proc/self/net, so through a HOST_PROC bind mount it would show the container's own netns, pid 1's net dir is the host's. Exported for the panel's conntrack flow sampler, which reads the same tree.
Types ¶
type HostCounters ¶
type HostCounters struct {
At time.Time
CPUIdle, CPUTotal uint64
RxBytes, TxBytes uint64
MemUsed int64
DiskUsed int64
// OK is false when /proc could not be read at all, which is every
// non-Linux platform. A sample is never derived from one.
OK bool
}
HostCounters is one raw reading of the machine's own counters: cumulative CPU jiffies, cumulative network bytes, and the point-in-time memory and disk figures. Two readings make a sample.
Exported because the node agent takes the same reading on a worker and posts it to the panel, the graphs on a worker's server page are the same graphs as the manager's, so the numbers behind them have to be measured the same way (docs/plans/32-multi-node-ui.md, host metrics).
func ReadHost ¶
func ReadHost(now time.Time) HostCounters
ReadHost takes one reading. Linux only; elsewhere OK is false.
func (HostCounters) Sample ¶
func (c HostCounters) Sample(prev HostCounters) (HostSample, bool)
Sample turns the previous reading and this one into rates. ok is false on the first reading, on a counter reset, and on any platform ReadHost cannot serve, in each case there is a reading to keep but no sample to store.