Count link and syscall failures instead of printing them

This commit is contained in:
flamingcow
2026-07-25 22:54:04 -07:00
parent 1b322bb32b
commit 0f16edfe61
5 changed files with 68 additions and 142 deletions
+26 -34
View File
@@ -44,7 +44,6 @@ type direction struct {
streams []lossWindow
txFDs []int
rxFDs []int
reports chan string
probeSpec *frameSpec
probeTxFD int
@@ -59,8 +58,8 @@ type direction struct {
drops uint64
errBase sample
dropBase uint64
nicTX nicCounters
nicRX nicCounters
nicNow uint64
nicBase uint64
}
// Smoothing has to be steady against high-frequency noise yet still chase a
@@ -192,6 +191,7 @@ type sample struct {
crcErr, badMagic uint64
badLen uint64
txErrs, txShort uint64
rxErrs uint64
}
func lookupEndpoint(name string) (endpoint, error) {
@@ -247,6 +247,7 @@ func (d *direction) snapshot() sample {
s.crcErr += r.crcErr.Load()
s.badMagic += r.badMagic.Load()
s.badLen += r.badLen.Load()
s.rxErrs += r.rxErrs.Load()
}
for i := range d.streams {
s.lost += d.streams[i].lost.Load()
@@ -264,6 +265,7 @@ func (d *direction) reset() {
d.dropBase = d.drops
d.heldFrames = heldValue{}
d.heldSent = heldValue{}
d.nicBase = d.nicNow
d.cable.reset()
}
@@ -300,7 +302,8 @@ var intervalCols = []colSpec{
{title: "CRC", width: 7, right: true},
{title: "BADMAG", width: 7, right: true},
{title: "KDROP", width: 11, right: true},
{title: "ERRORS", width: 11, right: true},
{title: "LINK", width: 13, right: true},
{title: "ERRORS", width: 13, right: true},
{title: "MIN ns", width: 9, right: true},
{title: "LEN m", width: 6, right: true},
}
@@ -314,7 +317,8 @@ type view struct {
rxFrames, rxGot uint64
lost, late uint64
crc, badMagic uint64
kdrop, errors uint64
kdrop, link uint64
errors uint64
cable cableView
}
@@ -334,7 +338,12 @@ func (d *direction) counters(now sample) view {
kdrop: d.drops - d.dropBase,
cable: d.cable.view(),
}
v.errors = v.lost + v.crc + v.badMagic + (now.badLen - b.badLen) + v.kdrop
// A frame the stack refused and a frame the driver dropped are the same
// failure seen from either side of the ring, and never the same frame twice:
// a send that fails never reaches the driver to be dropped.
v.link = d.nicNow - d.nicBase +
(now.txErrs - b.txErrs) + (now.rxErrs - b.rxErrs)
v.errors = v.lost + v.crc + v.badMagic + (now.badLen - b.badLen) + v.kdrop + v.link
return v
}
@@ -398,25 +407,19 @@ func (d *direction) row(elapsed time.Duration, v view, target float64, length st
statusCell(v.crc),
statusCell(v.badMagic),
statusCell(v.kdrop),
statusCell(v.link),
statusCell(v.errors),
paint(v.cable.minText(), cCyan),
paint(length, cCyan),
}
}
func (d *direction) reportNIC() []string {
// Read once a second rather than per frame, since these are sysfs files; the
// display uses whatever the last sample left behind.
func (d *direction) sampleNIC() {
tx := readNIC(d.tx.name)
rx := readNIC(d.rx.name)
var parts []string
if s := tx.diff(d.nicTX); s != "" {
parts = append(parts, paint(" ▲ "+d.tx.name+" tx-side: "+s, cYellow))
}
if s := rx.diff(d.nicRX); s != "" {
parts = append(parts, paint(" ▲ "+d.rx.name+" rx-side: "+s, cYellow))
}
d.nicTX = tx
d.nicRX = rx
return parts
d.nicNow = tx.tx + rx.rx + tx.carrierDown
}
func buildDirection(label string, tx, rx endpoint, patIdx int, sizes []int, cfg config) (*direction, error) {
@@ -426,10 +429,8 @@ func buildDirection(label string, tx, rx endpoint, patIdx int, sizes []int, cfg
tx: tx,
rx: rx,
streams: newLossWindows(cfg.streams),
reports: make(chan string, 64),
cable: newCableStats(),
}
d.nicTX = readNIC(tx.name)
d.nicRX = readNIC(rx.name)
for i := 0; i < cfg.streams; i++ {
et := uint16(etherBase + i)
@@ -453,7 +454,6 @@ func buildDirection(label string, tx, rx endpoint, patIdx int, sizes []int, cfg
// The probe carries its own ethertype so it lands on a socket of its own, but
// it is left unsteered: it is a few frames a second and does not need a queue
// to itself, and the stamps are taken at the wire either way.
d.cable = newCableStats()
d.probeSpec = newFrameSpec(patIdx, rx.mac, tx.mac, cfg.probeEther, []int{probeSize})
fd, err := openTxSocket(tx.idx)
if err != nil {
@@ -471,6 +471,10 @@ func buildDirection(label string, tx, rx endpoint, patIdx int, sizes []int, cfg
return nil, fmt.Errorf("%s probe rx timestamps: %w", label, err)
}
d.probeRxFD = fd
// Whatever the interfaces have counted before now is not ours.
d.sampleNIC()
d.nicBase = d.nicNow
return d, nil
}
@@ -497,7 +501,6 @@ func (d *direction) start(wg *sync.WaitGroup, doneTx, doneRx *atomic.Bool, cfg c
spec: d.specs[i],
stats: d.rxStats[i],
streams: d.streams,
reports: d.reports,
ready: rxReady,
}
wg.Add(1)
@@ -762,6 +765,7 @@ func run(aName, bName, sizesArg, patArg string,
// Length needs both directions, so every row is sampled before any of
// them is printed.
for i, d := range dirs {
d.sampleNIC()
rows[i] = d.view(&d.prevConsole, secs)
}
length := "-"
@@ -772,18 +776,6 @@ func run(aName, bName, sizesArg, patArg string,
for _, line := range stats.emit(d.row(elapsed, rows[i], target, length)) {
fmt.Println(line)
}
for _, line := range d.reportNIC() {
fmt.Println(line)
}
for {
select {
case msg := <-d.reports:
fmt.Println(paint(" ✗ "+d.short+": "+msg, cRed))
continue
default:
}
break
}
}
}
}