Bound what a refusal costs the engine, and let the busy tests fail instead of hang
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery delivered

A refusal waiting for its request parsed it into a 64 KiB reader for up to 5 s,
on a socket every account may open. It now drains at most 4 KiB through a small
buffer for at most 1 s and never past the accept-time deadline, only 32 wait at
once, and an unknown account is refused before it takes a slot. The busy tests
wait for the first line with a timeout and say what came of it (novox/hq issue 434).
This commit is contained in:
2026-10-11 02:47:11 +02:00
parent 70cb613614
commit f9805b2113
2 changed files with 132 additions and 21 deletions
+78 -4
View File
@@ -4,6 +4,7 @@ import (
"context"
"encoding/json"
"errors"
"io"
"net"
"os"
"os/user"
@@ -172,6 +173,25 @@ func keysOf(t *testing.T, body []byte) string {
return strings.Join(keys, " ")
}
// holdALine asks one line and returns once the engine is asking it, so it holds its account's slot. A first line
// that ends or never arrives fails the test at once, with what came of it, instead of hanging the suite (novox/hq
// issue 434).
func holdALine(t *testing.T, socket string, started <-chan struct{}) {
t.Helper()
first := make(chan error, 1)
go func() {
_, err := cli.Ask(context.Background(), socket, []string{"push"})
first <- err
}()
select {
case <-started:
case err := <-first:
t.Fatalf("the first line ended before it was asked: %v", err)
case <-time.After(5 * time.Second):
t.Fatal("the first line was never asked")
}
}
// **One account cannot hold the socket** (review of ADR 0272): its lines beyond its own bound are answered busy at
// once, nothing is asked, and the engine goes on accepting.
func TestOneAccountsLinesBeyondItsBoundAreAnsweredBusy(t *testing.T) {
@@ -187,8 +207,7 @@ func TestOneAccountsLinesBeyondItsBoundAreAnsweredBusy(t *testing.T) {
})
// Released before the engine is stopped: cleanups run last first, and the engine waits for its lines.
t.Cleanup(func() { close(release) })
go func() { _, _ = cli.Ask(context.Background(), socket, []string{"push"}) }()
<-started
holdALine(t, socket, started)
type asked struct {
a cli.Answer
err error
@@ -229,8 +248,7 @@ func TestARefusalWaitsForTheRequestItAnswers(t *testing.T) {
return cli.Answer{}
})
t.Cleanup(func() { close(release) })
go func() { _, _ = cli.Ask(context.Background(), socket, []string{"push"}) }()
<-started
holdALine(t, socket, started)
conn, err := net.Dial("unix", socket)
if err != nil {
t.Fatal(err)
@@ -256,6 +274,62 @@ func TestARefusalWaitsForTheRequestItAnswers(t *testing.T) {
}
}
// **Refusals that wait are bounded** (novox/hq issue 434, review of ADR 0272): with refusalsWaiting already full, a
// further refused line is answered at once, without waiting for its request.
func TestARefusalOverTheWaitingBoundIsAnsweredAtOnce(t *testing.T) {
was, wasWithin, wasWaiting := perAccount, refusedWithin, refusalsWaiting
perAccount, refusedWithin = 1, time.Minute
refusalsWaiting = make(chan struct{}, 1)
refusalsWaiting <- struct{}{}
t.Cleanup(func() { perAccount, refusedWithin, refusalsWaiting = was, wasWithin, wasWaiting })
release := make(chan struct{})
started := make(chan struct{}, 2)
socket := serving(t, func(context.Context, Asked) cli.Answer {
started <- struct{}{}
<-release
return cli.Answer{}
})
t.Cleanup(func() { close(release) })
holdALine(t, socket, started)
conn, err := net.Dial("unix", socket)
if err != nil {
t.Fatal(err)
}
defer conn.Close()
// Nothing is written: a refusal that waited would wait refusedWithin, a minute, for it.
_ = conn.SetDeadline(time.Now().Add(5 * time.Second))
a, err := cli.ReadAnswer(conn)
if err != nil || !strings.Contains(a.Refused, "busy") {
t.Fatalf("a refusal over the waiting bound reads %+v (%v)", a, err)
}
}
// countingReader is an endless request with no line end, counting what was read of it.
type countingReader struct{ read int }
func (c *countingReader) Read(p []byte) (int, error) {
for i := range p {
p[i] = 'x'
}
c.read += len(p)
return len(p), nil
}
// **A refused request is drained, not kept** (novox/hq issue 434): what a refused client writes is read up to
// drainCap bytes and no further, however much it sends.
func TestARefusedRequestIsDrainedNoFurtherThanItsCap(t *testing.T) {
r := &countingReader{}
drain(r)
if r.read != drainCap {
t.Fatalf("drained %d bytes of an endless request; the cap is %d", r.read, drainCap)
}
after := &countingReader{}
drain(io.MultiReader(strings.NewReader("{\"line\":[\"status\"]}\n"), after))
if after.read != 0 {
t.Fatalf("draining read %d bytes past the end of the request line", after.read)
}
}
// **A process in another namespace is not this machine's account** (review of ADR 0272): a container running with
// the operator's user id shares the id and not the machine; refused, and nothing is asked.
func TestAPeerInAnotherNamespaceIsRefused(t *testing.T) {