claude-code: the memory watch only notes a change, so it starts on a node with hundreds of files (issue 367)
mesh/merge-gate head e46eeb829853 merged into trunk 4d15a155e634 — pass: builds claude-code → ace, g14, novox, shanks; no bus step; every machine composes …
mesh/repo-check head e46eeb829853 merged into trunk 4d15a155e634 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered

The runtime hands a watch its current values one at a time and gives them ten
seconds. The handler read the state, waited for a running look and wrote files
for each value, so with about 450 memory values the watch never started on g14
or shanks, and a file written on g14 never reached shanks. The handler now
notes each key's newest revision and returns; a worker applies what was noted
at the state's current version.
This commit is contained in:
2026-10-13 16:40:41 +02:00
parent 5f491e0c09
commit e46eeb8298
2 changed files with 124 additions and 7 deletions
@@ -106,11 +106,16 @@ func (s *memState) pump(t *testing.T) {
}
for _, c := range todo {
for _, w := range s.watchers {
if err := w.Take(c.key, c.op, c.value, c.rev); err != nil && !errors.Is(err, errChanged) {
if err := w.Take(c.key, c.op, c.value, c.rev); err != nil {
t.Logf("%s taking %s: %v", w.Node, c.key, err)
}
}
}
for _, w := range s.watchers {
if err := w.Drain(); err != nil && !errors.Is(err, errChanged) {
t.Logf("%s applying: %v", w.Node, err)
}
}
}
t.Fatal("the state never settled")
}
@@ -137,6 +142,7 @@ func newMemNode(t *testing.T, root, name, kind string, state *memState, c *clock
}
}
n.MemoryCarrier = &MemoryCarrier{Home: home, Node: name, State: stateDir, Kind: kind, Store: state, Now: c.Now,
wake: make(chan struct{}, 1),
Notify: func(a map[string]any) error { n.notices = append(n.notices, a); return nil }}
state.watchers = append(state.watchers, n.MemoryCarrier)
return n
@@ -601,10 +607,13 @@ func (s *memState) replay(t *testing.T, n *memNode) {
keys, _ := s.Keys()
for _, k := range keys {
raw, rev, _, _ := s.Get(k)
if err := n.Take(k, "put", raw, rev); err != nil && !errors.Is(err, errChanged) {
if err := n.Take(k, "put", raw, rev); err != nil {
t.Logf("%s taking %s: %v", n.Node, k, err)
}
}
if err := n.Drain(); err != nil && !errors.Is(err, errChanged) {
t.Logf("%s applying: %v", n.Node, err)
}
}
// §6: a node that was away while a file changed twice takes the newest version, whose base it never saw, because
@@ -902,3 +911,58 @@ func TestARemovalReachesANodeThatWasAway(t *testing.T) {
t.Fatal("the version removed on the node that was away was not kept")
}
}
// slowState answers every read only once it is let go, as a state a whole bus away does under load.
type slowState struct {
*memState
gate chan struct{}
}
func (s slowState) Get(key string) (json.RawMessage, uint64, bool, error) {
<-s.gate
return s.memState.Get(key)
}
// Issue 367, seen live: the runtime hands a watch its current values one at a time and gives them all ten seconds;
// shanks's handler read the state and waited for its look for each of a few hundred values, so the watch never
// started and nothing written on g14 reached it. The handler now only notes a change: it returns at once while a
// look holds the carrier and while the state is slow to answer, and Drain applies what it noted.
func TestTheWatchHandlerReturnsAtOnceAndDrainAppliesWhatItNoted(t *testing.T) {
state, c, g14, shanks := twoNodes(t)
g14.write(t, project, "MEMORY.md", "index\n", c.now)
g14.look(t)
state.mu.Lock()
state.pending = nil
state.mu.Unlock()
slow := slowState{memState: state, gate: make(chan struct{})}
shanks.Store = slow
key := MemoryKey(HomeOperator, project, "MEMORY.md")
raw, rev, _, _ := state.Get(key)
shanks.mu.Lock() // a look running
done := make(chan error, 1)
start := time.Now()
for i := 0; i < 400; i++ {
go func() { done <- shanks.Take(key, "put", raw, rev) }()
}
for i := 0; i < 400; i++ {
select {
case err := <-done:
if err != nil {
t.Fatal(err)
}
case <-time.After(2 * time.Second):
t.Fatalf("the watch handler waited for the look or the state (%d of 400 returned)", i)
}
}
if took := time.Since(start); took > time.Second {
t.Fatalf("400 values took %s to hand over", took)
}
shanks.mu.Unlock()
close(slow.gate)
if err := shanks.Drain(); err != nil {
t.Fatal(err)
}
if got, _ := shanks.read(t, project, "MEMORY.md"); got != "index\n" {
t.Fatalf("what the watch noted was not applied: %q", got)
}
}