Files
mesh-host/internal/selfcompose/data_test.go
T
jschoubben e031758e4a
mesh/delivery merged: source.main moved to novox/mesh-host@233bb23bbcb1
mesh/delivery-group group feat/data-items merged: every member is merged
mesh/check pass e031758e4a37 on 6c1f7cc7f50f | unit 4510 run, 4259 kept, 49 skipped | integration 4589 run, 4361 kept, 51 skipped | over bound: unit, …
Forget a kept build when its key is written again, and retry data items on the composer's bound (review round 1)
The builder claims a done key again as a new revision, so a build kept as done is dropped on any
write of its key. Data items that could not be worked out or written stay due and owe a pass after
the growing bound, so a first write refused before the state or grant exists is tried again
without waiting for an unrelated composition.

Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
2026-11-01 00:27:29 +01:00

196 lines
9.0 KiB
Go

package selfcompose
import (
"context"
"errors"
"slices"
"strings"
"sync"
"testing"
"time"
"git.novox.be/novox/mesh-sdk/go/catalogue"
"git.novox.be/novox/mesh-sdk/go/compose/frombus"
"git.novox.be/novox/mesh-sdk/go/delivery"
"github.com/novox/mesh-host/internal/agent"
)
// fakeData is the engine's state data as a map, counting the writes and deletes made.
type fakeData struct {
mu sync.Mutex
kv map[string][]byte
puts []string
deletes []string
fail error
}
func (f *fakeData) Get(_ context.Context, key string) ([]byte, error) {
f.mu.Lock()
defer f.mu.Unlock()
return f.kv[key], nil
}
func (f *fakeData) Put(_ context.Context, key string, value []byte) error {
f.mu.Lock()
defer f.mu.Unlock()
if f.fail != nil {
return f.fail
}
f.kv[key] = value
f.puts = append(f.puts, key)
return nil
}
func (f *fakeData) Delete(_ context.Context, key string) error {
f.mu.Lock()
defer f.mu.Unlock()
delete(f.kv, key)
f.deletes = append(f.deletes, key)
return nil
}
func (f *fakeData) Keys(_ context.Context, pattern string) ([]string, error) {
f.mu.Lock()
defer f.mu.Unlock()
var out []string
for k := range f.kv {
if matches(pattern, k) {
out = append(out, k)
}
}
return out, nil
}
// ADR 0577 clause 1 (and its check in clause 5): after a pass that composed something new, the engine keeps each placed
// module's data items in its state data, key `<node>.<module>`, with its composition's digest: written when the module's
// items change, not written when they do not, and deleted when the module leaves the node; a key of another node is never
// touched, a pass that handed nothing new reads nothing, and items that could not be worked out or written are tried
// again with no new composition, after the composer's growing bound (the loop's owed pass).
func TestTheEngineKeepsEachPlacedModulesDataItems(t *testing.T) {
item := func(module, id, path string) catalogue.PlacedDataItem {
return catalogue.PlacedDataItem{Module: module, ID: id, Class: "valuable", Path: path, Measure: catalogue.MeasureWalk}
}
store := &fakeData{kv: map[string][]byte{
// Left by a module that went while the engine was down, and another node's key.
"n1.gone": []byte(`{"v":1}`), "n2.web": []byte(`{"v":1}`)}}
var items map[string][]catalogue.PlacedDataItem
var asked int
var itemsErr error
setFor(t, &dataItemsOf, func(context.Context, frombus.Reader, frombus.Mesh, frombus.Scope) (map[string][]catalogue.PlacedDataItem, []string, error) {
asked++
return items, nil, itemsErr
})
// The engine's grant lets it read its own node's keys, as the listing at a run's first pass does, and no other's.
if !grantAllows(t, "n1", delivery.EngineDataBucket, "n1.*") || grantAllows(t, "n1", delivery.EngineDataBucket, "n2.*") {
t.Error("the node's grant reads other keys of the engine's state data than its own")
}
c := New(Env{Node: "n1", Data: store})
webKey, _ := delivery.ComposedKey("n1", "web", delivery.SourcePrefix2+strings.Repeat("a", 64))
nodeKey, _ := delivery.ComposedNodeKey("n1")
for _, step := range []struct {
name string
due bool
items map[string][]catalogue.PlacedDataItem
digests map[string]string // composed key to digest handed
fail error
refused error // what working out the items says
owed time.Duration // the bound the loop waits before trying again; 0 when nothing is owed
puts []string
deletes []string
asked int
// want is each key's items and digest after the step.
want map[string]delivery.EngineData
}{
{name: "the run's first pass writes each module's items and deletes what left", due: true,
items: map[string][]catalogue.PlacedDataItem{"web": {item("web", "db", "/srv/web/db")}, "oldie": {item("oldie", "x", "/srv/oldie")}, "bare": nil},
digests: map[string]string{webKey: "sha256:w1", nodeKey: "sha256:n1"},
puts: []string{"n1.oldie", "n1.web"}, deletes: []string{"n1.gone"}, asked: 1,
want: map[string]delivery.EngineData{
"n1.web": {Module: "web", Digest: "sha256:w1", Items: []catalogue.PlacedDataItem{item("web", "db", "/srv/web/db")}},
"n1.oldie": {Module: "oldie", Digest: "sha256:n1", Items: []catalogue.PlacedDataItem{item("oldie", "x", "/srv/oldie")}}}},
{name: "a pass that handed nothing new reads nothing", due: false, asked: 1},
{name: "a pass that changes no module's items writes nothing", due: true,
items: map[string][]catalogue.PlacedDataItem{"web": {item("web", "db", "/srv/web/db")}, "oldie": {item("oldie", "x", "/srv/oldie")}},
digests: map[string]string{webKey: "sha256:w1", nodeKey: "sha256:n1"}, asked: 2},
{name: "items that cannot be worked out are tried again after the bound", due: true, refused: errors.New("no answer"),
digests: map[string]string{webKey: "sha256:w2", nodeKey: "sha256:n1"}, asked: 3, owed: LaterWait(0)},
{name: "a write that fails is tried again with no new composition, after a longer bound", due: false, fail: errors.New("the bus said no"),
items: map[string][]catalogue.PlacedDataItem{"web": {item("web", "db", "/srv/web/db2")}, "oldie": {item("oldie", "x", "/srv/oldie")}},
digests: map[string]string{webKey: "sha256:w2", nodeKey: "sha256:n1"}, asked: 4, owed: LaterWait(1)},
{name: "a module whose composition changed its items is written again", due: false,
items: map[string][]catalogue.PlacedDataItem{"web": {item("web", "db", "/srv/web/db2")}, "oldie": {item("oldie", "x", "/srv/oldie")}},
digests: map[string]string{webKey: "sha256:w2", nodeKey: "sha256:n1"}, puts: []string{"n1.web"}, asked: 5,
want: map[string]delivery.EngineData{
"n1.web": {Module: "web", Digest: "sha256:w2", Items: []catalogue.PlacedDataItem{item("web", "db", "/srv/web/db2")}}}},
{name: "a module that left the node is deleted", due: true,
items: map[string][]catalogue.PlacedDataItem{"oldie": {item("oldie", "x", "/srv/oldie")}},
digests: map[string]string{nodeKey: "sha256:n1"}, deletes: []string{"n1.web"}, asked: 6},
} {
t.Run(step.name, func(t *testing.T) {
store.mu.Lock()
store.puts, store.deletes, store.fail = nil, nil, step.fail
store.mu.Unlock()
items, itemsErr = step.items, step.refused
c.paid() // the pass before composed: nothing owed
c.mu.Lock()
c.handed = step.digests
c.dataDue = c.dataDue || step.due
c.mu.Unlock()
c.publishData(context.Background())
slices.Sort(store.puts)
if !slices.Equal(store.puts, step.puts) || !slices.Equal(store.deletes, step.deletes) || asked != step.asked {
t.Fatalf("wrote %v and deleted %v, items asked %d times; want %v, %v, %d", store.puts, store.deletes, asked,
step.puts, step.deletes, step.asked)
}
if wait, _ := c.waitNow(); c.owed.Load() != (step.owed > 0) || wait != step.owed {
t.Errorf("owed a pass: %v after %s; want after %s", c.owed.Load(), wait, step.owed)
}
for key, want := range step.want {
got, err := delivery.DecodeEngineData(store.kv[key])
if err != nil || got.Node != "n1" || got.Module != want.Module || got.Digest != want.Digest ||
!slices.Equal(got.Items, want.Items) {
t.Errorf("%s holds %+v (%v), want %+v", key, got, err, want)
}
}
if store.kv["n2.web"] == nil {
t.Error("another node's key was deleted")
}
})
}
// The items are worked out from what the pass read: a done build is read over the bus once, not again for the data
// items after the pass (a module in a running build is read again, since it may land).
b := newFakeBus()
done, _ := delivery.EncodeBuilderBuild(delivery.BuilderBuild{State: delivery.BuildDone, Manifest: delivery.ManifestDigest([]byte(lemurs))})
running, _ := delivery.EncodeBuilderBuild(delivery.BuilderBuild{State: "building"})
fp := delivery.SourcePrefix2 + strings.Repeat("b", 64)
doneKey, _ := delivery.BuilderBuildKey("web", fp)
runningKey, _ := delivery.BuilderBuildKey("pg", fp)
b.kv[delivery.BuilderBuildsBucket+"/"+doneKey], b.kv[delivery.BuilderBuildsBucket+"/"+runningKey] = done, running
r := reader{c: New(b.env())}
for range 3 {
if got, err := r.Build(context.Background(), "web", fp); err != nil || got.State != delivery.BuildDone {
t.Fatalf("the done build: %+v, %v", got, err)
}
if got, err := r.Build(context.Background(), "pg", fp); err != nil || got.State != "building" {
t.Fatalf("the running build: %+v, %v", got, err)
}
}
reads := map[string]int{}
for _, read := range b.reads {
reads[read[1]]++
}
if reads[doneKey] != 1 || reads[runningKey] != 3 {
t.Errorf("the builds were read %v times over the bus; want the done one once, the running one each time", reads)
}
// A done key the builder claims again (a gone step, issue 1937) and lands done again is read again, not served kept.
again, _ := delivery.EncodeBuilderBuild(delivery.BuilderBuild{State: delivery.BuildDone, Manifest: delivery.ManifestDigest([]byte("{}"))})
b.mu.Lock()
b.kv[delivery.BuilderBuildsBucket+"/"+doneKey] = again
b.mu.Unlock()
r.c.takeBuildDone(agent.Update{Key: doneKey, Value: running})
if got, err := r.Build(context.Background(), "web", fp); err != nil || got.Manifest != delivery.ManifestDigest([]byte("{}")) {
t.Errorf("a done build written again is served as %+v (%v), not read again", got, err)
}
}