A grant secret belongs to whoever provisions, and the sweep skips what it will not address

Issue 225. The mesh seals one credential per consumer beside the provider's
contributions file, and wrote it root-owned. That was right while a module's
own code ran in a container as root; ADR 0198 moved that code under the node's
runtime, as the node's account, and the secret stayed root's. On the control
machine two consumers went unprovisioned for three hours and the only sign
was a line reading 'secret not readable yet', 4330 times.

The same sentence is already written for a module's own secrets a few hundred
lines above — 'a root-owned 0600 file is one that process cannot read'. This
is that rule reaching the other kind of secret the mesh writes for a module.

Issue 226. The sweep met a reference recorded with the store's old address,
read 'I will not address this' as 'the store refuses everything', and
collected none of the 1681 it had found. Two changes: references from build
records are read through Recorded, where the provenance is known — not in
LetGo, which cannot tell one registry host from another and must stay strict
— and a reference the sweep will not address is now ErrNotOurs, skipped,
never a reason to stop. Only the store refusing ends a sweep.

make check: the two failures both fail on main as well — the resolver test
(hq 202/203) and the service-manager test, which reads this machine's own
shell environment.
This commit is contained in:
2026-10-04 12:21:49 +02:00
parent 912e9f4e85
commit 41b20b2782
7 changed files with 263 additions and 19 deletions
+30 -8
View File
@@ -4,6 +4,8 @@ import (
"context"
"encoding/json"
"strings"
"github.com/novox/mesh-controller/internal/catalogue"
)
// What the artifact store keeps, and what it may let go (novox/hq ADR 0189, issue 108).
@@ -71,11 +73,12 @@ func (i *Inventory) ToCollect(ctx context.Context) ([]string, error) {
continue
}
for _, a := range made {
if a.Reference == "" || keep[a.Reference] || collected[a.Reference] || seen[a.Reference] {
reference := asRecorded(a.Reference)
if reference == "" || keep[reference] || collected[reference] || seen[reference] {
continue
}
seen[a.Reference] = true
out = append(out, a.Reference)
seen[reference] = true
out = append(out, reference)
}
}
return out, rows.Err()
@@ -127,8 +130,8 @@ func (i *Inventory) keptReferences(ctx context.Context) (map[string]bool, error)
continue
}
for _, a := range made {
if a.Reference != "" {
keep[a.Reference] = true
if reference := asRecorded(a.Reference); reference != "" {
keep[reference] = true
}
}
}
@@ -186,16 +189,35 @@ func (i *Inventory) everyReferenceMade(ctx context.Context) ([]string, error) {
continue
}
for _, a := range made {
if a.Reference == "" || seen[a.Reference] {
reference := asRecorded(a.Reference)
if reference == "" || seen[reference] {
continue
}
seen[a.Reference] = true
out = append(out, a.Reference)
seen[reference] = true
out = append(out, reference)
}
}
return out, rows.Err()
}
// asRecorded is an artifact reference in the one vocabulary the sweep speaks (novox/hq issue 226).
//
// **Every reference here came from a build record, so every one of them is the mesh's own.** That
// is what makes it safe to normalise: references kept before the store's address stopped being
// written are `<host>:<port>/<path>@sha256:…` (04-ISSUES/102), and `Recorded` reads those as the
// `artifact-store://` references the rest of the mesh uses. Done here rather than when the store
// is asked, because `Recorded` cannot tell one registry host from another — only the provenance
// can, and the provenance is here.
//
// The oldest artifacts are exactly the ones recorded the old way, and exactly the ones a
// sweep reaches first. Untranslated, the first of them ended every sweep.
func asRecorded(reference string) string {
if reference == "" {
return ""
}
return catalogue.Recorded(reference)
}
// digestIn is the `sha256:<hex>` a reference names, empty when it names none.
func digestIn(reference string) string {
for _, marker := range []string{"@sha256:", "/sha256:"} {
+45
View File
@@ -145,3 +145,48 @@ func TestAFailedBuildNamesNothingToCollectAndEachModuleIsCountedOnItsOwn(t *test
t.Fatalf("offered %v; want only web's oldest — db's three are all within its five", go_)
}
}
// An artifact recorded with the store's old address is offered for collection, in the vocabulary
// the rest of the mesh speaks (novox/hq issue 226).
//
// Before references were kept without an address the mesh recorded
// `<host>:<port>/<path>@sha256:…` (04-ISSUES/102). Those are the oldest artifacts, which makes
// them exactly the ones an oldest-first sweep reaches first — and the first live run met one,
// read "I will not address this" as "the store refuses everything", and collected none of 1681.
func TestAnArtifactRecordedWithAnAddressIsOfferedAsTheMeshRecordsOne(t *testing.T) {
inv := fresh(t)
ctx := context.Background()
// The oldest build published the old way; five newer ones fill the module's five.
old := aBuild("a00", "tools", "")
old.Made = []Artifact{{Name: "build", Kind: "image",
Reference: "127.0.0.1:5100/tools/build@sha256:" + fmt.Sprintf("%064x", 1)}}
if err := inv.RecordBuild(ctx, old); err != nil {
t.Fatal(err)
}
for i := 2; i <= 6; i++ {
built(t, inv, fmt.Sprintf("a%02d", i), "tools", i)
}
go_, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
want := ref("tools", "build", 1)
if len(go_) != 1 || go_[0] != want {
t.Fatalf("offered %v; want %q — the address is a route to the artifact, not part of its "+
"name, and the sweep speaks the name", go_, want)
}
// And marking it collected uses that same name, so the next sweep does not offer it again
// under a spelling it has not seen.
if err := inv.MarkCollected(ctx, go_); err != nil {
t.Fatal(err)
}
again, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
if len(again) != 0 {
t.Fatalf("offered %v again after collecting it", again)
}
}