Send a merge's builds to every machine at once, with no first-machine gate (hq ADR 0360)
mesh/delivery superseded: a newer head of the same pull request
mesh/merge-gate head da8a5e9bc0f7 merged into trunk 76304d9db817 — pass: builds mesh-controller → novox; no bus step; every machine composes with the chang…
mesh/repo-check head da8a5e9bc0f7 merged into trunk 76304d9db817 — fail: its merge-check.sh failed: --- FAIL: TestAReusedCheckBuildIsRegisteredByItsDigestA…

The operator chose the leanest rollout: no gate, no one machine first, failures loud and fixed forward.
A walk sends every rolled-out module to every machine running it in one step, built-by orders no tier,
and a walk ends on its machines' reports. The backlog walk sends every machine at once and passes each
on its applied report.
This commit is contained in:
2026-10-14 19:19:49 +02:00
parent 76304d9db8
commit da8a5e9bc0
7 changed files with 168 additions and 315 deletions
+84 -52
View File
@@ -43,29 +43,15 @@ func tiersOf(set []string, edges []inventory.Edge) [][]string {
for _, m := range set {
deps[m] = map[string]bool{}
}
// The build seat's holders follow the controller that defines their worker (EdgeWorkerOf,
// novox/hq issue 206), so the built-by edge from that controller to such a holder yields: the
// controller is built by whichever build machine is running, as the runtime image always was.
worker := map[string]map[string]bool{}
for _, e := range edges {
if e.Kind == inventory.EdgeWorkerOf && in[e.From] && in[e.To] {
if worker[e.To] == nil {
worker[e.To] = map[string]bool{}
}
worker[e.To][e.From] = true
}
}
for _, e := range edges {
// A code dependency — B packages A's source — rebuilds B with A, in the same tier: B's
// build needs nothing of A's first. The other kinds order: stands-on and declared after
// the base is built, built-by after the build machine is built and running — except for
// what the build machine itself stands on, and for the controller whose worker the build
// machine binds. The runtime image is built by the builder and the builder is built on the
// runtime image; the image comes first, built by the builder that is running.
// build needs nothing of A's first. Stands-on and declared order after the base is built, and the
// build seat's holder follows the controller that defines its worker (EdgeWorkerOf, issue 206).
if !in[e.From] || !in[e.To] || e.From == e.To || e.Kind == inventory.EdgePackages {
continue
}
if e.Kind == inventory.EdgeBuiltBy && (isBaseOf(e.From, e.To, edges, in) || worker[e.From][e.To]) {
// **Being built by the builder orders nothing** (novox/hq ADR 0360 §3): whichever builder runs builds it.
if e.Kind == inventory.EdgeBuiltBy {
continue
}
deps[e.From][e.To] = true
@@ -106,29 +92,6 @@ func tiersOf(set []string, edges []inventory.Edge) [][]string {
return tiers
}
// isBaseOf says whether `to` stands on `base`, directly or through other bases in the set, along
// the build edges alone.
func isBaseOf(base, to string, edges []inventory.Edge, in map[string]bool) bool {
seen := map[string]bool{}
var walk func(string) bool
walk = func(m string) bool {
if m == base {
return true
}
if seen[m] {
return false
}
seen[m] = true
for _, e := range edges {
if e.From == m && in[e.To] && (e.Kind == inventory.EdgeStandsOn || e.Kind == inventory.EdgeDeclared) && walk(e.To) {
return true
}
}
return false
}
return walk(to)
}
// reachableFrom is the moved modules plus everything that depends on them, through every layer:
// what a merge rebuilds. Along the code and build edges only: a module *built by* the build machine
// is not changed by a new build machine, so a built-by edge orders and gates a plan and never
@@ -164,7 +127,7 @@ func reachableFrom(moved []string, edges []inventory.Edge) []string {
}
// hasCycle says whether the tiers' last tier holds modules that still depend on each other. A packages
// edge orders nothing (tiersOf), so a module and what packages its source share a tier by rule: that is
// or built-by edge orders nothing (tiersOf), so a module and what packages its source share a tier by rule: that is
// no cycle, and saying one was is a false report in every such plan's log.
func hasCycle(tiers [][]string, edges []inventory.Edge) bool {
if len(tiers) == 0 {
@@ -175,7 +138,7 @@ func hasCycle(tiers [][]string, edges []inventory.Edge) bool {
last[m] = true
}
for _, e := range edges {
if e.Kind == inventory.EdgePackages {
if e.Kind == inventory.EdgePackages || e.Kind == inventory.EdgeBuiltBy {
continue
}
if last[e.From] && last[e.To] {
@@ -221,12 +184,6 @@ func gates(p inventory.Plan, edges []inventory.Edge, rollsOut func(string) bool)
return nil
}
inTier := map[string]bool{}
all := map[string]bool{}
for _, tier := range p.Tiers {
for _, m := range tier {
all[m] = true
}
}
for _, m := range p.Tiers[p.Tier] {
inTier[m] = true
}
@@ -241,8 +198,8 @@ func gates(p inventory.Plan, edges []inventory.Edge, rollsOut func(string) bool)
for _, e := range edges {
// A delivery group's order inside a walk (novox/hq ADR 0276 decision 5) gates as built-by does: the
// member before is running on its machines before the member after is asked.
ordered := e.Kind == edgeGroupOrder ||
e.Kind == inventory.EdgeBuiltBy && !isBaseOf(e.From, e.To, edges, all)
// A built-by edge waits for nothing since novox/hq ADR 0360 §3.
ordered := e.Kind == edgeGroupOrder
if later[e.From] && inTier[e.To] && ordered && !seen[e.To] && rollsOut(e.To) {
seen[e.To] = true
out = append(out, e.To)
@@ -888,6 +845,9 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
if err != nil {
return false, err
}
// **Every machine at once** (novox/hq ADR 0360 §1, §2): a module that rolls out is sent to every machine
// running it in one step, with no first machine and no gate, whatever module it is.
policy.Together = true
var reports []inventory.Reported
if !policy.Together {
// Read for the choice of the first machine as well as for its report.
@@ -1117,15 +1077,87 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
return changed, nil
}
}
// **A walk ends on its machines' reports** (novox/hq ADR 0360 §4, §5): after its last tier, every machine it sent
// has reported on that send, or is silent past ApplySilentAfter; never a health reading.
end := ""
if p.Tier == len(p.Tiers)-1 {
reports, err := inv.LastReports(ctx)
if err != nil {
return false, err
}
waiting, failed, silent := walkReports(*p, reports, time.Now().UTC())
if len(waiting) > 0 {
note := "sent; waiting for " + strings.Join(waiting, ", ") + " to report on their send"
changed := p.State != inventory.PlanRolling || p.Note != note
p.State = inventory.PlanRolling
p.Note = note
return changed, nil
}
var said []string
if len(failed) > 0 {
said = append(said, "failed or refused its send (apply-failing says it): "+strings.Join(failed, ", "))
}
if len(silent) > 0 {
said = append(said, fmt.Sprintf("said nothing on its send within %s: %s", inventory.ApplySilentAfter,
strings.Join(silent, ", ")))
}
end = strings.Join(said, "; ")
}
p.Tier++
p.State = inventory.PlanBuilding
p.Note = ""
p.Note = end
if p.Tier < len(p.Tiers) {
fmt.Printf("%s: tier %d done; asking tier %d: %s\n", p.ID, p.Tier-1, p.Tier, strings.Join(p.Tiers[p.Tier], ", "))
}
return true, nil
}
// walkReports is, for every machine a walk sent its rest, whether it reported on that send (novox/hq ADR 0360 §4):
// those still waiting, those that reported a failed or refused apply, and those silent past ApplySilentAfter,
// each sorted. Pure.
func walkReports(p inventory.Plan, reports []inventory.Reported, now time.Time) (waiting, failed, silent []string) {
byNode := map[string]inventory.Reported{}
for _, r := range reports {
byNode[r.Node] = r
}
seen := map[string]bool{}
for _, s := range p.Modules {
if s == nil {
continue
}
for node, sent := range s.Rest {
if seen[node] {
continue
}
r := byNode[node]
at := sent.At
if at.IsZero() && s.SentAt != nil {
at = *s.SentAt
}
switch {
case r.At != nil && sent.ReportsOn(r):
seen[node] = true
if r.Outcome == inventory.OutcomeFailed || r.Outcome == inventory.OutcomeRefused {
failed = append(failed, node)
}
case !at.IsZero() && now.Sub(at) > inventory.ApplySilentAfter:
seen[node] = true
silent = append(silent, node)
default:
if !slices.Contains(waiting, node) {
waiting = append(waiting, node)
}
}
}
}
// A machine one module's send was reported on and another's not yet is the same send: reported.
waiting = slices.DeleteFunc(waiting, func(n string) bool { return seen[n] })
sort.Strings(waiting)
sort.Strings(failed)
sort.Strings(silent)
return waiting, failed, silent
}
// restOf is, for one module, every machine of its rest that the send reached and what it carried there.
func restOf(to, sent []string, what map[string]inventory.SentDeclaration) map[string]inventory.SentDeclaration {
out := map[string]inventory.SentDeclaration{}