Compare commits

..
6 Commits
Author SHA1 Message Date
mesh-tools 883d976b5f Merge pull request 'Builder: announce a done build again, so the controller composes a module whose composition and grant were lost (recovery of novox and ace, 2026-10-29)' (#111) from fix/announce-done-again into main 2026-10-29 19:47:58 +00:00
jschoubben ba4bfea0a4 Remove the PR description draft from the branch; it is kept outside the repository
mesh/check checking ba4bfea0a407 on 883d976b5f00
mesh/delivery merged: source.main moved to novox/mesh-delivery@883d976b5f00
2026-10-29 17:59:28 +01:00
jschoubben fb7ad5cd73 Add builder_announce_again so a done build can be taken in and composed again
Recovery of photos and link2pay on novox, 2026-10-29: their composition and grant were lost and no tool wrote a done key again.
Quick fix: rewrite the done key unchanged as a new revision. Thorough fix: ADR 0547, the node agent composes its own and the
assigner stops waiting on controller compositions (design 47).
2026-10-29 17:58:39 +01:00
mesh-tools 5e931eac19 Merge pull request 'The assigner never keeps a want back because a provider is unhealthy, and the canary gates only modules that run (issue 1960)' (#110) from fix/1960-no-hold-spread into main 2026-10-29 15:35:28 +00:00
jschoubben d54eab4052 Never keep a want back for an unhealthy provider, nor a first want for the canary where nothing runs: the module is wanted and its provider down is said (issue 1960)
mesh/delivery merged: source.main moved to novox/mesh-delivery@5e931eac191a
mesh/delivery-group group fix/1960-no-hold-spread merged: every member is merged
mesh/check pass d54eab4052d1 on c1b4533201c8 | unit 3318 run, 1951 kept, 38 skipped | integration 3994 run, 2125 kept, 38 skipped | over bound: unit, …
On novox on 2026-10-29 the assigner wrote no want for de-spiegel and link2pay (rule 5: no provider of route runs
healthy) nor photos (rule 5: no provider of s3-bucket), and none where no node of a module was online (rule 3). A
module with no want is on no list and nothing applies it; a site runs while the proxy in front of it is down.

- Quick fix: skip rules 3 and 5 for the three modules' first wants. Leaves rule 5 holding every move and the
  canary holding new assignments.
- Thorough fix, chosen: rule 5 holds no want; it is said per node (Outcome.Unreached, journaled once per words by the
  pass). The canary (rule 3) no longer holds the first want of a node that does not run the module. A node that runs
  it keeps the canary order: at an older composed and granted fingerprint of the new form its first want stays at what
  it runs; at an old-form build it waits for the canary as before (issue 1945); at an older fingerprint with no
  composition or grant it gets F. This departs from ADR 0429 §6, which makes every provider healthy a rule of every want.

Tests: rows changed in the existing rule tests, and rows for today's modules (route-proxy down: de-spiegel and
link2pay wanted; minio not healthy: photos moves; novox offline: a first want written, a move kept back).
2026-10-29 16:04:39 +01:00
mesh-tools c1b4533201 Merge pull request 'Revert L3: its node-engine removed every held module on novox, the bus server among them (issue 774)' (#109) from revert/774-l3 into main 2026-10-29 08:59:55 +00:00
10 changed files with 326 additions and 54 deletions
@@ -40,6 +40,9 @@ type passer struct {
// noWant is, by node and module, the reason last said why the node gets no want for the module (issue 1945): a reason is
// said in the journal when it appears or changes, and its clearing when the node has a want again; one line per change.
noWant map[string]string
// unreached is, by node and module, the words last said why a wanted module may not be reached there (issue 1960: its
// provider is down, which holds no want): said once per words, and once more when it is reached again.
unreached map[string]string
// nodeFailed is, by node and module, the words of the condition last raised for a failure that is the node's own (ADR 0557
// §1, issue 1496): raised once per words, cleared once when it ends.
nodeFailed map[string]string
@@ -196,6 +199,7 @@ func (p *passer) wants(ctx context.Context) {
gatherNeither(neither, name, out)
leftNeither(left, m, wanted[name], out)
p.sayNoWant(name, m.Nodes, wanted[name], out)
p.sayUnreached(name, m.Nodes, out)
p.sayNodeFailed(name, fp, m.Nodes, nodes, out, first)
if first {
startClears(stale, m, out)
@@ -274,6 +278,26 @@ func (p *passer) sayNoWant(module string, nodes []string, before map[string]deli
}
}
// sayUnreached says in the journal, once per words, that a module is wanted on a node and may not be reached there, since
// no provider of what it requires runs healthy (rule 5, which holds no want: issue 1960), and once when it is reached again.
func (p *passer) sayUnreached(module string, nodes []string, out assign.Outcome) {
if p.unreached == nil {
p.unreached = map[string]string{}
}
for _, node := range nodes {
key := node + "." + module
why, last := out.Unreached[node], p.unreached[key]
switch {
case why != "" && why != last:
p.logf("%s on %s is wanted and may not be reached: %s", module, node, why)
p.unreached[key] = why
case why == "" && last != "":
p.logf("%s on %s is reached: its providers run healthy", module, node)
delete(p.unreached, key)
}
}
}
// sayNodeFailed raises, once per node and words, the condition of a failed health check of a module's newest fingerprint fp
// that is the node's own, since fp runs healthy elsewhere and is not held (ADR 0557 §1, issue 1496), and says it in the journal;
// it clears the condition once when the node no longer fails alone, or is no longer one of the module's nodes. The first pass
@@ -89,14 +89,15 @@ func Canary(nodes []string, online map[string]bool) string {
// Holds is whether every rule holds for moving node's want for m to m.FP, and if not, the first rule that does not,
// with why (ADR 0429 §6). Rule 4, at most `step` others ahead, always holds: the mesh keeps no step, and step 0 is no
// limit (ADR 0528 clause 3).
// limit (ADR 0528 clause 3). Rule 5, every provider healthy, holds no want (issue 1960): a module whose provider is down
// is wanted all the same, and said as not reached (Unreached), since a site runs while the proxy in front of it is
// down and one module never blocks another.
func Holds(m Module, node string, n Nodes) (bool, int, string) {
return holds(m, node, n, true)
}
// holds is Holds; with ordering false it leaves out the two rules that order a move among nodes and modules, the canary (3)
// and the providers (5), and keeps those without which there is nothing to want: the build done, not held, the grant and the
// composition.
// holds is Holds; with ordering false it leaves out the canary (3), which orders a move among nodes, and keeps those
// without which there is nothing to want: the build done, not held, the grant and the composition.
func holds(m Module, node string, n Nodes, ordering bool) (bool, int, string) {
switch {
case !m.Built:
@@ -112,11 +113,6 @@ func holds(m Module, node string, n Nodes, ordering bool) (bool, int, string) {
case node != canary && !healthy(n.Haves[canary][m.Name], m.FP):
return false, delivery.RuleCanary, fmt.Sprintf("the canary %s does not run %s healthy yet", canary, short(m.FP))
}
for _, provision := range m.Manifest.Requires {
if ok, why := providedHealthy(provision, node, n); !ok {
return false, delivery.RuleProviders, why
}
}
}
if !m.Granted {
return false, delivery.RuleGrant, fmt.Sprintf("the controller has not issued the credential of %s for %s", m.Name, short(m.FP))
@@ -157,6 +153,17 @@ func providedHealthy(provision, node string, n Nodes) (bool, string) {
return false, fmt.Sprintf("no provider of %s runs healthy on %s or at its canary", provision, node)
}
// unreached is why m on node may be wanted and not reached: the first provision it requires that no provider runs
// healthy on node or at its canary (rule 5, said and never holding a want; issue 1960); empty when all do.
func unreached(m Module, node string, n Nodes) string {
for _, provision := range m.Manifest.Requires {
if ok, why := providedHealthy(provision, node, n); !ok {
return why
}
}
return ""
}
// keepsData says whether a module holds data a set-back would lose: an own item of a class that keeps (irreplaceable,
// valuable or rebuildable; cache and none do not).
func keepsData(m catalogue.Manifest) bool {
@@ -186,6 +193,9 @@ type Outcome struct {
Restore []string
// Rules is, by node, the rule holding the node back from m.FP and why.
Rules map[string]Holding
// Unreached is, by node, why the module may be wanted there and not reached: no provider of a provision it requires runs
// healthy (rule 5, issue 1960). It holds no want back; the pass says it.
Unreached map[string]string
// Neither is, by node, the nodes the module is assigned to that run it (or whose have names it), hold no want for it after this
// pass and are not held (a held module is on the held list), with the rule holding the first want back: the node's own composition does not hold a module with a fingerprint of the new form, so such a module is on
// neither list and nothing applies it on that node (issue 1945). A module with no fingerprint of the new form goes with the node's
@@ -301,26 +311,33 @@ func Wants(m Module, n Nodes, wants map[string]delivery.Want, lastHealthy map[st
}
ok, rule, words := Holds(mm, node, n)
target := m.FP
if have, ran := n.Haves[node][m.Name]; !ok && (!had || before.Removed) && (rule == delivery.RuleCanary || rule == delivery.RuleProviders) && runs(have) {
// A node that runs the module and has no want for it has no want the ordering rules could hold it back to: holding its first
// want back leaves the module on neither list, since the node's own composition gives up every module with a fingerprint of
// the new form (issue 1945: on g14 the bundles of clipmenu, screen-lock and every module requiring x11-display were removed
// and never made again, because xorg's health is unknown and screen-lock's canary never ran it healthy). Its first want names
// what it runs, never a newer fingerprint, so the canary is never passed over: F when it runs F (by the rules without which
// there is nothing to want), or an older fingerprint of the new form whose composition and grant exist. Rules 3 and 5 then
// order the move from it as from any want.
switch _, composed := m.composition(node, have.FP); {
case !ran:
case have.FP == m.FP:
ok, rule, words = holds(mm, node, n, false)
case !strings.HasPrefix(have.FP, delivery.SourcePrefix2):
// An old-form build (not src2:) has no composition, so no want can ever name it: its first want names F, and only by every
// rule as usual, the canary (3) and the providers (5) included, which here hold it back. Nothing is written and the module
// stays on neither list until the canary runs F healthy; then Holds passes and the want names F (issue 1945, a design
// limit set by the lead: the canary stays in force and old-form modules leave neither list by themselves).
case composed && m.GrantedAt != nil && m.GrantedAt(have.FP):
ok, target = true, have.FP
if have, ran := n.Haves[node][m.Name]; !ok && (!had || before.Removed) && rule == delivery.RuleCanary {
// The canary orders a move, never the first want of a node that does not run the module (issue 1960: de-spiegel,
// link2pay and photos on novox were left with no want). A node that runs an older fingerprint of the new form,
// composed and granted, gets its first want at what it runs, and rule 3 then orders its move to F as from any want.
// A node that runs an old-form build, which no want can name, stays under the canary (issue 1945, a design limit):
// its first want names F once the canary runs F healthy. Any other node that runs it (F itself, or an older
// fingerprint with no composition or grant) gets F by the rules without which there is nothing to want.
older := ran && runs(have) && have.FP != m.FP && strings.HasPrefix(have.FP, delivery.SourcePrefix2)
if older {
_, composed := m.composition(node, have.FP)
older = composed && m.GrantedAt != nil && m.GrantedAt(have.FP)
}
oldForm := ran && runs(have) && have.FP != m.FP && !strings.HasPrefix(have.FP, delivery.SourcePrefix2)
switch {
case older:
ok, target = true, have.FP
case oldForm:
// held back by rule 3 as written: on neither list until the canary runs F healthy
default:
ok, rule, words = holds(mm, node, n, false)
}
}
if why := unreached(m, node, n); why != "" {
if out.Unreached == nil {
out.Unreached = map[string]string{}
}
out.Unreached[node] = why
}
if ok && target == m.FP && len(out.NodeFailed) > 0 && !(had && !before.Removed && before.FP == m.FP) &&
n.Haves[node][m.Name].FP != m.FP {
@@ -65,9 +65,6 @@ func TestEachRuleHoldsTheWantBack(t *testing.T) {
delivery.RuleCanary: func(_ *Module, n *Nodes) {
n.Haves["ace"]["mail"] = delivery.Have{FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
},
delivery.RuleProviders: func(_ *Module, n *Nodes) {
n.Haves["ace"]["store"] = delivery.Have{FP: fpOf('s'), Phase: delivery.PhaseRunning, Health: delivery.HealthWaits}
},
delivery.RuleGrant: func(m *Module, _ *Nodes) { m.Granted = false },
delivery.RuleComposed: func(m *Module, _ *Nodes) {
m.Composed = func(string, string) (delivery.ComposedDigest, bool) { return delivery.ComposedDigest{}, false }
@@ -86,6 +83,50 @@ func TestEachRuleHoldsTheWantBack(t *testing.T) {
t.Errorf("rule %d broken: g14's want is %+v", rule, w)
}
}
// 1960: rule 5 holds no want, and no module is left without a first want for a provider down or a node offline; the
// provider down is said (Unreached). novox on 2026-10-29: route-proxy down held de-spiegel and link2pay, minio not
// healthy held photos, and novox offline left a module with no canary.
for _, c := range []struct {
name, module, provision, provider string
have *delivery.Have // what novox runs of the module; nil: a new assignment
before map[string]delivery.Want
online bool
want string
}{
{"de-spiegel, route-proxy down, old form", "de-spiegel", "route", "route-proxy",
&delivery.Have{FP: "build-1", Phase: delivery.PhaseRunning, Health: delivery.HealthUnknown}, nil, true, fNew},
{"link2pay, route-proxy down, new assignment", "link2pay", "route", "route-proxy", nil, nil, true, fNew},
{"photos, minio not healthy, moving its want", "photos", "s3-bucket", "minio",
&delivery.Have{FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK},
map[string]delivery.Want{"novox": {FP: fOld, Composed: dig(fOld)}}, true, fNew},
{"novox offline, no canary, new assignment", "link2pay", "route", "route-proxy", nil, nil, false, fNew},
{"novox offline, no canary, moving its want", "photos", "s3-bucket", "minio",
&delivery.Have{FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK},
map[string]delivery.Want{"novox": {FP: fOld, Composed: dig(fOld)}}, false, fOld},
} {
m := Module{Name: c.module, FP: fNew, Built: true, Granted: true, Composed: composedEverywhere, Nodes: []string{"novox"},
Manifest: catalogue.Manifest{Module: c.module, Requires: []string{c.provision}}}
n := Nodes{Online: map[string]bool{"novox": c.online}, Haves: map[string]map[string]delivery.Have{"novox": {
c.provider: {FP: fpOf('p'), Phase: delivery.PhaseFailed, Health: delivery.HealthFailed}}},
Providers: map[string]map[string][]string{c.provision: {c.provider: {"novox"}}}}
if c.have != nil {
n.Haves["novox"][c.module] = *c.have
}
out := Wants(m, n, c.before, nil)
w, written := out.Wants["novox"]
if !written {
w = c.before["novox"]
}
if w.FP != c.want || w.Removed {
t.Errorf("%s: novox's want = %+v, want %s", c.name, w, short(c.want))
}
if _, on := out.Neither["novox"]; on {
t.Errorf("%s: on neither list: %+v", c.name, out.Neither)
}
if c.online && !strings.Contains(out.Unreached["novox"], c.provision) {
t.Errorf("%s: the provider down is not said: %q", c.name, out.Unreached["novox"])
}
}
// 1126/AC-SW1, AC-SW2, AC-SW3: a want written Self while g14 was placed, kept by a rule after the placement is gone, names
// the controller's composition of the fingerprint the rule keeps (on 2026-10-27 g14's systemd-resolved, held, and forticlient,
@@ -108,7 +149,7 @@ func TestEachRuleHoldsTheWantBack(t *testing.T) {
{name: "rule 5 at F", before: fNew, set: func(m *Module, n *Nodes) {
atF(m, n)
n.Haves["ace"]["store"] = delivery.Have{FP: fpOf('s'), Phase: delivery.PhaseRunning, Health: delivery.HealthWaits}
}, want: delivery.Want{FP: fNew, Composed: dig(fNew)}, rule: delivery.RuleProviders},
}, want: delivery.Want{FP: fNew, Composed: dig(fNew)}},
{name: "rule 3 at F", before: fNew, set: func(m *Module, n *Nodes) {
atF(m, n)
n.Haves["ace"]["mail"] = delivery.Have{FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
@@ -962,10 +1003,12 @@ func x11() (Module, Nodes) {
func TestANodeRunningAModuleWithNoWantGetsItsFirstWantAtWhatItRuns(t *testing.T) {
m, n := x11()
out := Wants(m, n, nil, nil)
for _, node := range []string{"g14", "shanks"} {
if w := out.Wants[node]; w.FP != fOld || w.Composed != dig(fOld) || w.Newest != fNew || w.Rule == 0 {
t.Errorf("%s's want = %+v, want its first want at the fingerprint it runs, held back from F", node, w)
}
// 1960: the providers rule holds no want: the canary g14 gets F, its provider's health unknown said
if w := out.Wants["g14"]; w.FP != fNew || out.Unreached["g14"] == "" {
t.Errorf("g14's want = %+v (unreached %q), want F, the provider said", w, out.Unreached["g14"])
}
if w := out.Wants["shanks"]; w.FP != fOld || w.Composed != dig(fOld) || w.Newest != fNew || w.Rule != delivery.RuleCanary {
t.Errorf("shanks's want = %+v, want its first want at the fingerprint it runs, held back from F by the canary", w)
}
if len(out.Neither) != 0 {
t.Errorf("left on neither list: %+v", out.Neither)
@@ -1003,11 +1046,22 @@ func TestNoFirstWantWithoutTheCompositionAndGrantOfWhatItRuns(t *testing.T) {
m, n := x11()
change(&m, &n)
out := Wants(m, n, nil, nil)
if w, ok := out.Wants["shanks"]; ok {
t.Errorf("%s: shanks got a want: %+v", what, w)
if what == "an old-form build" {
// the canary g14 does not run F healthy: no want, on neither list by rule 3 (issue 1945)
if w, ok := out.Wants["shanks"]; ok {
t.Errorf("%s: shanks got a want: %+v", what, w)
}
if h, ok := out.Neither["shanks"]; !ok || h.Rule != delivery.RuleCanary {
t.Errorf("%s: shanks on neither list = %+v (named %v), want rule 3", what, h, ok)
}
continue
}
if h, ok := out.Neither["shanks"]; !ok || h.Rule == 0 {
t.Errorf("%s: shanks on neither list = %+v (named %v)", what, h, ok)
// 1960: no want can name what it runs, so its first want names F rather than none: never on neither list
if w := out.Wants["shanks"]; w.FP != fNew {
t.Errorf("%s: shanks's want = %+v, want F", what, w)
}
if h, ok := out.Neither["shanks"]; ok {
t.Errorf("%s: shanks on neither list = %+v", what, h)
}
}
m, n := x11()
@@ -1016,10 +1070,12 @@ func TestNoFirstWantWithoutTheCompositionAndGrantOfWhatItRuns(t *testing.T) {
if len(out.Neither) != 0 {
t.Errorf("nodes holding a want named on neither list: %+v", out.Neither)
}
for node, w := range out.Wants {
if w.FP != fOld {
t.Errorf("%s's want moved from its want past the ordering rules: %+v", node, w)
}
// the canary g14 moves to F (its provider's health no rule); shanks waits for it
if w := out.Wants["shanks"]; w.FP != "" && w.FP != fOld {
t.Errorf("shanks's want moved from its want past the canary: %+v", w)
}
if w := out.Wants["g14"]; w.FP != fNew {
t.Errorf("the canary g14's want = %+v, want F", w)
}
}
@@ -1030,11 +1086,12 @@ func TestANewAssignmentWaitingIsNotOnNeitherList(t *testing.T) {
m, n := x11()
delete(n.Haves["shanks"], "screen-lock")
out := Wants(m, n, nil, nil)
if w, ok := out.Wants["shanks"]; ok {
t.Errorf("a new assignment waiting for the canary got a want: %+v", w)
// 1960: the canary orders moves, never a first want: the new assignment is wanted at F
if w := out.Wants["shanks"]; w.FP != fNew {
t.Errorf("a new assignment while the canary does not run F: %+v, want F", w)
}
if _, ok := out.Neither["shanks"]; ok {
t.Errorf("a new assignment waiting for the canary is named on neither list: %+v", out.Neither)
t.Errorf("a new assignment is named on neither list: %+v", out.Neither)
}
m.Built = false
if out := Wants(m, n, nil, nil); len(out.Neither) != 1 || out.Neither["g14"].Rule != delivery.RuleBuildDone {
@@ -1076,7 +1133,7 @@ func TestANewAssignmentWaitingIsNotOnNeitherList(t *testing.T) {
// 1945/AC21: a node whose have names an old-form build (no fingerprint of the new form, so no composition of it) can never be
// named in a want at what it runs; its first want names F, under the canary rule as usual and with everything F needs and the
// providers rule. Given the canary runs F healthy, shanks's first want names F; given it does not yet, shanks gets no want and
// stays on neither list. The old-form node that is itself the canary gets F once its providers run healthy.
// stays on neither list; the providers hold no want (issue 1960). The old-form node that is itself the canary gets F once its providers run healthy.
func TestAnOldFormNodeGetsItsFirstWantAtFOnlyUnderTheCanary(t *testing.T) {
oldForm := func(n Nodes, node string) {
n.Haves[node]["screen-lock"] = delivery.Have{FP: "build-1792260876956849831", Phase: delivery.PhaseRunning, Health: delivery.HealthUnknown}
@@ -1110,16 +1167,16 @@ func TestAnOldFormNodeGetsItsFirstWantAtFOnlyUnderTheCanary(t *testing.T) {
if h, ok := out.Neither["shanks"]; !ok || h.Rule != delivery.RuleCanary {
t.Errorf("canary not healthy on F: shanks on neither list = %+v (named %v), want rule 3", h, ok)
}
// the canary healthy on F but the providers not healthy on shanks: rule 5 still holds it back
// 1960: the canary healthy on F but the providers not healthy on shanks: F all the same, the provider down said
m, n = x11()
oldForm(n, "shanks")
n.Haves["g14"]["screen-lock"] = delivery.Have{FP: fNew, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
out = Wants(m, n, nil, nil)
if w, ok := out.Wants["shanks"]; ok {
t.Errorf("providers not healthy: shanks got a want: %+v", w)
if w := out.Wants["shanks"]; w.FP != fNew || out.Unreached["shanks"] == "" {
t.Errorf("providers not healthy: shanks's want = %+v (unreached %q), want F, the provider said", w, out.Unreached["shanks"])
}
if h := out.Neither["shanks"]; h.Rule != delivery.RuleProviders {
t.Errorf("providers not healthy: shanks on neither list = %+v, want rule 5", h)
if _, on := out.Neither["shanks"]; on {
t.Errorf("providers not healthy: shanks on neither list: %+v", out.Neither)
}
// F not granted: nothing, even with the canary healthy
m, n = x11()
@@ -94,6 +94,34 @@ func Tools(r build.StuckReader, askAgain func(key string, value []byte) error, q
return map[string]any{"key": key, "revision": r.Revision, "asks": r.Asks, "answer": r.Answer,
"next": build.StuckTool + " shows the key and its asks while they wait"}, nil
}},
{Name: build.AnnounceAgainTool,
Description: "Announce a done build again (take in again, compose again): the builder writes the done key of module and fingerprint " +
"again, unchanged, as a new revision of the same key (its content, artifacts and manifest stay identical), so the controller's " +
"take-in hears it and composes the module and writes its grant again. Use it when a done build was taken in but its composition " +
"or grant is gone and every want is refused for it. Answers the new revision. Refuses a key that is not done: " + build.BuildAgainTool +
" builds a failed key again, " + build.AskAgainTool + " asks a waiting key's asks again.",
Input: map[string]any{"type": "object", "required": []string{"module", "fp", "why"}, "properties": map[string]any{
"module": map[string]any{"type": "string", "description": "the module's name"},
"fp": map[string]any{"type": "string", "description": "its new-form fingerprint, src2:<hex>"},
"why": map[string]any{"type": "string", "description": "why it is announced again, in plain words"}}},
Run: func(args map[string]any) (any, error) {
key, value, err := build.NewAnnounceAgain(str(args, "module"), str(args, "fp"), str(args, "why"), stdio.CallerOf(args), now())
if err != nil {
return nil, err
}
raw, err := buildAgain(key, value)
if err != nil {
return nil, err
}
r, err := build.DecodeReAsk(raw)
if err != nil {
return nil, err
}
if r.Refused || r.Revision == 0 {
return nil, errors.New(r.Answer)
}
return map[string]any{"key": key, "revision": r.Revision, "answer": r.Answer}, nil
}},
{Name: build.SeatQueueTool,
Description: "What waits on a build seat's stream (SEAT_NODE_BUILD_<LANGUAGE>), as the builder reads it: the stream's message count and " +
"sequences, and per kind (delivery, check, recheck) its messages oldest first, at most limit each, with sequence, ask id, when asked " +
@@ -19,6 +19,7 @@ import (
type askAgainer interface {
AskAgain(key, why string) []string
BuildAgain(ctx context.Context, module, fp, why, by string, bound time.Duration) build.BuildAgainResult
AnnounceAgain(ctx context.Context, module, fp, why, by string) build.BuildAgainResult
}
// buildAgainBound is how long the builder waits for the claim and the asks of a key built again by hand; the tool waits
@@ -45,6 +46,10 @@ func reAsked(ctx context.Context, b askAgainer, kv jetstream.KeyValue, e jetstre
go builtAgain(ctx, b, kv, e, r, logf)
return
}
if r.Mode == build.ModeAnnounce {
announcedAgain(ctx, b, kv, e, r, logf)
return
}
asked := b.AskAgain(e.Key(), r.Why)
if len(asked) == 0 {
r.Answer = "no ask of this key waits in this builder now (it is not building, or its build has not reached its ask yet): nothing was asked; builder_stuck shows where it stands"
@@ -114,3 +119,22 @@ func builtAgain(ctx context.Context, b askAgainer, kv jetstream.KeyValue, e jets
logf("%s.%s: the answer could not be written: %v", build.ReAsksState, e.Key(), err)
}
}
// announcedAgain answers a reasks value of mode announce (the tool builder_announce_again): the done key is written again
// unchanged, and the value is written back with the new revision, or why nothing was done.
func announcedAgain(ctx context.Context, b askAgainer, kv jetstream.KeyValue, e jetstream.KeyValueEntry, r build.ReAsk, logf func(string, ...any)) {
res := b.AnnounceAgain(ctx, r.Module, r.FP, r.Why, r.By)
if res.Refused != "" {
r.Answer, r.Refused = res.Refused, true
} else {
r.Revision, r.Answer = res.Revision, fmt.Sprintf("announced again: the done key is written again, unchanged, as revision %d", res.Revision)
}
r.Answered = time.Now().UTC()
data, err := json.Marshal(r)
if err != nil {
return
}
if _, err := kv.Update(ctx, e.Key(), data, e.Revision()); err != nil && !errors.Is(err, jetstream.ErrKeyExists) && !errors.Is(err, jetstream.ErrKeyRevisionMismatch) {
logf("%s.%s: the answer could not be written: %v", build.ReAsksState, e.Key(), err)
}
}
@@ -0,0 +1,53 @@
package build
import (
"context"
"errors"
"fmt"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// **A done build is announced again by hand** (recovery of photos and link2pay on novox, 2026-10-29). The controller takes a
// build in when a write of the build key reaches its take-in consumer, and composes then; a done key is never written again
// by the builder, so a composition or a grant lost afterwards could not be made again by any tool. AnnounceAgain writes the
// done key again with its own bytes, as a new revision of the same key: its content, state, artifacts and manifest stay
// identical, and only the write is new. Who asked and why is in the builder's log and in the reasks answer.
// AnnounceAgain rewrites the done key of module at fp unchanged, as a new revision, and answers the revision. A key that is
// not done is refused with why: failed (builder_build_again builds it) or waiting (builder_ask_again).
func (b *Builder) AnnounceAgain(ctx context.Context, module, fp, why, by string) BuildAgainResult {
b.init()
key, err := delivery.BuilderBuildKey(module, fp)
if err != nil {
return BuildAgainResult{Refused: err.Error()}
}
value, rev, found, err := b.Claimer.Store.Get(ctx, key)
if err != nil {
return BuildAgainResult{Refused: fmt.Sprintf("%s cannot be read: %v", key, err)}
}
if !found {
return BuildAgainResult{Refused: fmt.Sprintf("%s has no build key: nothing to announce", key)}
}
cur, err := delivery.DecodeBuilderBuild(value)
if err != nil {
return BuildAgainResult{Refused: fmt.Sprintf("%s holds what is not a build: %v", key, err)}
}
switch cur.State {
case delivery.BuildDone:
case delivery.BuildFailed:
return BuildAgainResult{Refused: fmt.Sprintf("%s is failed, not done: %s builds it again", key, BuildAgainTool)}
default:
return BuildAgainResult{Refused: fmt.Sprintf("%s is %q, not done: %s asks its waiting asks again", key, cur.State, AskAgainTool)}
}
// Compare-and-set at the revision read: a build written meanwhile is not overwritten.
nrev, err := b.Claimer.Store.Update(ctx, key, value, rev)
if err != nil {
if errors.Is(err, ErrConflict) {
return BuildAgainResult{Refused: fmt.Sprintf("%s changed while it was announced: read it again", key)}
}
return BuildAgainResult{Refused: fmt.Sprintf("%s could not be written: %v", key, err)}
}
b.logf("%s: done, announced again by %s (revision %d): %s", key, by, nrev, why)
return BuildAgainResult{Revision: nrev}
}
@@ -0,0 +1,49 @@
package build
import (
"bytes"
"context"
"strings"
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
func TestAnnounceAgainRewritesOnlyADoneKeyUnchanged(t *testing.T) { // recovery of photos and link2pay, 2026-10-29
for _, c := range []struct {
name string
key *delivery.BuilderBuild
fp string
refuse string
}{
{name: "a done key is written again, unchanged, as a new revision", key: &delivery.BuilderBuild{State: delivery.BuildDone, Step: "built"}},
{name: "a failed key is refused, builder_build_again is for it", refuse: "builder_build_again", key: &delivery.BuilderBuild{State: delivery.BuildFailed, Error: "x"}},
{name: "a building key is refused, builder_ask_again is for it", refuse: "builder_ask_again", key: &delivery.BuilderBuild{State: delivery.BuildBuilding, By: "novox.builder", Step: "asked"}},
{name: "a key with no build is refused", refuse: "has no build key"},
{name: "a fingerprint that is not one is refused", refuse: "fingerprint", fp: "nope"},
} {
t.Run(c.name, func(t *testing.T) {
r := newRig(t, goBundle)
var rev uint64
if c.key != nil {
c.key.Manifest = manifestD
rev = putBuild(t, r.st, r.key, *c.key)
}
before := bytes.Clone(r.st.vals[r.key])
fp := fp1
if c.fp != "" {
fp = c.fp
}
res := r.b.AnnounceAgain(context.Background(), "m", fp, "grant lost", "shanks.lead")
if c.refuse != "" {
if !strings.Contains(res.Refused, c.refuse) || res.Revision != 0 || r.st.revs[r.key] != rev || !bytes.Equal(r.st.vals[r.key], before) {
t.Fatalf("answer %+v, key at revision %d (was %d); want refused with %q and nothing written", res, r.st.revs[r.key], rev, c.refuse)
}
return
}
if res.Refused != "" || res.Revision <= rev || r.st.revs[r.key] != res.Revision || !bytes.Equal(r.st.vals[r.key], before) {
t.Fatalf("answer %+v, key at revision %d (was %d); want a new revision and the same bytes", res, r.st.revs[r.key], rev)
}
})
}
}
@@ -24,6 +24,9 @@ const (
AskAgainTool = "builder_ask_again"
// BuildAgainTool builds a failed key again by hand (issue 1126 note 72): it writes a reasks value of Mode ModeBuild.
BuildAgainTool = "builder_build_again"
// AnnounceAgainTool writes a done key again unchanged, so the controller takes it in and composes again: a reasks value
// of Mode ModeAnnounce.
AnnounceAgainTool = "builder_announce_again"
// ReAsksState is the builder's state of asks to ask again: the bucket mesh-delivery-builder_reasks, keyed as the build key.
ReAsksState = "reasks"
ReAsksBucket = delivery.BuilderModule + "_" + ReAsksState
@@ -56,6 +59,9 @@ type ReAsk struct {
// ModeBuild is the Mode of a reasks value that asks the key to be built again, not its waiting asks to be asked again.
const ModeBuild = "build"
// ModeAnnounce is the Mode of a reasks value that asks a done key to be announced again.
const ModeAnnounce = "announce"
// DecodeReAsk reads a reasks value.
func DecodeReAsk(data []byte) (ReAsk, error) {
var r ReAsk
@@ -94,6 +100,19 @@ func NewBuildAgain(module, fp, why, by string, at time.Time) (string, []byte, er
return key, data, err
}
// NewAnnounceAgain is the value builder_announce_again writes for a module's fingerprint: its key and the value.
func NewAnnounceAgain(module, fp, why, by string, at time.Time) (string, []byte, error) {
key, err := delivery.BuilderBuildKey(module, fp)
if err != nil {
return "", nil, err
}
if strings.TrimSpace(why) == "" {
return "", nil, errors.New("say why the done build is announced again (why)")
}
data, err := json.Marshal(ReAsk{V: 1, Why: strings.TrimSpace(why), By: by, At: at.UTC(), Mode: ModeAnnounce, Module: module, FP: fp})
return key, data, err
}
// StateRead is a state the tools read: its keys, and a key's value with when it was written (zero when not known).
type StateRead struct {
Keys func() ([]string, error)
@@ -116,7 +116,7 @@ func TestManifestDeclarations(t *testing.T) { // 1607/AC1
t.Errorf("secrets-owner = %q", m.SecretsOwner)
}
// 1945: the builder's tools (builder_seat_queue: note 62), served by its tools bundle, and the event it says when it asks an ask again.
if want := []string{"builder_ask_again", "builder_build_again", "builder_seat_queue", "builder_stuck"}; !reflect.DeepEqual(sorted(m.Tools), want) {
if want := []string{"builder_announce_again", "builder_ask_again", "builder_build_again", "builder_seat_queue", "builder_stuck"}; !reflect.DeepEqual(sorted(m.Tools), want) {
t.Errorf("tools = %v, want %v", m.Tools, want)
}
if want := []string{"build.asked-again"}; !reflect.DeepEqual(m.Emits, want) {
@@ -3,6 +3,7 @@
"version": "1",
"slug": "builder",
"tools": [
"builder_announce_again",
"builder_ask_again",
"builder_build_again",
"builder_seat_queue",