amadan
sim: hostlife: an abandoned cold admission is parked by the next op.exec; a writer inside an operation certifies its close; diagnostics
Found by the 100,000-seed baseline of rev 7.9's first model: a lost target's unfinished cold admission inside the landing blocked the source's rollback probe for good (seed 58795), and every clean close outside serving was dropped, so a rollback's reopened source was admitted cold at every wake and the rollback never released (seed 84087). The close is dropped only for a push, a move's cut, a delete or a purge (the first correction let a push's writer certify into the superseded generation, seed 4493 / 30,000). Logs for a refused admission, a failed rollback probe, a skipped obligation run and each park-local step.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
4 files changed,
+56
−5
internal/sim/hostlife/ltx_test.go+17 −4internal/sim/hostlife/moves_test.go+27 −0internal/sim/hostlife/protocol_test.go+10 −1internal/sim/hostlife/world_test.go+2 −0
diff --git a/internal/sim/hostlife/ltx_test.go b/internal/sim/hostlife/ltx_test.go| index 68065a9a4..4bcd5f384 100644 |
| --- a/internal/sim/hostlife/ltx_test.go |
| +++ b/internal/sim/hostlife/ltx_test.go |
| @@ -346,10 +346,23 @@ func (w *world) closeStep(b *box) { |
| if db == nil || !db.Closing || hasMain(b) { |
| return |
| } |
| - // the marker is the live generation's: once an operation has taken the |
| - // writer over (a push, a move, a delete), or the claim has moved on, |
| - // the generation is dead or being replaced and the close is dropped |
| - if c := w.claim; c == nil || c.Holder == nil || c.Op.Kind != "serving" || c.Op.Exec != b.ID || c.Holder.Gen != db.Gen || c.Holder.Inc != db.Inc { |
| + // the certificate is the live generation's: once an operation has |
| + // taken the writer over (a push, a move, a delete), or the claim has |
| + // moved on, the generation is dead or being replaced and the close is |
| + // dropped. A writer running inside an operation (a landing's probe or |
| + // its committed target, a rollback's reopened source, a push's probe) |
| + // certifies like any other (claim rev 6.8 §6.4a; seed 84087 / 100,000: |
| + // rev 7.9's first build dropped every close outside serving, so a |
| + // rollback's source was admitted cold at every wake and the rollback |
| + // never released) |
| + c := w.claim |
| + inside := c != nil && (c.Op.Kind == "serving" || c.Op.Kind == "landing" || c.Op.Kind == "restoring" || c.Op.Kind == "reverting") |
| + if c == nil || c.Holder == nil || !inside || c.Op.Exec != b.ID || c.Holder.Inc != db.Inc || (db.Gen != c.Holder.Gen && db.Gen != w.opGen(&c.Op)) { |
| + // a push, a move's cut, a delete or a purge is taking this writer |
| + // over: its generation is being replaced or parked, and a |
| + // certificate written into it now would outlive the park (seed |
| + // 4493 / 30,000: a certificate landed in a generation a push had |
| + // superseded) |
| db.Closing, db.CloseReq = false, nil |
| return |
| } |
diff --git a/internal/sim/hostlife/moves_test.go b/internal/sim/hostlife/moves_test.go| index 34afc0581..e79640d55 100644 |
| --- a/internal/sim/hostlife/moves_test.go |
| +++ b/internal/sim/hostlife/moves_test.go |
| @@ -21,10 +21,16 @@ func (w *world) workerMode(b *box, obl bool) { |
| } |
| c, ok := w.readClaim(b) |
| if !ok { |
| + if obl && w.claim != nil && len(w.claim.Obl) > 0 && w.calm { |
| + w.logf("box %d: obligation run skipped, the agent cannot read the claim (down=%v paused=%v frozen=%v partitioned=%v destroyed=%v)", b.ID, b.AgentDown, b.AgentPaused, b.Frozen, b.Partitioned, b.Destroyed) |
| + } |
| return |
| } |
| if obl { |
| if c == nil || (contains(c.Lost, b.ID) && on("lost-quarantine")) { |
| + if c != nil && w.calm { |
| + w.logf("box %d: obligation run refused, the box is declared lost", b.ID) |
| + } |
| return |
| } |
| for _, o := range c.Obl { |
| @@ -166,6 +172,26 @@ func (w *world) role(b *box, c *claim) { |
| if c.Holder == nil { |
| return |
| } |
| + if r := c.Op.Rec; r != nil && r.Box != c.Op.Exec && (c.Op.Exec == b.ID || c.Op.Exec == -1 && w.isParticipant(c, b)) { |
| + // the admitting box lost op.exec before its build finished (a |
| + // rollback's release of the target, a declared loss): its cold |
| + // admission is abandoned by whoever holds op.exec next, and the |
| + // generation it was building is parked as a chore (seed 58795 / |
| + // 100,000: the source's rollback probe could not be admitted behind |
| + // the lost target's build) |
| + if w.cas(c, func(n *claim) bool { |
| + if n.Op.Rec == nil || n.Op.Rec.Gen != r.Gen || n.Op.Rec.Box == n.Op.Exec { |
| + return false |
| + } |
| + n.Obl = append(n.Obl, w.newObl(n, "park-generation", r.Region, -1, c.Holder.Location, r.Gen, w.now+horizonH)) |
| + n.Op.Rec = nil |
| + return true |
| + }) { |
| + w.logf("box %d: cold admission %s of box %d abandoned, its exec is gone (op %s/%s exec %d)", b.ID, r.Gen, r.Box, c.Op.Kind, c.Op.ID, c.Op.Exec) |
| + w.stats["cold admissions abandoned"]++ |
| + } |
| + return |
| + } |
| if c.Op.Rec != nil && c.Op.Rec.Box == b.ID { |
| // a cold admission's build runs under the operation, by the box it |
| // admitted, before the operation's own step (claim rev 6.8 §6.4a, R8-2) |
| @@ -1326,6 +1352,7 @@ func (w *world) recoverSource(b *box, c *claim) { |
| w.cas(c, func(n *claim) bool { n.Op.SrcPhase = "probed"; return sameOp(n.Op.ID, op.ID) }) |
| return |
| } |
| + w.logf("box %d: the rollback's probe did not start (try %d, generation %q, db=%v)", b.ID, op.ProbeTries+1, want, b.DB != nil) |
| w.cas(c, func(n *claim) bool { |
| n.Op.ProbeTries++ |
| if n.Op.ProbeTries >= probeTries*3 { |
diff --git a/internal/sim/hostlife/protocol_test.go b/internal/sim/hostlife/protocol_test.go| index 7e6b8b1ed..49f0a1294 100644 |
| --- a/internal/sim/hostlife/protocol_test.go |
| +++ b/internal/sim/hostlife/protocol_test.go |
| @@ -386,6 +386,7 @@ func (w *world) spawn(b *box, c *claim, want string, probe bool) bool { |
| w.closeStep(b) |
| if b.DB.Closing { |
| w.stats["wake waited for a clean close"]++ |
| + w.logf("box %d: start waits for the previous session's clean close", b.ID) |
| return false |
| } |
| } |
| @@ -597,6 +598,7 @@ func (w *world) admitWriter(b *box, c *claim, gen string, probe bool) string { |
| h := c.Holder |
| if on("exec-writer") && c.Op.Exec != b.ID { |
| w.stats["admission refused: not op.exec"]++ |
| + w.logf("box %d: admission refused, op.exec is %d (%s/%s)", b.ID, c.Op.Exec, c.Op.Kind, c.Op.ID) |
| return "" |
| } |
| if !on("admit-everywhere") && c.Op.Kind != "serving" { |
| @@ -626,7 +628,7 @@ func (w *world) admitWriter(b *box, c *claim, gen string, probe bool) string { |
| n.Holder.Session = w.nextGen() |
| p := pred{Session: h.Session, Box: w.predBox(c), Gen: gen, Stamp: h.Stamp} |
| n.Op.Writers = append(n.Op.Writers, writerEntry{Session: n.Holder.Session, Box: b.ID, Gen: id, Mode: "cold", Pred: p}) |
| - n.Op.Rec = &recState{Gen: id, Base: gen, Session: n.Holder.Session, Box: b.ID, Phase: "build", Pred: p} |
| + n.Op.Rec = &recState{Gen: id, Base: gen, Session: n.Holder.Session, Box: b.ID, Region: b.Region, Phase: "build", Pred: p} |
| w.newGen(n.Holder.Location, gen, id) |
| if gen == n.Holder.Gen { |
| n.Holder.Gen = id // the claim's generation moves with the writer it admits |
| @@ -672,6 +674,8 @@ func (w *world) admitCommit(b *box, c *claim, gen, mode string) bool { |
| }) |
| if ok { |
| w.stats["admissions: "+mode]++ |
| + } else { |
| + w.logf("box %d: %s admission CAS refused (session %d, op %s/%s exec %d)", b.ID, mode, h.Session, c.Op.Kind, c.Op.ID, c.Op.Exec) |
| } |
| return ok |
| } |
| @@ -1010,6 +1014,7 @@ func (w *world) doObligation(b *box, c *claim, o obligation) { |
| switch o.Kind { |
| case "park-local": |
| if !w.evidence(b, o.Op) { |
| + w.logf("box %d: park-local %s waits for stop evidence", b.ID, o.ID) |
| return |
| } |
| inc := 0 |
| @@ -1020,8 +1025,12 @@ func (w *world) doObligation(b *box, c *claim, o obligation) { |
| inc = b.DB.Inc |
| } |
| if !w.localPark(b, inc) { |
| + if w.calm { |
| + w.logf("box %d: park-local %s parked an artifact of inc %d (stamp %d, db=%v files=%v cfg=%d live=%q)", b.ID, o.ID, inc, b.stampOf(inc), b.DB != nil, b.Files != nil, b.Config, b.Live) |
| + } |
| return // the next artifact on a later step |
| } |
| + w.logf("box %d: park-local %s done", b.ID, o.ID) |
| if b.Row != nil && b.Row.HoldOp == o.Op || b.Row != nil && (c.Holder == nil || b.ID != c.Op.Exec) { |
| b.Row = nil |
| } |
diff --git a/internal/sim/hostlife/world_test.go b/internal/sim/hostlife/world_test.go| index 5853375d5..d321f3ad9 100644 |
| --- a/internal/sim/hostlife/world_test.go |
| +++ b/internal/sim/hostlife/world_test.go |
| @@ -614,6 +614,8 @@ type pred struct { |
| // session and box admitted, the phase, the predecessor, and the snapshot |
| // identity once written. |
| type recState struct { |
| + Region int // the admitting box's region, where the generation under construction lives |
| + |
| Gen, Base string |
| Session, Box int |
| Phase string |