diff --git a/changelog.d/fixed-16023.md b/changelog.d/fixed-16023.md new file mode 100644 index 0000000000..fb35f82742 --- /dev/null +++ b/changelog.d/fixed-16023.md @@ -0,0 +1 @@ +- An adopted (or otherwise hive-owned) conflicting PR no longer sits on "Rebasing · attempt 1/2" while the fixer is busy: each rebase attempt is routed once and gets `governor.labels.stage.adoption.rebase_attempt_timeout` (default 15m) to push before it counts as failed and the next attempt is dispatched, the row shows the wait ("Rebasing · attempt 1/2 · waiting for scanner 12m"), and an in-place adoption whose attempts are spent re-homes to a hive-owned superseding PR at once instead of after the 3h dwell ([#16023](https://github.com/hivecommons/hive/issues/16023)). diff --git a/src/cmd/hive/adopted_pr_fixer_test.go b/src/cmd/hive/adopted_pr_fixer_test.go index 6f311de2f7..4fdb6ca2a1 100644 --- a/src/cmd/hive/adopted_pr_fixer_test.go +++ b/src/cmd/hive/adopted_pr_fixer_test.go @@ -1,6 +1,9 @@ package main import ( + "bytes" + "log/slog" + "path/filepath" "strings" "testing" "time" @@ -229,3 +232,82 @@ func TestWriteMergeEligible_UntrustedConflictIsNotHiveRebaseWork(t *testing.T) { t.Fatalf("ci_failing = %+v, want nothing when the trusted-author tier is off", failing) } } + +// sweepRebaseLogged runs one writeMergeEligible sweep at now and returns the +// ci-failing rows and the number of "routed to fixer for rebase" lines. +func sweepRebaseLogged(t *testing.T, prs []github.PullRequest, now time.Time) ([]failingEntry, int) { + t.Helper() + dir := t.TempDir() + origMerge, origFail, origNow := mergeEligiblePath, ciFailingPath, rebaseAttemptNow + mergeEligiblePath = filepath.Join(dir, "merge-eligible.json") + ciFailingPath = filepath.Join(dir, "ci-failing.json") + rebaseAttemptNow = func() time.Time { return now } + defer func() { mergeEligiblePath, ciFailingPath, rebaseAttemptNow = origMerge, origFail, origNow }() + var logs bytes.Buffer + actionable := &github.ActionableResult{PRs: github.PRResult{Items: prs}} + writeMergeEligible(actionable, github.HoldResult{}, "hivecommons", nil, false, nil, false, nil, nil, nil, slog.New(slog.NewTextHandler(&logs, nil))) + var payload struct { + CIFailing []failingEntry `json:"ci_failing"` + } + readJSON(t, ciFailingPath, &payload) + return payload.CIFailing, strings.Count(logs.String(), "PR routed to fixer for rebase") +} + +// hivecommons/hive#16023: an adopted conflicting PR whose fixer never +// pushes must not sit on attempt 1 forever. The attempt is routed (and +// logged) once; past the per-attempt deadline it counts as failed and +// attempt 2 is routed with one line; past the cap the row leaves the fixer +// list so the adoption lane can re-home it. +func TestWriteMergeEligible_AdoptedRebaseAttemptTimesOutWithoutPush(t *testing.T) { + store := useMemoryRebaseAttempts(t) + const repo = "hivecommons/hive" + adoptInPlace(t, repo, 15998, "f00dfac") + pr := github.PullRequest{Number: 15998, Repo: "hive", Author: "clanker-bot", HeadSHA: "f00dfac", + HeadRef: "feat/x", HeadRepo: "hivecommons/hive", BaseRef: "v5", + Mergeable: github.MergeableNo, MergeableState: "dirty", CIStatus: "pending", + Labels: []string{"needs-rebase", "stage/074-adopted-by-hive", "author/clanker"}} + t0 := time.Date(2026, 10, 31, 9, 0, 0, 0, time.UTC) + timeout := config.DefaultAdoptRebaseAttemptTimeout + + failing, routed := sweepRebaseLogged(t, []github.PullRequest{pr}, t0) + if len(failing) != 1 || failing[0].RebaseAttempt != 1 || routed != 1 { + t.Fatalf("first sweep: rows=%+v routed=%d, want attempt 1 routed once", failing, routed) + } + // Later sweeps inside the deadline keep the row listed, not re-routed. + failing, routed = sweepRebaseLogged(t, []github.PullRequest{pr}, t0.Add(5*time.Minute)) + if len(failing) != 1 || failing[0].RebaseAttempt != 1 || routed != 0 { + t.Fatalf("second sweep: rows=%+v routed=%d, want attempt 1 listed but not re-routed", failing, routed) + } + if a, _ := store.Get(repo, 15998); !a.DispatchedAt.Equal(t0) || a.Fixer != "scanner" { + t.Fatalf("dispatch = %+v, want the first sweep's dispatch to scanner kept", a) + } + if got := rebasemerge.AttemptTextAt(mustAttempt(t, store, repo, 15998), t0.Add(12*time.Minute)); got != "Rebasing · attempt 1/2 · waiting for scanner 12m" { + t.Fatalf("caption = %q", got) + } + // No push by the deadline: attempt 1 failed, attempt 2 routed once. + failing, routed = sweepRebaseLogged(t, []github.PullRequest{pr}, t0.Add(timeout)) + if len(failing) != 1 || failing[0].RebaseAttempt != 2 || routed != 1 { + t.Fatalf("deadline sweep: rows=%+v routed=%d, want attempt 2 routed once", failing, routed) + } + failing, routed = sweepRebaseLogged(t, []github.PullRequest{pr}, t0.Add(timeout+time.Minute)) + if len(failing) != 1 || failing[0].RebaseAttempt != 2 || routed != 0 { + t.Fatalf("attempt 2 re-sweep: rows=%+v routed=%d", failing, routed) + } + // Attempt 2 also times out: exhausted, off the fixer list. + failing, routed = sweepRebaseLogged(t, []github.PullRequest{pr}, t0.Add(2*timeout)) + if len(failing) != 0 || routed != 0 { + t.Fatalf("exhausted sweep: rows=%+v routed=%d, want nothing", failing, routed) + } + if a := mustAttempt(t, store, repo, 15998); !a.Exhausted() || a.Stalled != 2 { + t.Fatalf("episode = %+v, want exhausted by two timed-out attempts", a) + } +} + +func mustAttempt(t *testing.T, store *rebasemerge.Attempts, repo string, number int) rebasemerge.Attempt { + t.Helper() + a, ok := store.Get(repo, number) + if !ok { + t.Fatalf("no rebase episode for %s#%d", repo, number) + } + return a +} diff --git a/src/cmd/hive/merge_eligibility.go b/src/cmd/hive/merge_eligibility.go index 8c9a317198..992781e6d6 100644 --- a/src/cmd/hive/merge_eligibility.go +++ b/src/cmd/hive/merge_eligibility.go @@ -830,6 +830,19 @@ func hiveAuthoredPR(pr github.PullRequest) bool { // rebaseAttempts is the conflict loop-guard ledger; tests swap it. var rebaseAttempts = rebasemerge.SharedAttempts +// rebaseAttemptNow is the clock for the rebase episode; tests swap it. +var rebaseAttemptNow = time.Now + +// rebaseAttemptTimeout is how long one fixer rebase attempt may wait for a +// push before it counts as failed +// (governor.labels.stage.adoption.rebase_attempt_timeout, default 15m). +func rebaseAttemptTimeout(cfg *config.Config) time.Duration { + if cfg == nil { + return config.DefaultAdoptRebaseAttemptTimeout + } + return cfg.Governor.Labels.Stage.Adoption.EffectiveRebaseAttemptTimeout() +} + // mergeTrainRebaseRequester is the RequestedBy the merge train writes on a // rebase episode it opens for a member it could not update-branch // (hivecommons/hive#15117 → #15104). @@ -1236,13 +1249,30 @@ func writeMergeEligible(actionable *github.ActionableResult, hold github.HoldRes // Loop guard: attempt n of MaxRebaseAttempts per conflict // episode; a fixer's pushed head that is still dirty is the // next attempt, and past the cap the hive stops re-kicking. - attempt := rebaseAttempts().Observe(fullRepo, pr.Number, pr.HeadSHA, time.Now()) + // A fixer-owned attempt (adopted, trusted-author, requested) + // that saw no push within its deadline is the next attempt too + // (hivecommons/hive#16023): a busy fixer must not hold the + // episode at attempt 1 forever. + fixerRebase := adopted || trustedRebase || requested + rebaseNow := rebaseAttemptNow() + attempt := rebaseAttempts().Observe(fullRepo, pr.Number, pr.HeadSHA, rebaseNow) + var timedOut *rebasemerge.Attempt + if fixerRebase { + prev := attempt + if next, ok := rebaseAttempts().Expire(fullRepo, pr.Number, rebaseAttemptTimeout(cfg), rebaseNow); ok { + timedOut, attempt = &prev, next + } + } row.RebaseAttempt = attempt.Count() row.RebaseAttemptsMax = rebasemerge.MaxRebaseAttempts row.BaseRef = pr.BaseRef row.TrustedAuthor = trustedRebase if attempt.Exhausted() { - logger.Warn("rebase attempts exhausted; leaving the conflict to a person", "repo", fullRepo, "number", pr.Number, "head", pr.HeadSHA, "attempts", attempt.Count()-1, "max", rebasemerge.MaxRebaseAttempts, "mergeable_state", pr.MergeableState) + args := []any{"repo", fullRepo, "number", pr.Number, "head", pr.HeadSHA, "attempts", attempt.Count() - 1, "max", rebasemerge.MaxRebaseAttempts, "timed_out", attempt.Stalled, "mergeable_state", pr.MergeableState} + if timedOut != nil { + args = append(args, "waited_for", timedOut.Fixer, "waited", timedOut.Waiting(rebaseNow).Round(time.Second).String()) + } + logger.Warn("rebase attempts exhausted; leaving the conflict to a person", args...) continue } agent := prFixAgent(pr, prAgents[fmt.Sprintf("%s#%d", fullRepo, pr.Number)]) @@ -1262,8 +1292,17 @@ func writeMergeEligible(actionable *github.ActionableResult, hold github.HoldRes if trustedRebase || requested { row.ReachableAction = github.ReachableActionPush } - if adopted || trustedRebase || requested { - logger.Info("PR routed to fixer for rebase", "repo", fullRepo, "number", pr.Number, "work", "rebase", "adopted", adopted, "trusted_author", trustedRebase, "requested_by", attempt.RequestedBy, "attempt", attempt.Count(), "max", rebasemerge.MaxRebaseAttempts, "mergeable_state", pr.MergeableState, "fixer", agent, "sentinel_alert", row.SentinelAlert) + if fixerRebase { + // The row stays listed every sweep (the fixer's kick overlay + // reads it), but each attempt is routed and announced once; + // the dispatch time is the attempt's deadline clock. + if _, first := rebaseAttempts().Dispatch(fullRepo, pr.Number, agent, rebaseNow); first { + args := []any{"repo", fullRepo, "number", pr.Number, "work", "rebase", "adopted", adopted, "trusted_author", trustedRebase, "requested_by", attempt.RequestedBy, "attempt", attempt.Count(), "max", rebasemerge.MaxRebaseAttempts, "mergeable_state", pr.MergeableState, "fixer", agent, "sentinel_alert", row.SentinelAlert} + if timedOut != nil { + args = append(args, "previous_attempt_timed_out", true, "waited_for", timedOut.Fixer, "waited", timedOut.Waiting(rebaseNow).Round(time.Second).String()) + } + logger.Info("PR routed to fixer for rebase", args...) + } } failing = append(failing, row) continue diff --git a/src/docs/labels-and-control-signals/stage-labels.md b/src/docs/labels-and-control-signals/stage-labels.md index 5250872bc0..903ec7cd75 100644 --- a/src/docs/labels-and-control-signals/stage-labels.md +++ b/src/docs/labels-and-control-signals/stage-labels.md @@ -876,7 +876,13 @@ When a contributor-agent PR goes red the hive does not just relabel it (a human- adopted branch first reaches the normal fixer, which rebases onto the PR base and pushes to the actual fork with a head-bound lease, preserving authorship and sign-offs. If the head stays idle past the - 074 dwell, supersede in that sweep rather than escalate to 085. + 074 dwell, supersede in that sweep rather than escalate to 085. Each + fixer attempt is routed once and gets + `governor.labels.stage.adoption.rebase_attempt_timeout` (default 15m) + to push; one with no push counts as failed and the next attempt is + dispatched (the row reads "Rebasing · attempt 1/2 · waiting for + scanner 12m" meanwhile). Once both attempts are spent the adoption + supersedes at once instead of waiting out the dwell (#16023). - `reason=in-place-stalled` (always `mode=supersede`) — second takeover: an in-place-adopted head sat unchanged for the 074 dwell (3h) with CI idle (not pending) and the review swarm's latest verdict on it diff --git a/src/hive.yaml.example b/src/hive.yaml.example index f6b785d47f..57158a8f59 100644 --- a/src/hive.yaml.example +++ b/src/hive.yaml.example @@ -525,6 +525,8 @@ governor: # after_offline: 15m # default 15m; registered relay offline, measured from the later # # of its last heartbeat and the first stage/072 nudge # after_unregistered: 15m # default 15m; no contributor profile, wait after the nudge with no new head + # rebase_attempt_timeout: 15m # default 15m; a fixer rebase attempt with no push by then counts as failed; + # # once attempts are spent an in-place adoption re-homes to a superseding PR # supersede: # close_first: true # default true; close the original (hive/superseded, one note naming the # # copy) in the pass that opens the copy; reopened if the copy fails. diff --git a/src/pkg/config/config.go b/src/pkg/config/config.go index c2089268c7..7256747ed7 100644 --- a/src/pkg/config/config.go +++ b/src/pkg/config/config.go @@ -2930,6 +2930,13 @@ type StageAdoptionConfig struct { // 072/085 for a contributor who cannot act (default 30m; // hivecommons/hive#15307). OfflineGrace string `yaml:"offline_grace,omitempty"` + // RebaseAttemptTimeout is how long one fixer rebase attempt on an + // adopted (or otherwise hive-owned) conflicting PR may wait for the + // fixer's push before it counts as failed and the next attempt is + // dispatched; past the attempt cap an in-place adoption re-homes to a + // hive-owned superseding PR at once (default 15m; + // hivecommons/hive#16023). + RebaseAttemptTimeout string `yaml:"rebase_attempt_timeout,omitempty"` // InPlace lets the hive adopt by pushing to the PR's own branch when // GitHub allows it (same-repo head or maintainer_can_modify) instead of // opening a superseding App PR. nil = true. @@ -3057,11 +3064,17 @@ func (c StageAdoptionConfig) StrictAttributionEnabled() bool { // Defaults for the offline-adoption waits (stage 074, hivecommons/hive#12436). const ( - DefaultAdoptAfterOffline = 15 * time.Minute - DefaultAdoptAfterUnregistered = 15 * time.Minute - DefaultAdoptOfflineGrace = 30 * time.Minute + DefaultAdoptAfterOffline = 15 * time.Minute + DefaultAdoptAfterUnregistered = 15 * time.Minute + DefaultAdoptOfflineGrace = 30 * time.Minute + DefaultAdoptRebaseAttemptTimeout = 15 * time.Minute ) +// EffectiveRebaseAttemptTimeout parses RebaseAttemptTimeout (default 15m). +func (c StageAdoptionConfig) EffectiveRebaseAttemptTimeout() time.Duration { + return parseDurationOr(c.RebaseAttemptTimeout, DefaultAdoptRebaseAttemptTimeout) +} + // EffectiveOfflineGrace parses OfflineGrace (default 30m). func (c StageAdoptionConfig) EffectiveOfflineGrace() time.Duration { return parseDurationOr(c.OfflineGrace, DefaultAdoptOfflineGrace) diff --git a/src/pkg/config/stage_labels_getters_test.go b/src/pkg/config/stage_labels_getters_test.go index d53f793689..f2f74500b0 100644 --- a/src/pkg/config/stage_labels_getters_test.go +++ b/src/pkg/config/stage_labels_getters_test.go @@ -13,6 +13,7 @@ func TestStageLabelsConfigGetterDefaults(t *testing.T) { } if s.Adoption.EffectiveAfterOffline() != DefaultAdoptAfterOffline || s.Adoption.EffectiveAfterUnregistered() != DefaultAdoptAfterUnregistered || + s.Adoption.EffectiveRebaseAttemptTimeout() != 15*time.Minute || s.EffectiveContributorGraceOnline() != DefaultContributorGraceOnline || s.EffectiveContributorGraceOffline() != DefaultContributorGraceOffline || s.EffectivePresenceWindow() != DefaultPresenceWindow || @@ -33,7 +34,7 @@ func TestStageLabelsConfigGetterDefaults(t *testing.T) { ContributorNudge: ContributorNudgeConfig{Enabled: &off, RenudgeInterval: "30m"}, Adoption: StageAdoptionConfig{ Enabled: &off, InPlace: &off, StrictAttribution: &off, AutoMerge: &off, - AfterOffline: "2m", AfterUnregistered: "0s", + AfterOffline: "2m", AfterUnregistered: "0s", RebaseAttemptTimeout: "20m", }, Stalled: StageStalledConfig{RequireQuiet: &off}, } @@ -42,6 +43,7 @@ func TestStageLabelsConfigGetterDefaults(t *testing.T) { t.Fatal("explicit false turns each toggle off") } if s.EffectivePresenceWindow() != time.Minute || s.Adoption.EffectiveAfterOffline() != 2*time.Minute || + s.Adoption.EffectiveRebaseAttemptTimeout() != 20*time.Minute || s.ContributorNudge.EffectiveRenudgeInterval() != 30*time.Minute { t.Fatal("valid durations parse") } diff --git a/src/pkg/config/testdata/agent_runtime_golden.json b/src/pkg/config/testdata/agent_runtime_golden.json index 5ee4c4fb08..6c829f2337 100644 --- a/src/pkg/config/testdata/agent_runtime_golden.json +++ b/src/pkg/config/testdata/agent_runtime_golden.json @@ -203,6 +203,7 @@ "AfterOffline": "", "AfterUnregistered": "", "OfflineGrace": "", + "RebaseAttemptTimeout": "", "InPlace": null, "StrictAttribution": null, "AutoMerge": null, diff --git a/src/pkg/dashboard/rebase_merge_followthrough.go b/src/pkg/dashboard/rebase_merge_followthrough.go index cccda1a44b..f19ed4bc41 100644 --- a/src/pkg/dashboard/rebase_merge_followthrough.go +++ b/src/pkg/dashboard/rebase_merge_followthrough.go @@ -381,8 +381,9 @@ type StageTrackerRebase struct { } // applyRebaseAttempts stamps rebase onto PR rows with an open conflict -// episode in the fixer's attempts ledger. -func applyRebaseAttempts(items []StageTrackerItem) { +// episode in the fixer's attempts ledger; a dispatched attempt names the +// fixer and how long it has waited at now (hivecommons/hive#16023). +func applyRebaseAttempts(items []StageTrackerItem, now time.Time) { attempts := rebaseAttemptsStore() if attempts == nil { return @@ -396,7 +397,7 @@ func applyRebaseAttempts(items []StageTrackerItem) { if !ok { continue } - it.Rebase = &StageTrackerRebase{Attempt: min(max(a.Count(), 1), rebasemerge.MaxRebaseAttempts), Max: rebasemerge.MaxRebaseAttempts, Exhausted: a.Exhausted(), Head: rebaseMergeShortSHA(a.Head()), Text: rebasemerge.AttemptText(a)} + it.Rebase = &StageTrackerRebase{Attempt: min(max(a.Count(), 1), rebasemerge.MaxRebaseAttempts), Max: rebasemerge.MaxRebaseAttempts, Exhausted: a.Exhausted(), Head: rebaseMergeShortSHA(a.Head()), Text: rebasemerge.AttemptTextAt(a, now)} } } diff --git a/src/pkg/dashboard/rebase_merge_followthrough_test.go b/src/pkg/dashboard/rebase_merge_followthrough_test.go index 3ddf4bde54..8bab66b4e4 100644 --- a/src/pkg/dashboard/rebase_merge_followthrough_test.go +++ b/src/pkg/dashboard/rebase_merge_followthrough_test.go @@ -215,18 +215,24 @@ func TestRebaseAttemptsPillAndFollowThroughExhaustion(t *testing.T) { t.Cleanup(func() { rebaseMergeNow = time.Now }) items := []StageTrackerItem{{Repo: "hivecommons/hive", Number: 15087, Kind: "pr", Stage: 90, Modifiers: []string{modNeedsRebase}}, {Repo: "hivecommons/hive", Number: 1, Kind: "pr", Stage: 90}} - applyRebaseAttempts(items) + applyRebaseAttempts(items, now) if items[0].Rebase != nil || items[1].Rebase != nil { t.Fatalf("no episode, no pill: %+v %+v", items[0].Rebase, items[1].Rebase) } attempts.Observe("hivecommons/hive", 15087, "aaa1111aaa1111", now) - applyRebaseAttempts(items) + applyRebaseAttempts(items, now) if r := items[0].Rebase; r == nil || r.Attempt != 1 || r.Max != 2 || r.Exhausted || r.Head != "aaa1111" || r.Text != "Rebasing · attempt 1/2" { t.Fatalf("attempt 1 pill = %+v", r) } + // hivecommons/hive#16023: a dispatched attempt shows who it waits on. + attempts.Dispatch("hivecommons/hive", 15087, "scanner", now) + applyRebaseAttempts(items, now.Add(12*time.Minute)) + if r := items[0].Rebase; r == nil || r.Attempt != 1 || r.Text != "Rebasing · attempt 1/2 · waiting for scanner 12m" { + t.Fatalf("waiting pill = %+v", r) + } attempts.Observe("hivecommons/hive", 15087, "bbb2222", now) attempts.Observe("hivecommons/hive", 15087, "ccc3333", now) - applyRebaseAttempts(items) + applyRebaseAttempts(items, now) if r := items[0].Rebase; r == nil || !r.Exhausted || r.Attempt != 2 || r.Text != "Rebase failed · 2/2 — needs a person" { t.Fatalf("exhausted pill = %+v", r) } @@ -299,7 +305,7 @@ func TestRepoItemDecisionApproveOnDirtyPRQueuesRebase(t *testing.T) { t.Fatalf("a dirty head gets no update-branch: %d", mock.updates) } items := []StageTrackerItem{{Repo: "testorg/testrepo", Number: 8, Kind: "pr", Stage: 85}} - applyRebaseAttempts(items) + applyRebaseAttempts(items, time.Now()) if items[0].Rebase == nil || items[0].Rebase.Text != "Rebasing · attempt 1/2" { t.Fatalf("row = %+v", items[0].Rebase) } diff --git a/src/pkg/dashboard/stage_tracker_build.go b/src/pkg/dashboard/stage_tracker_build.go index a77175f644..58eca5d6d9 100644 --- a/src/pkg/dashboard/stage_tracker_build.go +++ b/src/pkg/dashboard/stage_tracker_build.go @@ -490,7 +490,7 @@ func (s *Server) buildStageTracker(status *StatusPayload, store *stagelabels.Sto resp.Items = stageTrackerDropOperatorClosed(resp.Items, now, operatorActionClosedGrace()) resp.OpenIssues, resp.OpenPRs = stageTrackerOpenCounts(resp.Items) s.applyDecidedProjection(resp.Items, now) - applyRebaseAttempts(resp.Items) + applyRebaseAttempts(resp.Items, now) s.applyRebaseMergeFollowThrough(resp.Items, now) resp.NeedsHuman, resp.OnHold = stageTrackerHumanCounts(resp.Items) resp.WaitingOnHuman = stageTrackerWaitingOnHuman(resp.Items) diff --git a/src/pkg/dashboard/static-src/js/config-governor.js b/src/pkg/dashboard/static-src/js/config-governor.js index 1d469f99d6..d24ca31645 100644 --- a/src/pkg/dashboard/static-src/js/config-governor.js +++ b/src/pkg/dashboard/static-src/js/config-governor.js @@ -799,7 +799,7 @@
-
Go duration after the comment-only nudge before an offline or external author's red PR is adopted. With live hub presence the hive instead waits ${yamlHint('governor.labels.stage.adoption.after_offline')} (default 30m) after a registered relay's last heartbeat, or ${yamlHint('governor.labels.stage.adoption.after_unregistered')} (default 2h) for an author with no profile here, and adopts in place on the PR's own branch when GitHub permits ${yamlHint('governor.labels.stage.adoption.in_place')} (default on) rather than opening a superseding PR. An in-place adoption keeps a changed head only when GitHub links the commit to the App bot ${yamlHint('governor.labels.stage.adoption.strict_attribution')} (default on); otherwise the hive hands the PR back to the contributor. Once the hive owns the head, the contributor lane merges it under the hive's own authority when the hive-authored gates pass (CI green on the adopted head, reviewer verdict, DCO, diff fences, soak), without requiring the original author's trust tier ${yamlHint('governor.labels.stage.adoption.auto_merge')} (default on). When the hive supersedes instead, it closes the original in the same pass that opens its copy — one note naming the copy, hive/superseded — and reopens it if the copy is closed unmerged or red at two heads ${yamlHint('governor.labels.stage.adoption.supersede.close_first')} (default on; off keeps the original open at stage 074 until the copy merges). A clanker- or hive-authored original closes in that pass by policy even without the repo's write-back opt-in label, since the copy carries its commits and sign-off ${yamlHint('governor.labels.stage.adoption.close_superseded')} (default on_adopt; on_merge waits for the copy to merge; a person's PR always waits unless the label is present). A copy that conflicts with the base is handed to the fixer's rebase-resolve episode (two attempts, then stage 085 rebase-failed-twice); an adoption attempt that opened nothing is retried once the author's clanker has stayed offline for ${yamlHint('governor.labels.stage.adoption.offline_grace')} (default 30m).
+
Go duration after the comment-only nudge before an offline or external author's red PR is adopted. With live hub presence the hive instead waits ${yamlHint('governor.labels.stage.adoption.after_offline')} (default 30m) after a registered relay's last heartbeat, or ${yamlHint('governor.labels.stage.adoption.after_unregistered')} (default 2h) for an author with no profile here, and adopts in place on the PR's own branch when GitHub permits ${yamlHint('governor.labels.stage.adoption.in_place')} (default on) rather than opening a superseding PR. An in-place adoption keeps a changed head only when GitHub links the commit to the App bot ${yamlHint('governor.labels.stage.adoption.strict_attribution')} (default on); otherwise the hive hands the PR back to the contributor. Once the hive owns the head, the contributor lane merges it under the hive's own authority when the hive-authored gates pass (CI green on the adopted head, reviewer verdict, DCO, diff fences, soak), without requiring the original author's trust tier ${yamlHint('governor.labels.stage.adoption.auto_merge')} (default on). When the hive supersedes instead, it closes the original in the same pass that opens its copy — one note naming the copy, hive/superseded — and reopens it if the copy is closed unmerged or red at two heads ${yamlHint('governor.labels.stage.adoption.supersede.close_first')} (default on; off keeps the original open at stage 074 until the copy merges). A clanker- or hive-authored original closes in that pass by policy even without the repo's write-back opt-in label, since the copy carries its commits and sign-off ${yamlHint('governor.labels.stage.adoption.close_superseded')} (default on_adopt; on_merge waits for the copy to merge; a person's PR always waits unless the label is present). A copy that conflicts with the base is handed to the fixer's rebase-resolve episode (two attempts, then stage 085 rebase-failed-twice); each fixer rebase attempt that sees no push within ${yamlHint('governor.labels.stage.adoption.rebase_attempt_timeout')} (default 15m) counts as failed, and an in-place adoption whose attempts are spent re-homes to a superseding PR at once; an adoption attempt that opened nothing is retried once the author's clanker has stayed offline for ${yamlHint('governor.labels.stage.adoption.offline_grace')} (default 30m).
${renderReporterTrustBlock()}

Writing guide

diff --git a/src/pkg/dashboard/static-src/js/config-state.js b/src/pkg/dashboard/static-src/js/config-state.js index 5dbbc34a06..433f6e0fb1 100644 --- a/src/pkg/dashboard/static-src/js/config-state.js +++ b/src/pkg/dashboard/static-src/js/config-state.js @@ -81,7 +81,7 @@ { tab: 'Labels', section: 'Labels', label: 'Stage dwell overrides', key: 'governor.labels.stage.dwell', text: 'stage dwell stalled threshold governor.labels.stage.dwell' }, { tab: 'Labels', section: 'Labels', label: 'Adoption of a maintainer\'s PR', key: 'governor.labels.stage.adoption.maintainer_policy', text: 'adoption maintainer policy adopt ask never maintainer own pr clanker disconnected offline adopt action governor.labels.stage.adoption.maintainer_policy' }, { tab: 'Labels', section: 'Labels', label: 'Contributor nudge', key: 'governor.labels.stage.contributor_nudge.enabled', text: 'contributor nudge returned-to-contributor red ci fix-pr relay interrupt renudge presence window governor.labels.stage.contributor_nudge.enabled governor.labels.stage.contributor_nudge.renudge_interval governor.labels.stage.presence_window' }, - { tab: 'Labels', section: 'Labels', label: 'Adopt red contributor PRs', key: 'governor.labels.stage.adoption.enabled', text: 'adoption adopted-by-hive stage 074 red contributor pr grace online offline supersede takeover in place after offline unregistered heartbeat governor.labels.stage.adoption.enabled governor.labels.stage.contributor_grace_online governor.labels.stage.contributor_grace_offline governor.labels.stage.adoption.after_offline governor.labels.stage.adoption.after_unregistered governor.labels.stage.adoption.in_place governor.labels.stage.adoption.strict_attribution strict attribution governor.labels.stage.adoption.auto_merge auto merge adopted head hive authority governor.labels.stage.adoption.supersede.close_first close first superseded reopen copy failed governor.labels.stage.adoption.close_superseded close superseded on adopt on merge clanker authored opt-in label governor.labels.stage.adoption.offline_grace offline grace rebase failed twice fixer resolving conflicts' }, + { tab: 'Labels', section: 'Labels', label: 'Adopt red contributor PRs', key: 'governor.labels.stage.adoption.enabled', text: 'adoption adopted-by-hive stage 074 red contributor pr grace online offline supersede takeover in place after offline unregistered heartbeat governor.labels.stage.adoption.enabled governor.labels.stage.contributor_grace_online governor.labels.stage.contributor_grace_offline governor.labels.stage.adoption.after_offline governor.labels.stage.adoption.after_unregistered governor.labels.stage.adoption.in_place governor.labels.stage.adoption.strict_attribution strict attribution governor.labels.stage.adoption.auto_merge auto merge adopted head hive authority governor.labels.stage.adoption.supersede.close_first close first superseded reopen copy failed governor.labels.stage.adoption.close_superseded close superseded on adopt on merge clanker authored opt-in label governor.labels.stage.adoption.offline_grace offline grace rebase failed twice fixer resolving conflicts governor.labels.stage.adoption.rebase_attempt_timeout rebase attempt timeout' }, { tab: 'Hub', section: 'Contributor Relay', label: 'Fix budget', key: 'contribute.max_parallel_fixes', text: 'fix budget parallel fixes focus window red pr fix-pr preempt spawn contribute.max_parallel_fixes contribute.min_focus_window' }, { tab: 'Logging', section: 'Logging', label: 'Governor logging', key: 'governor.logging.max_size_mb', text: 'governor logging directory level compression max size age backups governor.logging.compress governor.logging.dir governor.logging.level governor.logging.max_age_days governor.logging.max_backups governor.logging.max_size_mb' }, { tab: 'Advisory', section: 'Advisory', label: 'Advisory digest', key: 'governor.advisory.max_findings', text: 'advisory digest max findings show all staleness pr autoclose target linear issue retry backoff' }, diff --git a/src/pkg/dashboard/static/index.html b/src/pkg/dashboard/static/index.html index 02b1f175c6..cb3636afb3 100644 --- a/src/pkg/dashboard/static/index.html +++ b/src/pkg/dashboard/static/index.html @@ -50248,7 +50248,7 @@

Configure LiteLLM

{ tab: 'Labels', section: 'Labels', label: 'Stage dwell overrides', key: 'governor.labels.stage.dwell', text: 'stage dwell stalled threshold governor.labels.stage.dwell' }, { tab: 'Labels', section: 'Labels', label: 'Adoption of a maintainer\'s PR', key: 'governor.labels.stage.adoption.maintainer_policy', text: 'adoption maintainer policy adopt ask never maintainer own pr clanker disconnected offline adopt action governor.labels.stage.adoption.maintainer_policy' }, { tab: 'Labels', section: 'Labels', label: 'Contributor nudge', key: 'governor.labels.stage.contributor_nudge.enabled', text: 'contributor nudge returned-to-contributor red ci fix-pr relay interrupt renudge presence window governor.labels.stage.contributor_nudge.enabled governor.labels.stage.contributor_nudge.renudge_interval governor.labels.stage.presence_window' }, - { tab: 'Labels', section: 'Labels', label: 'Adopt red contributor PRs', key: 'governor.labels.stage.adoption.enabled', text: 'adoption adopted-by-hive stage 074 red contributor pr grace online offline supersede takeover in place after offline unregistered heartbeat governor.labels.stage.adoption.enabled governor.labels.stage.contributor_grace_online governor.labels.stage.contributor_grace_offline governor.labels.stage.adoption.after_offline governor.labels.stage.adoption.after_unregistered governor.labels.stage.adoption.in_place governor.labels.stage.adoption.strict_attribution strict attribution governor.labels.stage.adoption.auto_merge auto merge adopted head hive authority governor.labels.stage.adoption.supersede.close_first close first superseded reopen copy failed governor.labels.stage.adoption.close_superseded close superseded on adopt on merge clanker authored opt-in label governor.labels.stage.adoption.offline_grace offline grace rebase failed twice fixer resolving conflicts' }, + { tab: 'Labels', section: 'Labels', label: 'Adopt red contributor PRs', key: 'governor.labels.stage.adoption.enabled', text: 'adoption adopted-by-hive stage 074 red contributor pr grace online offline supersede takeover in place after offline unregistered heartbeat governor.labels.stage.adoption.enabled governor.labels.stage.contributor_grace_online governor.labels.stage.contributor_grace_offline governor.labels.stage.adoption.after_offline governor.labels.stage.adoption.after_unregistered governor.labels.stage.adoption.in_place governor.labels.stage.adoption.strict_attribution strict attribution governor.labels.stage.adoption.auto_merge auto merge adopted head hive authority governor.labels.stage.adoption.supersede.close_first close first superseded reopen copy failed governor.labels.stage.adoption.close_superseded close superseded on adopt on merge clanker authored opt-in label governor.labels.stage.adoption.offline_grace offline grace rebase failed twice fixer resolving conflicts governor.labels.stage.adoption.rebase_attempt_timeout rebase attempt timeout' }, { tab: 'Hub', section: 'Contributor Relay', label: 'Fix budget', key: 'contribute.max_parallel_fixes', text: 'fix budget parallel fixes focus window red pr fix-pr preempt spawn contribute.max_parallel_fixes contribute.min_focus_window' }, { tab: 'Logging', section: 'Logging', label: 'Governor logging', key: 'governor.logging.max_size_mb', text: 'governor logging directory level compression max size age backups governor.logging.compress governor.logging.dir governor.logging.level governor.logging.max_age_days governor.logging.max_backups governor.logging.max_size_mb' }, { tab: 'Advisory', section: 'Advisory', label: 'Advisory digest', key: 'governor.advisory.max_findings', text: 'advisory digest max findings show all staleness pr autoclose target linear issue retry backoff' }, @@ -57127,7 +57127,7 @@

Stage labels

-
Go duration after the comment-only nudge before an offline or external author's red PR is adopted. With live hub presence the hive instead waits ${yamlHint('governor.labels.stage.adoption.after_offline')} (default 30m) after a registered relay's last heartbeat, or ${yamlHint('governor.labels.stage.adoption.after_unregistered')} (default 2h) for an author with no profile here, and adopts in place on the PR's own branch when GitHub permits ${yamlHint('governor.labels.stage.adoption.in_place')} (default on) rather than opening a superseding PR. An in-place adoption keeps a changed head only when GitHub links the commit to the App bot ${yamlHint('governor.labels.stage.adoption.strict_attribution')} (default on); otherwise the hive hands the PR back to the contributor. Once the hive owns the head, the contributor lane merges it under the hive's own authority when the hive-authored gates pass (CI green on the adopted head, reviewer verdict, DCO, diff fences, soak), without requiring the original author's trust tier ${yamlHint('governor.labels.stage.adoption.auto_merge')} (default on). When the hive supersedes instead, it closes the original in the same pass that opens its copy — one note naming the copy, hive/superseded — and reopens it if the copy is closed unmerged or red at two heads ${yamlHint('governor.labels.stage.adoption.supersede.close_first')} (default on; off keeps the original open at stage 074 until the copy merges). A clanker- or hive-authored original closes in that pass by policy even without the repo's write-back opt-in label, since the copy carries its commits and sign-off ${yamlHint('governor.labels.stage.adoption.close_superseded')} (default on_adopt; on_merge waits for the copy to merge; a person's PR always waits unless the label is present). A copy that conflicts with the base is handed to the fixer's rebase-resolve episode (two attempts, then stage 085 rebase-failed-twice); an adoption attempt that opened nothing is retried once the author's clanker has stayed offline for ${yamlHint('governor.labels.stage.adoption.offline_grace')} (default 30m).
+
Go duration after the comment-only nudge before an offline or external author's red PR is adopted. With live hub presence the hive instead waits ${yamlHint('governor.labels.stage.adoption.after_offline')} (default 30m) after a registered relay's last heartbeat, or ${yamlHint('governor.labels.stage.adoption.after_unregistered')} (default 2h) for an author with no profile here, and adopts in place on the PR's own branch when GitHub permits ${yamlHint('governor.labels.stage.adoption.in_place')} (default on) rather than opening a superseding PR. An in-place adoption keeps a changed head only when GitHub links the commit to the App bot ${yamlHint('governor.labels.stage.adoption.strict_attribution')} (default on); otherwise the hive hands the PR back to the contributor. Once the hive owns the head, the contributor lane merges it under the hive's own authority when the hive-authored gates pass (CI green on the adopted head, reviewer verdict, DCO, diff fences, soak), without requiring the original author's trust tier ${yamlHint('governor.labels.stage.adoption.auto_merge')} (default on). When the hive supersedes instead, it closes the original in the same pass that opens its copy — one note naming the copy, hive/superseded — and reopens it if the copy is closed unmerged or red at two heads ${yamlHint('governor.labels.stage.adoption.supersede.close_first')} (default on; off keeps the original open at stage 074 until the copy merges). A clanker- or hive-authored original closes in that pass by policy even without the repo's write-back opt-in label, since the copy carries its commits and sign-off ${yamlHint('governor.labels.stage.adoption.close_superseded')} (default on_adopt; on_merge waits for the copy to merge; a person's PR always waits unless the label is present). A copy that conflicts with the base is handed to the fixer's rebase-resolve episode (two attempts, then stage 085 rebase-failed-twice); each fixer rebase attempt that sees no push within ${yamlHint('governor.labels.stage.adoption.rebase_attempt_timeout')} (default 15m) counts as failed, and an in-place adoption whose attempts are spent re-homes to a superseding PR at once; an adoption attempt that opened nothing is retried once the author's clanker has stayed offline for ${yamlHint('governor.labels.stage.adoption.offline_grace')} (default 30m).
${renderReporterTrustBlock()}

Writing guide

diff --git a/src/pkg/github/automerge/automerge_adoption_escape_test.go b/src/pkg/github/automerge/automerge_adoption_escape_test.go index 6e6da71e6f..1a61b222c5 100644 --- a/src/pkg/github/automerge/automerge_adoption_escape_test.go +++ b/src/pkg/github/automerge/automerge_adoption_escape_test.go @@ -302,7 +302,7 @@ func TestAdoptionEscapeHelpers(t *testing.T) { } for _, tc := range tests { t.Run(tc.name, func(t *testing.T) { - if got := inPlaceRehomeReason(tc.obs, tc.rec, policy, tc.at); got != tc.wantRehome { + if got := inPlaceRehomeReason(tc.obs, tc.rec, policy, tc.at, false); got != tc.wantRehome { t.Fatalf("inPlaceRehomeReason=%q, want %q", got, tc.wantRehome) } if got := adoptionStalledReason(tc.obs, tc.rec, policy, tc.at); got != tc.wantStall { @@ -312,6 +312,65 @@ func TestAdoptionEscapeHelpers(t *testing.T) { } } +// hivecommons/hive#16023: an in-place adoption whose fixer rebase episode +// is exhausted re-homes to a superseding PR at once instead of waiting out +// the 074 dwell; a live episode, a non-conflicting head or a moved head +// does not. +func TestInPlaceRehomeOnExhaustedRebaseEpisode(t *testing.T) { + policy := adoptPolicy() + now := time.Date(2026, 10, 31, 9, 0, 0, 0, time.UTC) + inPlace := stagelabels.Record{AdoptedHead: "aaa", AdoptMode: stagelabels.AdoptModeInPlace, AdoptedAt: now, AdoptedHeadAt: now} + conflict := stageObservation{Owner: "acme", Repo: "widget", Number: 7, HeadSHA: "aaa", Reason: "conflicting", LaneReason: "conflicting", Labels: []string{"needs-rebase"}} + red := stageObservation{Owner: "acme", Repo: "widget", Number: 7, HeadSHA: "aaa", Reason: "check-failing:unit", LaneReason: "check-failing:unit"} + moved := conflict + moved.HeadSHA = "bbb" + at := now.Add(30 * time.Minute) + tests := []struct { + name string + obs stageObservation + exhausted bool + want string + }{ + {"exhausted conflict re-homes before the dwell", conflict, true, AdoptReasonRebaseRehome}, + {"live episode waits for the dwell", conflict, false, ""}, + {"exhausted but not conflicting stays", red, true, ""}, + {"exhausted on a moved head stays", moved, true, ""}, + } + for _, tc := range tests { + t.Run(tc.name, func(t *testing.T) { + if got := inPlaceRehomeReason(tc.obs, inPlace, policy, at, tc.exhausted); got != tc.want { + t.Fatalf("inPlaceRehomeReason=%q, want %q", got, tc.want) + } + }) + } + if got := inPlaceRehomeReason(conflict, inPlace, policy, now.Add(adoptedDwell), false); got != AdoptReasonRebaseRehome { + t.Fatalf("dwell path still re-homes: %q", got) + } + + episodes := map[int]RebaseEpisode{7: {Attempt: 3, MaxAttempts: 2, Exhausted: true}, 8: {Attempt: 1, MaxAttempts: 2}} + e := &Engine{rebaseEpisode: func(repo string, number int) (RebaseEpisode, bool) { + if repo != "acme/widget" { + return RebaseEpisode{}, false + } + ep, ok := episodes[number] + return ep, ok + }} + live := conflict + live.Number = 8 + none := conflict + none.Number = 9 + supersede := stagelabels.Record{AdoptedHead: "aaa", AdoptMode: stagelabels.AdoptModeSupersede, AdoptedBy: 90, AdoptedAt: now} + if !e.inPlaceRebaseExhausted(conflict, inPlace) { + t.Fatal("exhausted episode on the in-place adoption not reported") + } + if e.inPlaceRebaseExhausted(live, inPlace) || e.inPlaceRebaseExhausted(none, inPlace) || e.inPlaceRebaseExhausted(conflict, supersede) { + t.Fatal("live, missing or supersede-mode episode reported exhausted") + } + if (&Engine{}).inPlaceRebaseExhausted(conflict, inPlace) { + t.Fatal("no episode reader reports exhausted") + } +} + func TestAdoptionBodyCauseAndBlockerText(t *testing.T) { for reason, want := range map[string]string{ AdoptReasonOffline: "adopted after CI went red", diff --git a/src/pkg/github/automerge/automerge_contributor_adopt.go b/src/pkg/github/automerge/automerge_contributor_adopt.go index 6198c0424a..b508953c34 100644 --- a/src/pkg/github/automerge/automerge_contributor_adopt.go +++ b/src/pkg/github/automerge/automerge_contributor_adopt.go @@ -1152,6 +1152,17 @@ func (c *Engine) adoptionCopyRebaseFailedTwice(obs stageObservation) bool { return ok && ep.Exhausted } +// inPlaceRebaseExhausted reports whether the fixer's rebase-resolve episode +// on this in-place adoption used every attempt without the head reading +// mergeable (hivecommons/hive#16023). +func (c *Engine) inPlaceRebaseExhausted(obs stageObservation, rec stagelabels.Record) bool { + if c.rebaseEpisode == nil || !rec.AdoptedInPlace() { + return false + } + ep, ok := c.rebaseEpisode(obs.Owner+"/"+obs.Repo, obs.Number) + return ok && ep.Exhausted +} + // priorCopyInputs keys a claim by the closed copy it follows, so a // re-adoption is not answered by the first adoption's journal entry. func priorCopyInputs(prior int, inputs map[string]string) map[string]string { @@ -1926,8 +1937,11 @@ func adoptedHeadSignoffUnfixable(obs stageObservation, rec stagelabels.Record) b // supersession this sweep, or "" to stay in place. A DCO-red head is left to // settleAdoption's DCO re-home, which carries the repair instructions. A // squash-unsafe hive sign-off re-homes immediately; rebase work first gets -// the normal fixer window on the contributor branch. -func inPlaceRehomeReason(obs stageObservation, rec stagelabels.Record, policy StageLabelPolicy, now time.Time) string { +// the normal fixer window on the contributor branch, which ends at the 074 +// dwell or as soon as the fixer's rebase-resolve episode on the adopted PR +// is exhausted (rebaseExhausted: every attempt pushed a dirty head or timed +// out with no push; hivecommons/hive#16023). +func inPlaceRehomeReason(obs stageObservation, rec stagelabels.Record, policy StageLabelPolicy, now time.Time, rebaseExhausted bool) string { if !rec.AdoptedInPlace() || obs.HeadSHA == "" || obs.HeadSHA != rec.AdoptedHead || adoptedHeadDCORed(obs, rec) || adoptionBlocked(obs.Labels) != "" || verdictNeedsHuman(obs.Detail) { return "" } @@ -1935,7 +1949,7 @@ func inPlaceRehomeReason(obs stageObservation, rec stagelabels.Record, policy St return AdoptReasonSignoffRehome } if humanConflictReasons[obs.Reason] || humanConflictReasons[obs.LaneReason] || hasLabelFold(obs.Labels, "needs-rebase") { - if adoptionStale(obs, rec, policy, now) { + if rebaseExhausted || adoptionStale(obs, rec, policy, now) { return AdoptReasonRebaseRehome } } @@ -1996,7 +2010,7 @@ func (c *Engine) escapeStalledAdoption(ctx context.Context, store *stagelabels.S c.info("stage labels: adopted head had no clock; starting the 074 dwell from this sweep", "repo", obs.Display, "pr", obs.Number, "head", obs.HeadSHA, "adopted_at", rec.AdoptedAt) rec, _ = store.Get(key) } - rehomeReason := inPlaceRehomeReason(obs, rec, policy, now) + rehomeReason := inPlaceRehomeReason(obs, rec, policy, now, c.inPlaceRebaseExhausted(obs, rec)) if rehomeReason == "" && rec.AdoptedInPlace() && obs.HeadSHA == rec.AdoptedHead && !adoptedHeadDCORed(obs, rec) && adoptionBlocked(obs.Labels) == "" && !verdictNeedsHuman(obs.Detail) && c.pushAccessLost(ctx, obs) { rehomeReason = AdoptReasonPushRehome } diff --git a/src/pkg/rebasemerge/attempts.go b/src/pkg/rebasemerge/attempts.go index 6568b20688..8042f170e3 100644 --- a/src/pkg/rebasemerge/attempts.go +++ b/src/pkg/rebasemerge/attempts.go @@ -22,13 +22,23 @@ var AttemptsPath = "/data/rebase-attempts.json" // Attempt is one PR's conflict episode: the dirty heads seen, oldest // first. The first dirty head is attempt 1 (the fixer's first task); a -// pushed head that is still dirty is the next attempt; past -// MaxRebaseAttempts dirty heads the episode is exhausted. A PR that stops -// conflicting ends its episode (Clear) so a later conflict starts fresh. +// pushed head that is still dirty is the next attempt, and so is an +// attempt the fixer never pushed for within the per-attempt deadline +// (Expire; hivecommons/hive#16023). Past MaxRebaseAttempts the episode is +// exhausted. A PR that stops conflicting ends its episode (Clear) so a +// later conflict starts fresh. type Attempt struct { Repo string `json:"repo"` Number int `json:"number"` Heads []string `json:"heads"` + // Stalled counts attempts that timed out with no push from the fixer; + // each one is a failed attempt. + Stalled int `json:"stalled,omitempty"` + // DispatchedAttempt is the attempt the router last handed to Fixer, + // at DispatchedAt; the router announces an attempt once. + DispatchedAttempt int `json:"dispatched_attempt,omitempty"` + DispatchedAt time.Time `json:"dispatched_at,omitempty"` + Fixer string `json:"fixer,omitempty"` // RequestedBy is the operator whose approval (or merge-now) asked for // the rebase from the dashboard, so the fixer router lists the PR even // when neither the hive nor a trusted author opened it. @@ -37,11 +47,26 @@ type Attempt struct { UpdatedAt time.Time `json:"updated_at"` } -// Count is the attempt number the current head represents (1-based). -func (a Attempt) Count() int { return len(a.Heads) } +// Count is the current attempt number (1-based): dirty heads seen plus +// attempts that timed out waiting for the fixer. +func (a Attempt) Count() int { return len(a.Heads) + a.Stalled } -// Exhausted: more dirty heads than the cap allows rebases. -func (a Attempt) Exhausted() bool { return len(a.Heads) > MaxRebaseAttempts } +// Exhausted: more attempts than the cap allows rebases. +func (a Attempt) Exhausted() bool { return a.Count() > MaxRebaseAttempts } + +// Dispatched reports whether the current attempt was handed to a fixer. +func (a Attempt) Dispatched() bool { + return a.DispatchedAttempt > 0 && a.DispatchedAttempt == a.Count() && !a.DispatchedAt.IsZero() +} + +// Waiting is how long the current attempt has waited for the fixer's +// push since it was dispatched; 0 when it was not dispatched. +func (a Attempt) Waiting(now time.Time) time.Duration { + if !a.Dispatched() || now.Before(a.DispatchedAt) { + return 0 + } + return now.Sub(a.DispatchedAt) +} // Head is the newest dirty head seen. func (a Attempt) Head() string { @@ -157,6 +182,48 @@ func (s *Attempts) Request(repo string, number int, head, actor string, now time return a } +// Dispatch records that the current attempt of repo#number's open episode +// was routed to fixer. first is true only the first time an attempt is +// dispatched, so the router announces each attempt once and later sweeps +// keep the original dispatch time. +func (s *Attempts) Dispatch(repo string, number int, fixer string, now time.Time) (a Attempt, first bool) { + s.mu.Lock() + defer s.mu.Unlock() + s.load() + key := Key(repo, number) + a, ok := s.items[key] + if !ok || a.Exhausted() || a.Dispatched() { + return a, false + } + a.DispatchedAttempt = a.Count() + a.DispatchedAt = now + a.Fixer = strings.TrimSpace(fixer) + a.UpdatedAt = now + s.items[key] = a + s.persist() + return a, true +} + +// Expire counts the dispatched attempt of repo#number as failed once it +// has waited timeout with no push from the fixer (a push is a new head, +// which Observe already counts as the next attempt). advanced is true when +// the episode moved to its next attempt; a timeout <= 0 never expires. +func (s *Attempts) Expire(repo string, number int, timeout time.Duration, now time.Time) (a Attempt, advanced bool) { + s.mu.Lock() + defer s.mu.Unlock() + s.load() + key := Key(repo, number) + a, ok := s.items[key] + if !ok || timeout <= 0 || a.Exhausted() || !a.Dispatched() || now.Sub(a.DispatchedAt) < timeout { + return a, false + } + a.Stalled++ + a.UpdatedAt = now + s.items[key] = a + s.persist() + return a, true +} + // Get returns the episode for repo#number, if one is open. func (s *Attempts) Get(repo string, number int) (Attempt, bool) { s.mu.Lock() @@ -192,3 +259,25 @@ func AttemptText(a Attempt) string { } return "Rebasing · attempt " + strconv.Itoa(n) + "/" + strconv.Itoa(MaxRebaseAttempts) } + +// AttemptTextAt is AttemptText with the wait on a dispatched attempt: +// "Rebasing · attempt 1/2 · waiting for scanner 12m". +func AttemptTextAt(a Attempt, now time.Time) string { + text := AttemptText(a) + if a.Exhausted() || !a.Dispatched() { + return text + } + fixer := a.Fixer + if fixer == "" { + fixer = "the fixer" + } + return text + " · waiting for " + fixer + " " + waitingText(a.Waiting(now)) +} + +func waitingText(d time.Duration) string { + m := int(d / time.Minute) + if m < 1 { + return "<1m" + } + return strconv.Itoa(m) + "m" +} diff --git a/src/pkg/rebasemerge/rebasemerge_test.go b/src/pkg/rebasemerge/rebasemerge_test.go index 91bfe644de..83dc28bfeb 100644 --- a/src/pkg/rebasemerge/rebasemerge_test.go +++ b/src/pkg/rebasemerge/rebasemerge_test.go @@ -211,6 +211,78 @@ func TestAttemptsLoopGuard(t *testing.T) { } } +// hivecommons/hive#16023: a dispatched attempt the fixer never pushes for +// counts as failed past the deadline, so the episode advances (and is +// exhausted) without a new dirty head; each attempt is dispatched once. +func TestAttemptsExpireAdvancesStalledAttempt(t *testing.T) { + path := filepath.Join(t.TempDir(), "attempts.json") + s := NewAttempts(path) + now := time.Date(2026, 10, 31, 9, 0, 0, 0, time.UTC) + const timeout = 15 * time.Minute + s.Observe("o/r", 1, "aaa", now) + if _, ok := s.Expire("o/r", 1, timeout, now.Add(time.Hour)); ok { + t.Fatal("an attempt never dispatched must not expire") + } + a, first := s.Dispatch("o/r", 1, "scanner", now) + if !first || !a.Dispatched() || a.Fixer != "scanner" || a.DispatchedAttempt != 1 { + t.Fatalf("first dispatch: %+v first=%v", a, first) + } + if a, first = s.Dispatch("o/r", 1, "scanner", now.Add(5*time.Minute)); first || !a.DispatchedAt.Equal(now) { + t.Fatalf("re-dispatch of the same attempt: %+v first=%v", a, first) + } + if got := AttemptTextAt(a, now.Add(12*time.Minute)); got != "Rebasing · attempt 1/2 · waiting for scanner 12m" { + t.Fatalf("caption: %q", got) + } + if got := AttemptTextAt(a, now.Add(30*time.Second)); got != "Rebasing · attempt 1/2 · waiting for scanner <1m" { + t.Fatalf("fresh caption: %q", got) + } + if _, ok := s.Expire("o/r", 1, timeout, now.Add(14*time.Minute)); ok { + t.Fatal("expired before the deadline") + } + if _, ok := s.Expire("o/r", 1, 0, now.Add(time.Hour)); ok { + t.Fatal("a zero timeout must never expire") + } + a, ok := s.Expire("o/r", 1, timeout, now.Add(timeout)) + if !ok || a.Count() != 2 || a.Stalled != 1 || a.Exhausted() || a.Dispatched() || a.Head() != "aaa" { + t.Fatalf("timed-out attempt 1: %+v ok=%v", a, ok) + } + if got := AttemptTextAt(a, now.Add(timeout)); got != "Rebasing · attempt 2/2" { + t.Fatalf("undispatched attempt 2 caption: %q", got) + } + if a, first = s.Dispatch("o/r", 1, "", now.Add(timeout)); !first || a.DispatchedAttempt != 2 { + t.Fatalf("attempt 2 dispatch: %+v first=%v", a, first) + } + if got := AttemptTextAt(a, now.Add(timeout+3*time.Minute)); got != "Rebasing · attempt 2/2 · waiting for the fixer 3m" { + t.Fatalf("no-fixer caption: %q", got) + } + a, ok = s.Expire("o/r", 1, timeout, now.Add(2*timeout)) + if !ok || !a.Exhausted() || AttemptTextAt(a, now.Add(2*timeout)) != "Rebase failed · 2/2 — needs a person" { + t.Fatalf("timed-out attempt 2: %+v ok=%v", a, ok) + } + if _, first = s.Dispatch("o/r", 1, "scanner", now.Add(2*timeout)); first { + t.Fatal("an exhausted episode must not be dispatched") + } + if _, ok = s.Expire("o/r", 1, timeout, now.Add(3*timeout)); ok { + t.Fatal("an exhausted episode must not expire again") + } + if got, ok := NewAttempts(path).Get("o/r", 1); !ok || got.Stalled != 2 || !got.Exhausted() { + t.Fatalf("reload: %+v ok=%v", got, ok) + } + // A push (new dirty head) is the next attempt and needs a fresh dispatch. + s.Clear("o/r", 1) + s.Observe("o/r", 1, "bbb", now) + s.Dispatch("o/r", 1, "scanner", now) + if a = s.Observe("o/r", 1, "ccc", now.Add(5*time.Minute)); a.Count() != 2 || a.Dispatched() { + t.Fatalf("pushed head: %+v", a) + } + if _, ok = s.Expire("o/r", 1, timeout, now.Add(time.Hour)); ok { + t.Fatal("a pushed attempt not yet dispatched must not expire") + } + if _, first = s.Dispatch("o/r", 9, "scanner", now); first { + t.Fatal("dispatch without an episode") + } +} + // Request opens the episode on an operator's behalf (approve / merge-now on // a dirty PR): the actor is recorded once, a blank actor records nothing, // and the request rides the same head count as the sweep's Observe.