diff --git a/modules/claude-licence-manager/cmd/tools/tools.go b/modules/claude-licence-manager/cmd/tools/tools.go index 6f4e647fb..ed4f9e2d4 100644 --- a/modules/claude-licence-manager/cmd/tools/tools.go +++ b/modules/claude-licence-manager/cmd/tools/tools.go @@ -92,6 +92,21 @@ func (r *Reader) Tools() []stdio.Tool { } } +// carriedForm gives an entry carried over from the old table `audit` (§12) the fields of §37: the old manager wrote a +// switch as `consumer` (one), `licence` (the one moved to), `from` and `by`, and no `consumers` or `to`. +func carriedForm(e map[string]any) { + if _, ok := e["to"]; !ok { + if l, ok := e["licence"]; ok { + e["to"] = l + } + } + if _, ok := e["consumers"]; !ok { + if c, ok := e["consumer"]; ok { + e["consumers"] = []any{c} + } + } +} + // Switches are the audit's switches, newest first: of a licence (from or to it) and a consumer when named, none before // since, at most limit (§37). func (r *Reader) Switches(licence, consumer string, since time.Time, limit int) ([]map[string]any, error) { @@ -118,6 +133,7 @@ func (r *Reader) Switches(licence, consumer string, since time.Time, limit int) if json.Unmarshal(all[i].Value, &e) != nil { continue } + carriedForm(e) at, _ := time.Parse(time.RFC3339Nano, fmt.Sprint(e["at"])) if at.Before(since) { continue diff --git a/modules/claude-licence-manager/cmd/tools/tools_test.go b/modules/claude-licence-manager/cmd/tools/tools_test.go index 6df0bf48b..0245e9895 100644 --- a/modules/claude-licence-manager/cmd/tools/tools_test.go +++ b/modules/claude-licence-manager/cmd/tools/tools_test.go @@ -229,6 +229,26 @@ func TestSwitchesAnswersBothCauses(t *testing.T) { } } +// design 39 bus-and-switch §37, §12: a switch carried over from the old table keeps the old field names (consumer, +// licence) and is answered with `to` and `consumers`, and found by them. +func TestSwitchesAnswersCarriedEntries(t *testing.T) { + rt := newRuntime(t) + rt.seedLog(licences.EntryAudit, "switched", map[string]any{"at": "2026-11-11T21:10:33Z", "consumer": "n1", + "licence": "d@example.org", "from": "c@example.org", "by": "a@example.org"}) + r := rt.reader() + for _, args := range []map[string]any{nil, {"consumer": "n1"}, {"licence": "d@example.org"}} { + got := run(t, r, "licences_switches", args).([]any) + if len(got) != 1 { + t.Fatalf("%v: %v", args, got) + } + e := got[0].(map[string]any) + if e["to"] != "d@example.org" || e["from"] != "c@example.org" || e["by"] != "a@example.org" || + len(e["consumers"].([]any)) != 1 || e["consumers"].([]any)[0] != "n1" { + t.Fatalf("%v: %v", args, e) + } + } +} + // design 39 bus-and-switch §40/AC1, B18, design 39 parts §3: a value set at node level is answered with its level, the // rest with the default, each with its value and default, and an invalid one saying why, read from the settings file; // mesh-config not answering is an error, never the default. diff --git a/modules/mesh-files/README.md b/modules/mesh-files/README.md index b2e492154..812a4b183 100644 --- a/modules/mesh-files/README.md +++ b/modules/mesh-files/README.md @@ -58,7 +58,7 @@ an override the override and the owned paths it went through. The caller is the ## What it refuses - A place a secret is kept (`MESH_SECRET_PLACES`, and the known credential files in `refuse.go`) is refused by - name for `read`, `list` and `search`. Copying or moving such a place is allowed: nothing is printed. + name for `read` and `search`; `list` answers its entries, never content, marked `secret-place`. Copying or moving such a place is allowed: nothing is printed. - A copy or move to or from `:/path` is refused: copying between nodes comes in step 2 (issue 903). - A write without a reason; a path that is not absolute; a mode that is not a mode. diff --git a/modules/mesh-files/cmd/mesh-files/files_test.go b/modules/mesh-files/cmd/mesh-files/files_test.go index d21eee723..aa809d1e3 100644 --- a/modules/mesh-files/cmd/mesh-files/files_test.go +++ b/modules/mesh-files/cmd/mesh-files/files_test.go @@ -24,6 +24,7 @@ type fakeRoot struct { calls []string cat map[string]string out map[string]string // the answer to a whole call, "bin arg arg" + fail map[string]error // the error of a whole call err error } @@ -31,6 +32,15 @@ func (r *fakeRoot) Run(bin string, args ...string) ([]byte, error) { r.mu.Lock() defer r.mu.Unlock() r.calls = append(r.calls, bin+" "+strings.Join(args, " ")) + for _, a := range append([]string{bin}, args...) { + if strings.ContainsRune(a, 0) { + // The kernel refuses a program argument holding a NUL byte, as exec does on a node. + return nil, fmt.Errorf("fork/exec /usr/bin/sudo: invalid argument") + } + } + if e, ok := r.fail[bin+" "+strings.Join(args, " ")]; ok { + return nil, e + } if o, ok := r.out[bin+" "+strings.Join(args, " ")]; ok { return []byte(o), nil } @@ -668,7 +678,7 @@ func TestTheReadsFallBackToRoot(t *testing.T) { t.Errorf("readlink through root: %+v %v", rl, err) } // list - g.root.out["/usr/bin/find -H "+in+" -mindepth 1 -maxdepth 1 -printf %y\t%U\t%G\t%m\t%s\t%l\t%f\x00"] = + g.root.out["/usr/bin/find -H "+in+" -mindepth 1 -maxdepth 1 -printf %y\t%U\t%G\t%m\t%s\t%l\t%f\\0"] = "f\t1000\t998\t640\t2\t\tf\x00d\t1000\t1000\t755\t4096\t\tsub dir\x00l\t1000\t1000\t777\t1\tf\tl\x00" ls, err := g.List(in) if err != nil || ls.Count != 3 { @@ -685,7 +695,7 @@ func TestTheReadsFallBackToRoot(t *testing.T) { t.Errorf("entries: %+v", ls.Entries) } // search: the files are listed by find, the secret ones skipped by name, the rest read through cat - g.root.out["/usr/bin/find -H "+locked+" -maxdepth 5 -xdev -type f -printf %p\x00"] = + g.root.out["/usr/bin/find -H "+locked+" -maxdepth 5 -xdev -type f -printf %p\\0"] = in + "/f\x00" + locked + "/.ssh/id_rsa\x00" g.root.cat[in+"/f"] = "one\nneedle here\n" g.root.reach(in + "/f") @@ -699,11 +709,20 @@ func TestTheReadsFallBackToRoot(t *testing.T) { if g.root.called("/usr/bin/cat -- " + locked + "/.ssh/id_rsa") { t.Error("root read a credential file") } - // a secret place is refused by name, and root is never asked - g.root.calls = nil - if _, err := g.List(g.path("home/.ssh")); err == nil { - t.Error("a secret place was listed") + // a secret place is listed, as root lists it, its entries marked and never their content (ADR 0394 §7); a path + // that is not there under a directory the account may not enter is said missing (design 39 parts §15) + secretDir := locked + "/.ssh" + g.root.out["/usr/bin/find -H "+secretDir+" -mindepth 1 -maxdepth 1 -printf %y\t%U\t%G\t%m\t%s\t%l\t%f\\0"] = + "f\t1000\t1000\t600\t9\t\tid_rsa\x00" + if ls, err := g.List(secretDir); err != nil || ls.Secret == "" || len(ls.Entries) != 1 || ls.Entries[0].Name != "id_rsa" { + t.Errorf("a secret place was not listed through root: %+v %v", ls, err) } + g.root.fail = map[string]error{"/usr/bin/stat -c " + statFormat + " -- " + in + "/gone": errors.New( + "sudo -n /usr/bin/stat: stat: cannot statx '" + in + "/gone': No such file or directory")} + if st, err := g.Stat([]string{in + "/gone"}, false); err != nil || !strings.Contains(st.Paths[0].Stopped, "does not exist") { + t.Errorf("a missing path under a closed directory: %+v %v", st, err) + } + g.root.calls = nil if _, err := g.Read(locked+"/.ssh/id_rsa", ReadAsk{}); err == nil || len(g.root.calls) != 0 { t.Errorf("a credential file was read: %v %v", err, g.root.calls) } diff --git a/modules/mesh-files/cmd/mesh-files/main.go b/modules/mesh-files/cmd/mesh-files/main.go index c6b7af439..68c6eb232 100644 --- a/modules/mesh-files/cmd/mesh-files/main.go +++ b/modules/mesh-files/cmd/mesh-files/main.go @@ -171,7 +171,9 @@ func object(props map[string]any, required ...string) map[string]any { // The descriptions of the reads are the controller's (internal/catalogue/files.go): the reads' own, with // where they are read as restated for the operator's account. -const asAccount = "as the node's operator account, through sudo -n where its rights do not reach; a place a " + +const readsAs = "as the node's operator account, through sudo -n where its rights do not reach" + +const asAccount = readsAs + "; a place a " + "secret is kept, or a known credential file, is refused by name" func boolArg(a map[string]any, k string) bool { return arg(a, k) == "true" } @@ -266,7 +268,7 @@ func tools(f *Files) []stdio.Tool { func readTools(r Reader) []stdio.Tool { ts := rawReadTools(r) for i := range ts { - if !strings.Contains(ts[i].Description, asAccount) { + if !strings.Contains(ts[i].Description, readsAs) { ts[i].Description += " Read " + asAccount + "." } } @@ -286,8 +288,9 @@ func rawReadTools(r Reader) []stdio.Tool { object(map[string]any{"path": str("an absolute path")}, "path"), func(a map[string]any) (any, error) { return r.Readlink(arg(a, "path")) }), verb("list", "Every entry of one directory with its type, owner, group, mode, size and, for a link, its "+ - "target. At most 2000 entries; a longer directory is cut and says so. A place a secret is kept is "+ - "refused by name.", + "target. At most 2000 entries; a longer directory is cut and says so. A place a secret is kept, at any depth "+ + "beneath it, is listed too and marked secret-place: its entries, never their content; what reaches a "+ + "process's memory or files (/proc//environ, mem, fd, map_files, root, cwd) is refused by name. Read "+readsAs+".", object(map[string]any{"path": str("an absolute path to a directory")}, "path"), func(a map[string]any) (any, error) { return r.List(arg(a, "path")) }), verb("read", "One text file, or a part of it by line: at most 256 KiB in one answer (64 KiB unless more "+ diff --git a/modules/mesh-files/cmd/mesh-files/reads.go b/modules/mesh-files/cmd/mesh-files/reads.go index db438d4db..e68452047 100644 --- a/modules/mesh-files/cmd/mesh-files/reads.go +++ b/modules/mesh-files/cmd/mesh-files/reads.go @@ -100,6 +100,12 @@ var errRefusedByName = errors.New("refused by name") // (/proc/self/fd) is judged by name like any path asked. A deleted file, or one the kernel names no path for, is // refused. func (r Reader) openJudged(path string, nofollow, dir bool) (*os.File, fs.FileInfo, error) { + return r.openJudgedIn(path, nofollow, dir, false) +} + +// openJudgedIn is openJudged where a list may open a place a secret is kept (listing it reads no content; +// novox/hq ADR 0394 §7): what reaches a process's memory or files stays refused. +func (r Reader) openJudgedIn(path string, nofollow, dir, listing bool) (*os.File, fs.FileInfo, error) { stat := os.Stat if nofollow { stat = os.Lstat @@ -136,9 +142,9 @@ func (r Reader) openJudged(path string, nofollow, dir bool) (*os.File, fs.FileIn return nil, nil, fmt.Errorf("%s is refused: the kernel names no path for what was opened, so it cannot be "+ "judged by name (novox/hq ADR 0394 §5)", path) } - if why := r.Refused(real); why != "" { + if why := r.Refused(real); why != "" && !(listing && !reachesProcess(real)) { f.Close() - return nil, nil, fmt.Errorf("%s is %s, which is refused by name: %s (novox/hq ADR 0394 §5): %w", path, real, why, errRefusedByName) + return nil, nil, fmt.Errorf("%s is %s, which is refused by name: %s (novox/hq ADR 0394 §7): %w", path, real, why, errRefusedByName) } return f, info, nil } @@ -308,6 +314,10 @@ var rootTypes = map[string]string{"regular file": "file", "regular empty file": func (r Reader) componentAsRoot(p string, names accounts, mounts map[string]Mount) (Component, bool) { out, err := r.Root.Run("/usr/bin/stat", "-c", statFormat, "--", p) f := strings.Split(strings.TrimSpace(string(out)), "|") + if err != nil && strings.Contains(err.Error(), "No such file or directory") { + // root reached the place and it is not there: said as missing, not as refused by the file modes. + return Component{Path: p, Unreachd: why(fs.ErrNotExist)}, true + } if err != nil || len(f) != 6 { return Component{}, false } @@ -478,6 +488,9 @@ type ListAnswer struct { Count int `json:"count"` Cut string `json:"cut,omitempty"` Withheld []Withheld `json:"withheld,omitempty"` + // Secret says why a read under the place is refused, when the directory is a place a secret is kept: its + // entries are answered, never their content (novox/hq ADR 0394 §7). + Secret string `json:"secret-place,omitempty"` } // refusedRead is a refusal by name, before any read: the path asked and where it leads are both judged, so a @@ -494,17 +507,45 @@ func (r Reader) refusedRead(path string) error { return nil } +// listedPlace judges a directory asked to be listed, and where it leads: a place a secret is kept is listed +// (its entries, never content) and said so; what reaches a process's memory or files is refused by name +// (novox/hq ADR 0394 §7). +func (r Reader) listedPlace(path string) (string, error) { + var secret string + judge := func(p, how string) error { + why := r.Refused(p) + switch { + case why == "": + case reachesProcess(p): + return fmt.Errorf("%s%s is refused by name: %s (novox/hq ADR 0394 §7)", path, how, why) + case secret == "": + secret = why + } + return nil + } + if err := judge(path, ""); err != nil { + return "", err + } + if leads, err := filepath.EvalSymlinks(path); err == nil && leads != path { + if err := judge(leads, " leads to "+leads+", which"); err != nil { + return "", err + } + } + return secret, nil +} + func (r Reader) List(raw string) (ListAnswer, error) { path, err := absolute(raw) if err != nil { return ListAnswer{}, err } - if err := r.refusedRead(path); err != nil { + secret, err := r.listedPlace(path) + if err != nil { return ListAnswer{}, err } names := r.names() var all []Entry - dir, _, err := r.openJudged(path, false, true) + dir, _, err := r.openJudgedIn(path, false, true, true) switch { case err != nil && r.Root != nil && errors.Is(err, fs.ErrPermission): if all, err = r.listAsRoot(path, names); err != nil { @@ -540,15 +581,15 @@ func (r Reader) List(raw string) (ListAnswer, error) { } all = append(all, en) } - return r.finishList(path, all, len(entries)), nil + return r.finishList(path, all, len(entries), secret), nil } - return r.finishList(path, all, len(all)), nil + return r.finishList(path, all, len(all), secret), nil } // finishList bounds a list and holds each link's target to the content rule. total is how many entries the // directory holds. -func (r Reader) finishList(path string, all []Entry, total int) ListAnswer { - a := ListAnswer{Path: path, Count: total, Entries: []Entry{}} +func (r Reader) finishList(path string, all []Entry, total int, secret string) ListAnswer { + a := ListAnswer{Path: path, Count: total, Entries: []Entry{}, Secret: secret} if total > MaxEntries { a.Cut = fmt.Sprintf("cut at %d of %d entries (the bound of one list), in name order", MaxEntries, total) } @@ -569,7 +610,7 @@ func (r Reader) finishList(path string, all []Entry, total int) ListAnswer { // a name cannot confuse it: the fields end with a NUL, and the name is the last. func (r Reader) listAsRoot(path string, names accounts) ([]Entry, error) { out, err := r.Root.Run("/usr/bin/find", "-H", path, "-mindepth", "1", "-maxdepth", "1", "-printf", - "%y\t%U\t%G\t%m\t%s\t%l\t%f\x00") + "%y\t%U\t%G\t%m\t%s\t%l\t%f\\0") // find writes the NUL: an argument may not hold one if err != nil { return nil, fmt.Errorf("%s cannot be listed, not even through sudo: %v", path, err) } @@ -926,7 +967,7 @@ func (r Reader) Search(ask SearchAsk) (SearchAnswer, error) { // and searches each: the names are judged first, so no credential file is read. func (r Reader) searchBelowAsRoot(p string, depth int, needle string, fold bool, bound int, deadline time.Time, a *SearchAnswer) error { out, err := r.Root.Run("/usr/bin/find", "-H", p, "-maxdepth", strconv.Itoa(depth), "-xdev", "-type", "f", - "-printf", "%p\x00") + "-printf", "%p\\0") // find writes the NUL: an argument may not hold one if err != nil { a.Unreadable++ return nil diff --git a/modules/mesh-files/cmd/mesh-files/reads_test.go b/modules/mesh-files/cmd/mesh-files/reads_test.go index 162e5f720..bb94ef3d4 100644 --- a/modules/mesh-files/cmd/mesh-files/reads_test.go +++ b/modules/mesh-files/cmd/mesh-files/reads_test.go @@ -85,8 +85,33 @@ func TestSecretPlacesAreRefusedByNameBeforeAnyRead(t *testing.T) { t.Errorf("%s was read: %v", p, err) } } - if _, err := r.List(vault); err == nil || !strings.Contains(err.Error(), "refused by name") { - t.Errorf("the vault's directory was listed: %v", err) + // a list of a place a secret is kept answers its entries, never content, marked (ADR 0394 §7) + if a, err := r.List(vault); err != nil || a.Secret == "" || a.Count != 1 || a.Entries[0].Name != "ledger" { + t.Errorf("the vault's directory was not listed as a secret place: %+v %v", a, err) + } + if a, err := r.List(filepath.Join(vault, "ledger")); err != nil || a.Secret == "" || a.Count != 1 { + t.Errorf("beneath the vault's directory: %+v %v", a, err) + } + if a, err := r.List(filepath.Join(dir, "home/ops")); err != nil || a.Secret != "" { + t.Errorf("a plain directory is marked: %+v %v", a, err) + } + // a link in a secret place: the entry is answered, its target held to the content rule (ADR 0394 §8) + if err := os.Symlink(aToken, filepath.Join(vault, "ledger/tokenlink")); err != nil { + t.Fatal(err) + } + if a, err := r.List(filepath.Join(vault, "ledger")); err != nil || a.Secret == "" || len(a.Withheld) != 1 || + a.Withheld[0].At != "a link's target" || strings.Contains(a.Entries[0].Link+a.Entries[1].Link, aToken) { + t.Errorf("a secret-shaped target in a secret place: %+v %v", a, err) + } + // the served text of list says it lists a secret place, and no longer that it refuses one + for _, tl := range readTools(r) { + if strings.HasSuffix(tl.Name, ".list") && (strings.Contains(tl.Description, "known credential file, is refused") || + !strings.HasSuffix(tl.Description, "where its rights do not reach.")) { + t.Errorf("the list text: %s", tl.Description) + } + } + if _, err := r.List("/proc/self/fd"); err == nil || !strings.Contains(err.Error(), "refused by name") { + t.Errorf("a process's open files were listed: %v", err) } link := filepath.Join(dir, "tmp/innocent") write(t, filepath.Join(dir, "tmp/x"), "") diff --git a/modules/mesh-files/cmd/mesh-files/refuse.go b/modules/mesh-files/cmd/mesh-files/refuse.go index df2d55865..afee44679 100644 --- a/modules/mesh-files/cmd/mesh-files/refuse.go +++ b/modules/mesh-files/cmd/mesh-files/refuse.go @@ -7,7 +7,7 @@ import ( "git.novox.be/novox/mesh-sdk/go/outward" ) -// What is refused by name, before any read (novox/hq ADR 0394 §5), and what an answer withholds (§4). +// What is refused by name, before any read (novox/hq ADR 0394 §7), and what an answer withholds (§4). // // **Two lists of places.** The places the mesh knows a secret is kept on this node — every module's own-secret // directory, every credential file and grant directory, the vault's directories — come from the controller, @@ -103,6 +103,12 @@ func (r Refuser) Refused(path string) string { return "" } +// reachesProcess says whether a path is an entry of a process that reaches its memory, environment or files. +func reachesProcess(path string) bool { + parts := strings.Split(strings.TrimPrefix(filepath.Clean(path), "/"), "/") + return len(parts) >= 3 && parts[0] == "proc" && procPrivate[parts[2]] +} + func within(path, place string) bool { return path == place || strings.HasPrefix(path, strings.TrimSuffix(place, "/")+"/") } diff --git a/modules/mesh-provisioner/cmd/mesh-provisioner/holder.go b/modules/mesh-provisioner/cmd/mesh-provisioner/holder.go index a9f3d1d6f..476dae2e5 100644 --- a/modules/mesh-provisioner/cmd/mesh-provisioner/holder.go +++ b/modules/mesh-provisioner/cmd/mesh-provisioner/holder.go @@ -28,6 +28,8 @@ package main import ( "bytes" "context" + "crypto/sha256" + "encoding/hex" "encoding/json" "errors" "fmt" @@ -105,8 +107,8 @@ type Composer func(ctx context.Context, node string, keptAt map[string]string) C // Composition is a node's plans at its wanted builds and at its newest, and where the time went. type Composition struct { AtWants, Newest HeldPlans - // Unpublished is each module whose composition at the build it is read at is not published yet: the node's picture - // is not whole, so no retirement is judged on it (issue 940). + // Unpublished is each module whose composition at the build it is read at is not published yet: its own plan and keys + // are kept as they were, and nothing else on the node waits for it (issue 940). Unpublished []string // UnpublishedNewest is each module whose composition at its newest build (which is not its wanted build) is not // published yet: the newest build's needs are unread, so no key only that build requires is judged (issue 1966). @@ -1070,6 +1072,25 @@ const askHeldWithin = time.Minute // askWaits is what a retiring key's why says while its ask waits for an open one to end. const askWaits = "its ask waits until fewer than 3 of mesh-provisioner's asks stand open, the router showing at most 3 per asker" +// askRefusedNote opens what a retiring key's why says once its ask is refused: the ask is not raised, and the key says +// so here, and provisioner_health names the key with those words. +const askRefusedNote = "its ask is refused: " + +// askID is a retiring ask's id: letters, digits and - only, of bounded length (the router refuses over 64), made from the +// key by its digest and the time, so one key's asks differ and no key's length matters. +func askID(key string, at time.Time) string { + sum := sha256.Sum256([]byte(key)) + return "retire-" + hex.EncodeToString(sum[:8]) + "-" + fmt.Sprint(at.Unix()) +} + +// withoutAskNotes is why without what ask() said of its ask (askWaits, askRefusedNote). +func withoutAskNotes(why string) string { + if i := strings.Index(why, askRefusedNote); i >= 0 { + why = why[:i] + } + return strings.TrimSuffix(strings.TrimSuffix(why, askWaits), "; ") +} + // errAskWaits is an ask held back by askOpenAtMost: no fault, raised once one ends. var errAskWaits = errors.New(askWaits) @@ -1095,21 +1116,26 @@ func (h *Holder) ask(p *meshprovisionerclient.Provision, key string) error { } if h.openAsks() >= askOpenAtMost { if !strings.Contains(p.Why, askWaits) { - p.Why = strings.TrimPrefix(p.Why+"; "+askWaits, "; ") + p.Why = strings.TrimPrefix(withoutAskNotes(p.Why)+"; "+askWaits, "; ") } return errAskWaits } now := h.Now().UTC() - id := "retire-" + strings.ReplaceAll(key, ".", "-") + "-" + fmt.Sprint(now.Unix()) + id := askID(key, now) a := retiringAsk(key, *p, now, id) - if err := occ.Check(a, now); err != nil { + // A refused ask is said in the key's why, where the operator reads why it stands, and on provisioner_health. + refused := func(err error) error { + p.Why = strings.TrimPrefix(withoutAskNotes(p.Why)+"; "+askRefusedNote+err.Error(), "; ") return err } + if err := occ.Check(a, now); err != nil { + return refused(err) + } if err := h.Asks.Publish(oc.AcceptSubjectAsk(Module), a, id); err != nil { - return fmt.Errorf("the retiring ask of %s could not be published: %w", key, err) + return refused(fmt.Errorf("the retiring ask of %s could not be published: %w", key, err)) } p.Ask = id - p.Why = strings.TrimSuffix(strings.TrimSuffix(p.Why, askWaits), "; ") + p.Why = withoutAskNotes(p.Why) return nil } @@ -1340,6 +1366,22 @@ func (h *Holder) Faults() map[string]string { return out } +// RefusedAsks is each retiring key whose ask the router's rules refuse, as its key with its why: read from the keys +// themselves, so a restart keeps it (ask). +func (h *Holder) RefusedAsks() map[string]string { + out := map[string]string{} + keys, _ := h.Store.Keys(ProvisionsState) + for _, k := range keys { + if e, err := h.Store.Get(ProvisionsState, k); err == nil && e != nil { + if p, err := meshprovisionerclient.Decode(e.Value); err == nil && p.State == meshprovisionerclient.Retiring && + p.Ask == "" && strings.Contains(p.Why, askRefusedNote) { + out[k] = p.Why + } + } + } + return out +} + // LastingHolds is each module held unpublished at one build for longer than unpublishedFor, as "." with // the hold's words, which name its build, and since when. func (h *Holder) LastingHolds() map[string]string { @@ -1514,8 +1556,8 @@ func orSaid(words ...string) string { // builds a module is read at the fingerprint keptAt names, else at the catalog entry's fingerprint (FP2); at the newest, each placed // module at FP2. A key whose engine refused the module holds it with the engine's words; a key not published yet holds // it at its wanted build (its last plan kept, as for a refusal) and at its newest too (issue 1966), whose build the engine -// writes once it is done (ADR 0593 §6), and says it unpublished, so the pass judges no retirement on the node (issue -// 940). A placed module no build of which is known is held too. A plan absent from a key is a module that hands the +// writes once it is done (ADR 0593 §6), and says it unpublished, so only that module's own plan and keys are kept as they were and nothing else on the +// node waits for it (issue 940). A placed module no build of which is known is held too. A plan absent from a key is a module that hands the // provisioner nothing. func FromCompositions(s Store, cat CatalogHalf) Composer { return func(_ context.Context, node string, keptAt map[string]string) Composition { diff --git a/modules/mesh-provisioner/cmd/mesh-provisioner/holder_test.go b/modules/mesh-provisioner/cmd/mesh-provisioner/holder_test.go index dc18625ef..dcd74f76f 100644 --- a/modules/mesh-provisioner/cmd/mesh-provisioner/holder_test.go +++ b/modules/mesh-provisioner/cmd/mesh-provisioner/holder_test.go @@ -3,6 +3,7 @@ package main import ( "context" "encoding/json" + "errors" "fmt" "go/ast" "go/parser" @@ -33,6 +34,8 @@ type router struct { cancelled []string // failCancel is a router that refuses a cancel. failCancel bool + // refuses is a router whose rules refuse every ask, with these words. + refuses string } func (r *router) Publish(subject string, body any, id string) error { @@ -47,6 +50,9 @@ func (r *router) Publish(subject string, body any, id string) error { delete(r.records, c.ID) return nil } + if r.refuses != "" { + return errors.New(r.refuses) + } r.asked[id] = body.(oc.Ask) if !r.drop { r.records[id], _ = json.Marshal(map[string]any{"state": "open", "ask": body}) @@ -852,3 +858,33 @@ func TestAnAskTheRouterDoesNotHoldIsRaisedAgain(t *testing.T) { } } } + +// Issue 1966, U7 (design 47 fix-unpublished-placed-module): a retiring ask's id is bounded and made from the key, so a +// long key is asked; an ask the router refuses is said in the key's why and in provisioner_health, and the key is +// asked once the refusal ends. +func TestARetiringAskIsBoundedAndARefusedOneIsSaid(t *testing.T) { + m := newMesh() + long := "claude-licence-manager-ab" // with this key the old id "retire-g14-...-" is 65 characters + key := "g14." + long + ".postgres-database" + m.atWant["g14"] = map[string]catalogue.Plan{long: planOf(pgDB), "dunst": planOf(oidc)} + m.compose(t, "g14") + m.atWant["g14"] = map[string]catalogue.Plan{"dunst": planOf(oidc)} + m.r.refuses = "its id is refused" + c := m.compose(t, "g14") + p := m.state(t, key) + if p.Ask != "" || !strings.Contains(p.Why, askRefusedNote) || !strings.Contains(p.Why, "its id is refused") || !strings.Contains(strings.Join(c.Problems, " "), "its id is refused") { + t.Errorf("a refused ask: ask %q, why %q, problems %v", p.Ask, p.Why, c.Problems) + } + if a := m.h.health(); a.Healthy || !strings.Contains(a.Why, key) || !strings.Contains(a.Why, "its id is refused") { + t.Errorf("health with a refused ask: %+v", a) + } + m.r.refuses = "" + m.compose(t, "g14") + p = m.state(t, key) + if len(p.Ask) == 0 || len(p.Ask) > 64 || !occ.UsableID(p.Ask) || strings.Contains(p.Why, askRefusedNote) { + t.Errorf("the ask once the refusal ends: %q, why %q", p.Ask, p.Why) + } + if a := m.h.health(); !a.Healthy { + t.Errorf("health once the ask is raised: %+v", a) + } +} diff --git a/modules/mesh-provisioner/cmd/mesh-provisioner/holdertools.go b/modules/mesh-provisioner/cmd/mesh-provisioner/holdertools.go index 778e488f2..227695642 100644 --- a/modules/mesh-provisioner/cmd/mesh-provisioner/holdertools.go +++ b/modules/mesh-provisioner/cmd/mesh-provisioner/holdertools.go @@ -91,17 +91,17 @@ func holderTools(h *Holder) []stdio.Tool { "a pass held back a retirement of most of a node's consumers at once as a fault of the mesh (issue 940), " + "naming the node, the consumers and the likely cause, or while a module's composition has stayed " + "unpublished at one build, wanted or newest, for longer than 45 minutes (longer than a build takes), naming the module, " + - "the build and since when.", + "the build and since when, or while a retiring provision's ask is refused by the operator channel's rules, naming the key and the refusal.", Run: func(map[string]any) (any, error) { return h.health(), nil }}, } } // health is the holder's runtime health: unhealthy while a retirement is held back as a fault of the mesh. func (h *Holder) health() health.Answer { - faults, holds := h.Faults(), h.LastingHolds() - if len(faults)+len(holds) == 0 { + faults, holds, refused := h.Faults(), h.LastingHolds(), h.RefusedAsks() + if len(faults)+len(holds)+len(refused) == 0 { return health.Healthy("No pass holds back a retirement of most of a node's consumers, and no module's " + - "composition has stayed unpublished at one build for longer than " + unpublishedFor.String() + ".") + "composition has stayed unpublished at one build for longer than " + unpublishedFor.String() + ", and no retiring ask is refused.") } var said []string for _, n := range sortedKeys(faults) { @@ -110,6 +110,9 @@ func (h *Holder) health() health.Answer { for _, k := range sortedKeys(holds) { said = append(said, k+": "+holds[k]+"; its plan and its provisions are kept as they were.") } + for _, k := range sortedKeys(refused) { + said = append(said, k+" is retiring and "+refused[k]+".") + } return health.Unhealthy(strings.Join(said, " ")) }