| 1 | package serve |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "log/slog" |
| 7 | "reasonix/internal/control" |
| 8 | "reasonix/internal/secrets" |
| 9 | "sync" |
| 10 | "time" |
| 11 | ) |
| 12 | |
| 13 | // One event-driven retry owner per foreground binding. A failed construction |
| 14 | // is not requeued; the next explicit retry or saved revision initiates it. |
| 15 | type modelApplicationRetry struct { |
| 16 | mu sync.Mutex |
| 17 | owner control.SessionAPI |
| 18 | pending, running, dirty bool |
| 19 | failedRevision string |
| 20 | failure string |
| 21 | } |
| 22 | |
| 23 | func (s *Server) cachedModelApplicationFailureLocked(ctx context.Context) error { |
| 24 | r := &s.modelApplicationRetry |
| 25 | owner := s.ctl() |
| 26 | r.mu.Lock() |
| 27 | revision, message, previous := r.failedRevision, r.failure, r.owner |
| 28 | r.mu.Unlock() |
| 29 | if previous != owner || revision == "" { |
| 30 | return nil |
| 31 | } |
| 32 | d := s.modelApplicationDetailsLocked(ctx) |
| 33 | if d != nil && d.DesiredRevision == revision { |
| 34 | return errors.New(message) |
| 35 | } |
| 36 | return nil |
| 37 | } |
| 38 | |
| 39 | func (s *Server) recordModelApplicationFailureLocked(ctx context.Context, err error) { |
| 40 | var failure *modelConstructionError |
| 41 | if !errors.As(err, &failure) { |
| 42 | return |
| 43 | } |
| 44 | owner := s.ctl() |
| 45 | r := &s.modelApplicationRetry |
| 46 | r.mu.Lock() |
| 47 | r.owner, r.failedRevision, r.failure = owner, failure.revision, secrets.RedactCredentials(err.Error()) |
| 48 | r.mu.Unlock() |
| 49 | slog.Warn("model configuration application failed", "phase", "build") |
| 50 | } |
| 51 | |
| 52 | type modelConstructionError struct { |
| 53 | cause error |
| 54 | revision string |
| 55 | } |
| 56 | |
| 57 | func (e *modelConstructionError) Error() string { return e.cause.Error() } |
| 58 | func (e *modelConstructionError) Unwrap() error { return e.cause } |
| 59 | |
| 60 | func (s *Server) modelConstructionFailure(err error) error { |
| 61 | revision := "" |
| 62 | if settings := s.managedModels; settings != nil && settings.SourceToken != "" { |
| 63 | revision = settings.Revision |
| 64 | } else if c, ok := s.ctl().(*control.Controller); ok { |
| 65 | _, revision, _ = c.ModelSettingsState() |
| 66 | } |
| 67 | return &modelConstructionError{cause: err, revision: revision} |
| 68 | } |
| 69 | |
| 70 | func (s *Server) deferModelApplicationLocked() { |
| 71 | r := &s.modelApplicationRetry |
| 72 | owner := s.ctl() |
| 73 | r.mu.Lock() |
| 74 | r.owner = owner |
| 75 | r.pending = true |
| 76 | r.failedRevision = "" |
| 77 | r.failure = "" |
| 78 | r.mu.Unlock() |
| 79 | s.kickModelApplication() |
| 80 | } |
| 81 | |
| 82 | func (s *Server) kickModelApplication() { |
| 83 | r := &s.modelApplicationRetry |
| 84 | r.mu.Lock() |
| 85 | if !r.pending { |
| 86 | r.mu.Unlock() |
| 87 | return |
| 88 | } |
| 89 | r.dirty = true |
| 90 | if r.running { |
| 91 | r.mu.Unlock() |
| 92 | return |
| 93 | } |
| 94 | r.running = true |
| 95 | r.mu.Unlock() |
| 96 | go func() { |
| 97 | for { |
| 98 | s.bindMu.Lock() |
| 99 | r.mu.Lock() |
| 100 | owner := r.owner |
| 101 | r.dirty = false |
| 102 | r.mu.Unlock() |
| 103 | if s.ctl() != owner || !control.ModelReplacementBlocked(owner) { |
| 104 | r.mu.Lock() |
| 105 | r.pending = false |
| 106 | r.mu.Unlock() |
| 107 | if s.ctl() == owner { |
| 108 | ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second) |
| 109 | _ = s.refreshRunModelSettingsLocked(ctx) |
| 110 | cancel() |
| 111 | } |
| 112 | } |
| 113 | s.bindMu.Unlock() |
| 114 | r.mu.Lock() |
| 115 | again := r.pending && r.dirty |
| 116 | if !again { |
| 117 | r.running = false |
| 118 | } |
| 119 | r.mu.Unlock() |
| 120 | if !again { |
| 121 | return |
| 122 | } |
| 123 | } |
| 124 | }() |
| 125 | } |
| 126 |