| 1 | package extension |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "sync" |
| 7 | "sync/atomic" |
| 8 | ) |
| 9 | |
| 10 | // Backend is a replaceable provider or MCP backend behind a StableProxy. |
| 11 | type Backend interface { |
| 12 | ID() string |
| 13 | // Close drains in-flight work owned by this backend. |
| 14 | Close(context.Context) error |
| 15 | } |
| 16 | |
| 17 | // StableProxy presents a stable consumer-facing handle while backends roll. |
| 18 | // It does not alter provider-visible prompt/tool prefixes; cache identity |
| 19 | // remains owned by RuntimeSnapshot.CacheHash. |
| 20 | type StableProxy struct { |
| 21 | mu sync.RWMutex |
| 22 | active Backend |
| 23 | draining []Backend |
| 24 | closed atomic.Bool |
| 25 | // generation of the currently active backend registration. |
| 26 | generation uint64 |
| 27 | // inFlight tracks CallCtx cancel funcs so Replace/Close/drain can abort |
| 28 | // mid-call work instead of hanging across a backend roll. |
| 29 | inFlight map[uint64]context.CancelFunc |
| 30 | nextCall atomic.Uint64 |
| 31 | } |
| 32 | |
| 33 | // NewStableProxy returns an empty proxy. |
| 34 | func NewStableProxy() *StableProxy { |
| 35 | return &StableProxy{inFlight: make(map[uint64]context.CancelFunc)} |
| 36 | } |
| 37 | |
| 38 | // Active returns the current backend, or nil. |
| 39 | func (p *StableProxy) Active() Backend { |
| 40 | if p == nil { |
| 41 | return nil |
| 42 | } |
| 43 | p.mu.RLock() |
| 44 | defer p.mu.RUnlock() |
| 45 | return p.active |
| 46 | } |
| 47 | |
| 48 | // Generation returns the active backend generation. |
| 49 | func (p *StableProxy) Generation() uint64 { |
| 50 | if p == nil { |
| 51 | return 0 |
| 52 | } |
| 53 | p.mu.RLock() |
| 54 | defer p.mu.RUnlock() |
| 55 | return p.generation |
| 56 | } |
| 57 | |
| 58 | // Replace swaps in a new backend and begins draining the previous one. |
| 59 | // Rolling replacement keeps the consumer pointer stable. In-flight CallCtx |
| 60 | // work is cancelled before the previous backend is closed. |
| 61 | func (p *StableProxy) Replace(ctx context.Context, next Backend, generation uint64) error { |
| 62 | if p == nil { |
| 63 | return fmt.Errorf("extension: nil StableProxy") |
| 64 | } |
| 65 | p.mu.Lock() |
| 66 | if p.closed.Load() { |
| 67 | p.mu.Unlock() |
| 68 | if next != nil { |
| 69 | _ = next.Close(ctx) |
| 70 | } |
| 71 | return fmt.Errorf("extension: proxy closed") |
| 72 | } |
| 73 | prev := p.active |
| 74 | p.active = next |
| 75 | p.generation = generation |
| 76 | if prev != nil { |
| 77 | p.draining = append(p.draining, prev) |
| 78 | } |
| 79 | cancels := p.takeInFlightLocked() |
| 80 | p.mu.Unlock() |
| 81 | for _, c := range cancels { |
| 82 | c() |
| 83 | } |
| 84 | if prev != nil { |
| 85 | if err := prev.Close(ctx); err != nil { |
| 86 | return fmt.Errorf("drain previous backend: %w", err) |
| 87 | } |
| 88 | p.mu.Lock() |
| 89 | out := p.draining[:0] |
| 90 | for _, b := range p.draining { |
| 91 | if b != prev { |
| 92 | out = append(out, b) |
| 93 | } |
| 94 | } |
| 95 | p.draining = out |
| 96 | p.mu.Unlock() |
| 97 | } |
| 98 | return nil |
| 99 | } |
| 100 | |
| 101 | // Close drains the active and any remaining backends after cancelling in-flight calls. |
| 102 | func (p *StableProxy) Close(ctx context.Context) error { |
| 103 | if p == nil { |
| 104 | return nil |
| 105 | } |
| 106 | if !p.closed.CompareAndSwap(false, true) { |
| 107 | return nil |
| 108 | } |
| 109 | p.mu.Lock() |
| 110 | active := p.active |
| 111 | p.active = nil |
| 112 | draining := append([]Backend(nil), p.draining...) |
| 113 | p.draining = nil |
| 114 | cancels := p.takeInFlightLocked() |
| 115 | p.mu.Unlock() |
| 116 | for _, c := range cancels { |
| 117 | c() |
| 118 | } |
| 119 | var first error |
| 120 | if active != nil { |
| 121 | first = active.Close(ctx) |
| 122 | } |
| 123 | for _, b := range draining { |
| 124 | if err := b.Close(ctx); err != nil && first == nil { |
| 125 | first = err |
| 126 | } |
| 127 | } |
| 128 | return first |
| 129 | } |
| 130 | |
| 131 | // Call invokes fn with the active backend. If no backend is registered the |
| 132 | // call fails fast so consumers do not hang across a crash/replace window. |
| 133 | func (p *StableProxy) Call(fn func(Backend) error) error { |
| 134 | return p.CallCtx(context.Background(), func(_ context.Context, b Backend) error { |
| 135 | return fn(b) |
| 136 | }) |
| 137 | } |
| 138 | |
| 139 | // CallCtx is Call with a parent context. The call is cancelled when the proxy |
| 140 | // is Replaced or Closed (drain of in-flight work). |
| 141 | func (p *StableProxy) CallCtx(ctx context.Context, fn func(context.Context, Backend) error) error { |
| 142 | if p == nil || p.closed.Load() { |
| 143 | return fmt.Errorf("extension: proxy unavailable") |
| 144 | } |
| 145 | if ctx == nil { |
| 146 | ctx = context.Background() |
| 147 | } |
| 148 | ctx, cancel := context.WithCancel(ctx) |
| 149 | id := p.nextCall.Add(1) |
| 150 | p.mu.Lock() |
| 151 | if p.inFlight == nil { |
| 152 | p.inFlight = make(map[uint64]context.CancelFunc) |
| 153 | } |
| 154 | p.inFlight[id] = cancel |
| 155 | b := p.active |
| 156 | p.mu.Unlock() |
| 157 | defer func() { |
| 158 | cancel() |
| 159 | p.mu.Lock() |
| 160 | delete(p.inFlight, id) |
| 161 | p.mu.Unlock() |
| 162 | }() |
| 163 | if b == nil { |
| 164 | return fmt.Errorf("extension: no active backend") |
| 165 | } |
| 166 | return fn(ctx, b) |
| 167 | } |
| 168 | |
| 169 | // CancelInFlight aborts every outstanding CallCtx. Used by generation drain. |
| 170 | func (p *StableProxy) CancelInFlight() { |
| 171 | if p == nil { |
| 172 | return |
| 173 | } |
| 174 | p.mu.Lock() |
| 175 | cancels := p.takeInFlightLocked() |
| 176 | p.mu.Unlock() |
| 177 | for _, c := range cancels { |
| 178 | c() |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | func (p *StableProxy) takeInFlightLocked() []context.CancelFunc { |
| 183 | if p == nil || len(p.inFlight) == 0 { |
| 184 | return nil |
| 185 | } |
| 186 | out := make([]context.CancelFunc, 0, len(p.inFlight)) |
| 187 | for id, c := range p.inFlight { |
| 188 | out = append(out, c) |
| 189 | delete(p.inFlight, id) |
| 190 | } |
| 191 | return out |
| 192 | } |
| 193 |