| 1 | package providerext |
| 2 | |
| 3 | import ( |
| 4 | "crypto/rand" |
| 5 | "encoding/hex" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "log/slog" |
| 9 | "sync" |
| 10 | "time" |
| 11 | |
| 12 | "reasonix/internal/extension/protocol" |
| 13 | "reasonix/internal/extension/providerconv" |
| 14 | "reasonix/internal/provider" |
| 15 | "reasonix/internal/secrets" |
| 16 | ) |
| 17 | |
| 18 | // extensionStream is one in-flight sidecar provider stream. Its fields are |
| 19 | // guarded by Resolver.mu and mirror the broker's hostStream: chunks arrive |
| 20 | // with 1-based seqs, buffer out of order, and flush contiguously; stream/end |
| 21 | // freezes the terminal boundary (LastSeq) and a gap timer converts a missing |
| 22 | // tail chunk into an interruption instead of a hang. |
| 23 | type extensionStream struct { |
| 24 | client ProviderClient |
| 25 | out chan provider.Chunk |
| 26 | done chan struct{} |
| 27 | abortDelivery chan struct{} |
| 28 | nextSeq int64 |
| 29 | pending map[int64]provider.Chunk |
| 30 | ended bool |
| 31 | endSeq int64 |
| 32 | endError string |
| 33 | interrupted bool |
| 34 | gapTimer bool |
| 35 | closeOnce sync.Once |
| 36 | delivery []provider.Chunk |
| 37 | deliveryWake chan struct{} |
| 38 | deliveryFinal bool |
| 39 | activity chan struct{} |
| 40 | unregisterDrainCancel func() |
| 41 | } |
| 42 | |
| 43 | // deliveryQueueLimit bounds the per-stream delivery queue without applying |
| 44 | // backpressure while Resolver.mu is held. One slot is reserved for the |
| 45 | // terminal chunk; mirroring the broker's hostDeliveryQueueLimit. |
| 46 | const deliveryQueueLimit = 256 |
| 47 | |
| 48 | // pendingWindowLimit bounds out-of-order buffering: chunks with a sequence at |
| 49 | // or beyond nextSeq+pendingWindowLimit never enter pending. A sidecar with a |
| 50 | // sequencing bug (emitting ever-higher seqs without the missing one or an |
| 51 | // end) must not grow host memory without limit — the stream is failed |
| 52 | // interrupted instead. |
| 53 | const pendingWindowLimit = 256 |
| 54 | |
| 55 | // RouteStreamChunk implements sidecar.StreamRouter. Unknown stream IDs are |
| 56 | // dropped with a debug log — a sidecar can legitimately race a late chunk |
| 57 | // against the host's cancel or its own crash teardown. Stale-generation |
| 58 | // chunks (after publish of a newer runtime) are also dropped. |
| 59 | func (r *Resolver) RouteStreamChunk(p protocol.StreamChunkParams) { |
| 60 | gen := p.Generation |
| 61 | if gen == 0 { |
| 62 | gen = p.Chunk.Generation |
| 63 | } |
| 64 | if r.owner.Gate.DropStale(gen, "provider_chunk") { |
| 65 | slog.Debug("providerext: dropping stale-generation chunk", "stream", p.StreamID, "seq", p.Seq, "generation", gen) |
| 66 | return |
| 67 | } |
| 68 | r.mu.Lock() |
| 69 | defer r.mu.Unlock() |
| 70 | stream := r.streams[p.StreamID] |
| 71 | if stream == nil { |
| 72 | slog.Debug("providerext: dropping chunk for unknown stream", "stream", p.StreamID, "seq", p.Seq) |
| 73 | return |
| 74 | } |
| 75 | signalStreamActivity(stream) |
| 76 | if p.Seq < stream.nextSeq { |
| 77 | return // duplicate or already delivered |
| 78 | } |
| 79 | if stream.ended && p.Seq > stream.endSeq { |
| 80 | r.finishLocked(p.StreamID, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 81 | Err: fmt.Errorf("extension stream %s: chunk seq %d exceeds frozen LastSeq %d", p.StreamID, p.Seq, stream.endSeq), |
| 82 | }}) |
| 83 | return |
| 84 | } |
| 85 | if p.Seq >= stream.nextSeq+pendingWindowLimit { |
| 86 | r.finishLocked(p.StreamID, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 87 | Err: fmt.Errorf("extension stream %s: chunk seq %d exceeds the pending window of stream seq %d", p.StreamID, p.Seq, stream.nextSeq), |
| 88 | }}) |
| 89 | return |
| 90 | } |
| 91 | stream.pending[p.Seq] = providerconv.ChunkFromProtocol(p.Chunk) |
| 92 | r.flushLocked(p.StreamID, stream) |
| 93 | } |
| 94 | |
| 95 | // RouteStreamEnd implements sidecar.StreamRouter. LastSeq freezes the |
| 96 | // terminal boundary: the stream completes only after chunks 1..LastSeq have |
| 97 | // been delivered, and a missing chunk trips the gap timer. |
| 98 | func (r *Resolver) RouteStreamEnd(p protocol.StreamEndParams) { |
| 99 | r.mu.Lock() |
| 100 | stream := r.streams[p.StreamID] |
| 101 | if stream == nil { |
| 102 | r.mu.Unlock() |
| 103 | slog.Debug("providerext: dropping stream end for unknown stream", "stream", p.StreamID) |
| 104 | return |
| 105 | } |
| 106 | signalStreamActivity(stream) |
| 107 | if stream.ended { |
| 108 | if stream.endSeq != p.LastSeq || stream.endError != p.Error || stream.interrupted != p.Interrupted { |
| 109 | r.finishLocked(p.StreamID, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 110 | Err: fmt.Errorf("extension stream %s: conflicting duplicate end changed frozen LastSeq or terminal state", p.StreamID), |
| 111 | }}) |
| 112 | r.mu.Unlock() |
| 113 | return |
| 114 | } |
| 115 | r.flushLocked(p.StreamID, stream) |
| 116 | r.mu.Unlock() |
| 117 | return |
| 118 | } |
| 119 | stream.ended = true |
| 120 | stream.endSeq = p.LastSeq |
| 121 | stream.endError = p.Error |
| 122 | stream.interrupted = p.Interrupted |
| 123 | for seq := range stream.pending { |
| 124 | if seq > stream.endSeq { |
| 125 | r.finishLocked(p.StreamID, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 126 | Err: fmt.Errorf("extension stream %s: buffered chunk seq %d exceeds frozen LastSeq %d", p.StreamID, seq, stream.endSeq), |
| 127 | }}) |
| 128 | r.mu.Unlock() |
| 129 | return |
| 130 | } |
| 131 | } |
| 132 | r.flushLocked(p.StreamID, stream) |
| 133 | if r.streams[p.StreamID] == stream && stream.nextSeq <= stream.endSeq && !stream.gapTimer { |
| 134 | stream.gapTimer = true |
| 135 | go r.expireGap(p.StreamID, stream) |
| 136 | } |
| 137 | r.mu.Unlock() |
| 138 | } |
| 139 | |
| 140 | func signalStreamActivity(stream *extensionStream) { |
| 141 | if stream == nil || stream.activity == nil { |
| 142 | return |
| 143 | } |
| 144 | select { |
| 145 | case stream.activity <- struct{}{}: |
| 146 | default: |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | // flushLocked delivers every contiguous pending chunk, then completes the |
| 151 | // stream once the end boundary is fully delivered. The terminal chunk mirrors |
| 152 | // the broker: a reported error is defensively credential-redacted even though |
| 153 | // the protocol also requires producer-side redaction, an interruption becomes |
| 154 | // StreamInterruptedError, and a clean end closes the channel. |
| 155 | func (r *Resolver) flushLocked(id string, stream *extensionStream) { |
| 156 | for !stream.ended || stream.nextSeq <= stream.endSeq { |
| 157 | chunk, ok := stream.pending[stream.nextSeq] |
| 158 | if !ok { |
| 159 | break |
| 160 | } |
| 161 | delete(stream.pending, stream.nextSeq) |
| 162 | stream.nextSeq++ |
| 163 | if !r.enqueueDeliveryLocked(stream, chunk) { |
| 164 | r.finishLocked(id, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 165 | Err: errors.New("extension provider stream output overflow"), |
| 166 | }}) |
| 167 | return |
| 168 | } |
| 169 | } |
| 170 | if !stream.ended || stream.nextSeq <= stream.endSeq { |
| 171 | return |
| 172 | } |
| 173 | if stream.endError != "" || stream.interrupted { |
| 174 | redactedEndError := secrets.RedactCredentials(stream.endError) |
| 175 | err := errors.New(redactedEndError) |
| 176 | if stream.endError == "" { |
| 177 | err = errors.New("extension provider stream failed") |
| 178 | } |
| 179 | if stream.interrupted { |
| 180 | message := redactedEndError |
| 181 | if message == "" { |
| 182 | message = "extension provider stream was interrupted" |
| 183 | } |
| 184 | err = &provider.StreamInterruptedError{Err: errors.New(message)} |
| 185 | } |
| 186 | r.finishLocked(id, stream, provider.Chunk{Type: provider.ChunkError, Err: err}) |
| 187 | return |
| 188 | } |
| 189 | r.finishLocked(id, stream, provider.Chunk{}) |
| 190 | } |
| 191 | |
| 192 | // finishLocked appends the terminal chunk (when non-zero) and marks delivery |
| 193 | // final. The first finish wins; later finishes for a replaced or completed |
| 194 | // stream are no-ops. |
| 195 | func (r *Resolver) finishLocked(id string, stream *extensionStream, terminal provider.Chunk) { |
| 196 | if r.streams[id] != stream { |
| 197 | return |
| 198 | } |
| 199 | delete(r.streams, id) |
| 200 | stream.closeOnce.Do(func() { |
| 201 | if stream.unregisterDrainCancel != nil { |
| 202 | stream.unregisterDrainCancel() |
| 203 | stream.unregisterDrainCancel = nil |
| 204 | } |
| 205 | if terminal.Err != nil || terminal.Type != 0 { |
| 206 | stream.delivery = append(stream.delivery, terminal) |
| 207 | } |
| 208 | stream.deliveryFinal = true |
| 209 | close(stream.done) |
| 210 | r.signalDeliveryLocked(stream) |
| 211 | }) |
| 212 | } |
| 213 | |
| 214 | // removeStream tears down a stream whose open never completed, aborting the |
| 215 | // delivery loop without a terminal chunk. |
| 216 | func (r *Resolver) removeStream(id string, stream *extensionStream) { |
| 217 | r.mu.Lock() |
| 218 | defer r.mu.Unlock() |
| 219 | if r.streams[id] == stream { |
| 220 | r.abortDeliveryLocked(stream) |
| 221 | r.finishLocked(id, stream, provider.Chunk{}) |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | func (r *Resolver) abortDeliveryLocked(stream *extensionStream) { |
| 226 | if stream.abortDelivery == nil { |
| 227 | return |
| 228 | } |
| 229 | select { |
| 230 | case <-stream.abortDelivery: |
| 231 | default: |
| 232 | close(stream.abortDelivery) |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | func (r *Resolver) enqueueDeliveryLocked(stream *extensionStream, chunk provider.Chunk) bool { |
| 237 | if stream.deliveryFinal || len(stream.delivery) >= deliveryQueueLimit-1 { |
| 238 | return false |
| 239 | } |
| 240 | stream.delivery = append(stream.delivery, chunk) |
| 241 | r.signalDeliveryLocked(stream) |
| 242 | return true |
| 243 | } |
| 244 | |
| 245 | func (r *Resolver) signalDeliveryLocked(stream *extensionStream) { |
| 246 | select { |
| 247 | case stream.deliveryWake <- struct{}{}: |
| 248 | default: |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | // deliverStream is the sole sender and closer of stream.out. It may wait for |
| 253 | // a slow consumer, but never while holding Resolver.mu, so routing, |
| 254 | // cancellation, disconnects, gap expiry, and unrelated streams keep moving. |
| 255 | func (r *Resolver) deliverStream(stream *extensionStream) { |
| 256 | defer close(stream.out) |
| 257 | for { |
| 258 | r.mu.Lock() |
| 259 | if len(stream.delivery) > 0 { |
| 260 | chunk := stream.delivery[0] |
| 261 | stream.delivery[0] = provider.Chunk{} |
| 262 | stream.delivery = stream.delivery[1:] |
| 263 | r.mu.Unlock() |
| 264 | select { |
| 265 | case stream.out <- chunk: |
| 266 | case <-stream.abortDelivery: |
| 267 | return |
| 268 | } |
| 269 | continue |
| 270 | } |
| 271 | if stream.deliveryFinal { |
| 272 | r.mu.Unlock() |
| 273 | return |
| 274 | } |
| 275 | wake := stream.deliveryWake |
| 276 | r.mu.Unlock() |
| 277 | select { |
| 278 | case <-wake: |
| 279 | case <-stream.abortDelivery: |
| 280 | return |
| 281 | } |
| 282 | } |
| 283 | } |
| 284 | |
| 285 | // expireGap converts a missing tail chunk into an interruption one second |
| 286 | // after stream/end: the LastSeq boundary froze, so the absent seq will never |
| 287 | // legitimately arrive. |
| 288 | func (r *Resolver) expireGap(id string, stream *extensionStream) { |
| 289 | timer := time.NewTimer(time.Second) |
| 290 | defer timer.Stop() |
| 291 | select { |
| 292 | case <-timer.C: |
| 293 | case <-stream.done: |
| 294 | return |
| 295 | } |
| 296 | r.mu.Lock() |
| 297 | defer r.mu.Unlock() |
| 298 | if r.streams[id] != stream || !stream.ended || stream.nextSeq > stream.endSeq { |
| 299 | return |
| 300 | } |
| 301 | r.finishLocked(id, stream, provider.Chunk{Type: provider.ChunkError, Err: &provider.StreamInterruptedError{ |
| 302 | Err: fmt.Errorf("extension provider stream missing chunk %d of %d", stream.nextSeq, stream.endSeq), |
| 303 | }}) |
| 304 | } |
| 305 | |
| 306 | func randomID(n int) string { |
| 307 | b := make([]byte, n) |
| 308 | if _, err := rand.Read(b); err != nil { |
| 309 | return fmt.Sprintf("%d", time.Now().UnixNano()) |
| 310 | } |
| 311 | return hex.EncodeToString(b) |
| 312 | } |
| 313 |