File
Blob: src/workerd/io/trace.h
| 1 | // Copyright (c) 2017-2022 Cloudflare, Inc. |
| 2 | // Licensed under the Apache 2.0 license found in the LICENSE file or at: |
| 3 | // https://opensource.org/licenses/Apache-2.0 |
| 4 | |
| 5 | #pragma once |
| 6 | |
| 7 | #include <workerd/io/outcome.capnp.h> |
| 8 | #include <workerd/io/trace.capnp.h> |
| 9 | #include <workerd/io/worker-interface.capnp.h> |
| 10 | #include <workerd/jsg/memory.h> |
| 11 | #include <workerd/util/own-util.h> |
| 12 | |
| 13 | #include <kj/map.h> |
| 14 | #include <kj/one-of.h> |
| 15 | #include <kj/refcount.h> |
| 16 | #include <kj/string.h> |
| 17 | #include <kj/time.h> |
| 18 | #include <kj/vector.h> |
| 19 | |
| 20 | #include <concepts> |
| 21 | #include <initializer_list> |
| 22 | |
| 23 | namespace kj { |
| 24 | enum class HttpMethod; |
| 25 | class EntropySource; |
| 26 | } // namespace kj |
| 27 | |
| 28 | namespace workerd { |
| 29 | |
| 30 | using kj::byte; |
| 31 | using kj::uint; |
| 32 | |
| 33 | using LogLevel = rpc::Trace::Log::Level; |
| 34 | using ExecutionModel = rpc::Trace::ExecutionModel; |
| 35 | |
| 36 | class Trace; |
| 37 | |
| 38 | namespace tracing { |
| 39 | // A 128-bit globally unique trace identifier. This will be used for both |
| 40 | // external and internal tracing. Specifically, for internal tracing, this |
| 41 | // is used to represent tracing IDs for jaeger traces. For external tracing, |
| 42 | // this is used for both the trace ID and invocation ID for tail workers. |
| 43 | class TraceId final { |
| 44 | public: |
| 45 | // A null trace ID. This is only acceptable for use in tests. |
| 46 | constexpr TraceId(decltype(nullptr)) {} |
| 47 | |
| 48 | // A trace ID with the given low and high values. |
| 49 | constexpr TraceId(uint64_t low, uint64_t high): low(low), high(high) {} |
| 50 | |
| 51 | constexpr TraceId(const TraceId& other) = default; |
| 52 | constexpr TraceId& operator=(const TraceId& other) = default; |
| 53 | |
| 54 | constexpr TraceId(TraceId&& other): low(other.low), high(other.high) { |
| 55 | other.low = 0; |
| 56 | other.high = 0; |
| 57 | } |
| 58 | |
| 59 | constexpr TraceId& operator=(TraceId&& other) { |
| 60 | low = other.low; |
| 61 | high = other.high; |
| 62 | other.low = 0; |
| 63 | other.high = 0; |
| 64 | return *this; |
| 65 | } |
| 66 | |
| 67 | constexpr TraceId& operator=(decltype(nullptr)) { |
| 68 | low = 0; |
| 69 | high = 0; |
| 70 | return *this; |
| 71 | } |
| 72 | |
| 73 | constexpr bool operator==(const TraceId& other) const { |
| 74 | return low == other.low && high == other.high; |
| 75 | } |
| 76 | constexpr bool operator==(decltype(nullptr)) const { |
| 77 | return low == 0 && high == 0; |
| 78 | } |
| 79 | constexpr operator bool() const { |
| 80 | return low || high; |
| 81 | } |
| 82 | |
| 83 | operator kj::String() const { |
| 84 | return toGoString(); |
| 85 | } |
| 86 | |
| 87 | // Replicates Jaeger go library's string serialization. |
| 88 | kj::String toGoString() const; |
| 89 | |
| 90 | // Replicates Jaeger go library's protobuf serialization. |
| 91 | kj::Array<byte> toProtobuf() const; |
| 92 | |
| 93 | // Replicates W3C Serialization |
| 94 | kj::String toW3C() const; |
| 95 | |
| 96 | // Creates a random Trace Id, optionally using a given entropy source. If an |
| 97 | // entropy source is not given, then we fallback to using BoringSSL's RAND_bytes. |
| 98 | static TraceId fromEntropy(kj::Maybe<kj::EntropySource&> entropy = kj::none); |
| 99 | |
| 100 | // Replicates Jaeger go library's string serialization. |
| 101 | static kj::Maybe<TraceId> fromGoString(kj::ArrayPtr<const char> s); |
| 102 | |
| 103 | // Replicates Jaeger go library's protobuf serialization. |
| 104 | static kj::Maybe<TraceId> fromProtobuf(kj::ArrayPtr<const kj::byte> buf); |
| 105 | |
| 106 | // A null trace ID. This is really only acceptable for use in tests. |
| 107 | static const TraceId nullId; |
| 108 | |
| 109 | inline uint64_t getLow() const { |
| 110 | return low; |
| 111 | } |
| 112 | inline uint64_t getHigh() const { |
| 113 | return high; |
| 114 | } |
| 115 | |
| 116 | static TraceId fromCapnp(rpc::TraceId::Reader reader); |
| 117 | void toCapnp(rpc::TraceId::Builder writer) const; |
| 118 | |
| 119 | private: |
| 120 | uint64_t low = 0; |
| 121 | uint64_t high = 0; |
| 122 | }; |
| 123 | constexpr TraceId TraceId::nullId = nullptr; |
| 124 | |
| 125 | // A 64-bit span identifier. |
| 126 | class SpanId final { |
| 127 | public: |
| 128 | // A null span ID. This is only acceptable for use in tests. |
| 129 | constexpr SpanId(decltype(nullptr)): id(0) {} |
| 130 | |
| 131 | constexpr SpanId(uint64_t id): id(id) {} |
| 132 | constexpr SpanId(const SpanId& other) = default; |
| 133 | constexpr SpanId& operator=(const SpanId& other) = default; |
| 134 | constexpr SpanId(SpanId&& other): id(other.id) { |
| 135 | other.id = 0; |
| 136 | } |
| 137 | constexpr SpanId& operator=(SpanId&& other) { |
| 138 | id = other.id; |
| 139 | other.id = 0; |
| 140 | return *this; |
| 141 | } |
| 142 | constexpr operator bool() const { |
| 143 | return id != 0; |
| 144 | } |
| 145 | constexpr bool operator==(const SpanId& other) const { |
| 146 | return id == other.id; |
| 147 | } |
| 148 | constexpr bool operator==(decltype(nullptr)) const { |
| 149 | return id == 0; |
| 150 | } |
| 151 | |
| 152 | inline operator kj::String() const { |
| 153 | return toGoString(); |
| 154 | } |
| 155 | |
| 156 | inline operator uint64_t() const { |
| 157 | return id; |
| 158 | } |
| 159 | |
| 160 | kj::String toGoString() const; |
| 161 | |
| 162 | static const SpanId nullId; |
| 163 | |
| 164 | constexpr uint64_t getId() const { |
| 165 | return id; |
| 166 | } |
| 167 | |
| 168 | static SpanId fromEntropy(kj::Maybe<kj::EntropySource&> entropy = kj::none); |
| 169 | |
| 170 | private: |
| 171 | uint64_t id; |
| 172 | }; |
| 173 | constexpr SpanId SpanId::nullId = nullptr; |
| 174 | // Fixed spanId value to be used for tests |
| 175 | constexpr uint64_t staticSpanId = 0x2a2a2a2a2a2a2a2aULL; |
| 176 | |
| 177 | // W3C trace flags propagated from an upstream traceparent. Wrapped in kj::Maybe |
| 178 | // at usage sites: kj::none means no upstream decision was made. |
| 179 | class TraceFlags final { |
| 180 | public: |
| 181 | explicit constexpr TraceFlags(uint8_t flags): flags(flags) {} |
| 182 | |
| 183 | constexpr bool isSampled() const { |
| 184 | return flags & SAMPLED; |
| 185 | } |
| 186 | |
| 187 | constexpr operator uint8_t() const { |
| 188 | return flags; |
| 189 | } |
| 190 | |
| 191 | constexpr bool operator==(const TraceFlags& other) const { |
| 192 | return flags == other.flags; |
| 193 | } |
| 194 | |
| 195 | private: |
| 196 | // W3C trace-flags bit: the caller requested this trace be recorded. |
| 197 | static constexpr uint8_t SAMPLED = 0x01; |
| 198 | |
| 199 | uint8_t flags; |
| 200 | }; |
| 201 | |
| 202 | // The InvocationSpanContext is a tuple of a trace id, invocation id, and span id. |
| 203 | // The trace id represents a top-level request and should be shared across all |
| 204 | // invocation spans and events within those spans. The invocation id identifies |
| 205 | // a specific worker invocation. The span id identifies a specific span within an |
| 206 | // invocation. Every invocation of every worker should have an InvocationSpanContext. |
| 207 | // That may or may not have a trigger InvocationSpanContext. |
| 208 | class InvocationSpanContext final { |
| 209 | public: |
| 210 | // The constructor is public only so kj::rc can see it and create a new instance. |
| 211 | // User code should use the static factory methods or the newChild method. |
| 212 | InvocationSpanContext(kj::Badge<InvocationSpanContext>, |
| 213 | kj::Maybe<kj::EntropySource&> entropySource, |
| 214 | TraceId traceId, |
| 215 | TraceId invocationId, |
| 216 | SpanId spanId, |
| 217 | kj::Maybe<const InvocationSpanContext&> parentSpanContext, |
| 218 | kj::Maybe<TraceFlags> traceFlags); |
| 219 | // Still need a constructor to be available as long as span context is not propagated everywhere |
| 220 | // we need it. |
| 221 | InvocationSpanContext( |
| 222 | TraceId traceId, TraceId invocationId, SpanId spanId, kj::Maybe<TraceFlags> traceFlags) |
| 223 | : traceId(traceId), |
| 224 | invocationId(invocationId), |
| 225 | spanId(spanId), |
| 226 | traceFlags(traceFlags) {}; |
| 227 | |
| 228 | KJ_DISALLOW_COPY(InvocationSpanContext); |
| 229 | |
| 230 | InvocationSpanContext(InvocationSpanContext&& other) = default; |
| 231 | InvocationSpanContext& operator=(InvocationSpanContext&& other) = default; |
| 232 | |
| 233 | inline bool operator==(const InvocationSpanContext& other) const { |
| 234 | return traceId == other.traceId && invocationId == other.invocationId && spanId == other.spanId; |
| 235 | } |
| 236 | |
| 237 | inline const TraceId& getTraceId() const { |
| 238 | return traceId; |
| 239 | } |
| 240 | |
| 241 | inline const TraceId& getInvocationId() const { |
| 242 | return invocationId; |
| 243 | } |
| 244 | |
| 245 | inline const SpanId& getSpanId() const { |
| 246 | return spanId; |
| 247 | } |
| 248 | |
| 249 | inline kj::Maybe<const InvocationSpanContext&> getParent() const { |
| 250 | KJ_IF_SOME(p, parentSpanContext) { |
| 251 | return *p; |
| 252 | } |
| 253 | return kj::none; |
| 254 | } |
| 255 | |
| 256 | // W3C trace flags propagated from an upstream traceparent. kj::none when |
| 257 | // no upstream sampling decision exists. |
| 258 | inline kj::Maybe<TraceFlags> getTraceFlags() const { |
| 259 | return traceFlags; |
| 260 | } |
| 261 | |
| 262 | // Creates a new child span. If the current context does not have an entropy |
| 263 | // source this will assert. If isTrigger() is true then it will not have an |
| 264 | // entropy source. |
| 265 | InvocationSpanContext newChild() const; |
| 266 | |
| 267 | // An InvocationSpanContext is a trigger context if it has no entropy source. |
| 268 | // This generally means the SpanContext was create from a capnp message and |
| 269 | // represents an InvocationSpanContext that was propagated from a parent |
| 270 | // or triggering context. |
| 271 | bool isTrigger() const { |
| 272 | return entropySource == kj::none; |
| 273 | } |
| 274 | |
| 275 | // Creates a new InvocationSpanContext. If the triggerContext is given, then its |
| 276 | // traceId is used as the traceId for the newly created context. Otherwise a new |
| 277 | // traceId is generated. The invocationId is always generated new and the spanId |
| 278 | // will be 0 with no parent span. |
| 279 | static InvocationSpanContext newForInvocation( |
| 280 | kj::Maybe<const InvocationSpanContext&> triggerContext = kj::none, |
| 281 | kj::Maybe<kj::EntropySource&> entropySource = kj::none); |
| 282 | |
| 283 | // Creates a new InvocationSpanContext from a capnp message. The returned |
| 284 | // InvocationSpanContext will not be capable of creating child spans and |
| 285 | // is considered only a "trigger" span. |
| 286 | static kj::Maybe<InvocationSpanContext> fromCapnp(rpc::InvocationSpanContext::Reader reader); |
| 287 | void toCapnp(rpc::InvocationSpanContext::Builder writer) const; |
| 288 | InvocationSpanContext clone() const; |
| 289 | |
| 290 | private: |
| 291 | // If there is no entropy source, then child spans cannot be created from |
| 292 | // this InvocationSpanContext. |
| 293 | kj::Maybe<kj::EntropySource&> entropySource; |
| 294 | TraceId traceId; |
| 295 | TraceId invocationId; |
| 296 | SpanId spanId; |
| 297 | |
| 298 | // The parentSpanContext can be either a direct parent or a trigger |
| 299 | // context. If it is a trigger context, then it should have the same |
| 300 | // traceId but a different invocationId (unless predictable mode for |
| 301 | // testing is enabled). The isTrigger() should also return true. |
| 302 | kj::Maybe<kj::Own<InvocationSpanContext>> parentSpanContext; |
| 303 | |
| 304 | // W3C trace flags from an upstream traceparent, propagated through the |
| 305 | // invocation chain. kj::none when no upstream sampling decision was made. |
| 306 | kj::Maybe<TraceFlags> traceFlags; |
| 307 | }; |
| 308 | |
| 309 | // SpanContext as used for streaming tail worker tail events. spanId is always set except for Onset |
| 310 | // events that don't inherit context from another invocation. |
| 311 | struct SpanContext { |
| 312 | SpanContext( |
| 313 | TraceId traceId, kj::Maybe<SpanId> spanId, kj::Maybe<TraceFlags> traceFlags = kj::none) |
| 314 | : traceId(traceId), |
| 315 | spanId(spanId), |
| 316 | traceFlags(traceFlags) {}; |
| 317 | KJ_DISALLOW_COPY(SpanContext); |
| 318 | SpanContext(SpanContext&& other) = default; |
| 319 | SpanContext& operator=(SpanContext&& other) = default; |
| 320 | |
| 321 | inline bool operator==(const SpanContext& other) const { |
| 322 | return traceId == other.traceId && spanId == other.spanId; |
| 323 | } |
| 324 | |
| 325 | inline const TraceId& getTraceId() const { |
| 326 | return traceId; |
| 327 | } |
| 328 | |
| 329 | inline kj::Maybe<SpanId> getSpanId() const { |
| 330 | return spanId; |
| 331 | } |
| 332 | |
| 333 | // W3C trace flags from an upstream traceparent. kj::none when no upstream |
| 334 | // sampling decision was made (e.g. subrequest propagation without a |
| 335 | // worker_tracing header). |
| 336 | inline kj::Maybe<TraceFlags> getTraceFlags() const { |
| 337 | return traceFlags; |
| 338 | } |
| 339 | |
| 340 | static SpanContext fromCapnp(rpc::SpanContext::Reader reader); |
| 341 | void toCapnp(rpc::SpanContext::Builder writer) const; |
| 342 | static SpanContext clone(const SpanContext& ctx) { |
| 343 | return SpanContext(ctx.traceId, ctx.spanId, ctx.traceFlags); |
| 344 | } |
| 345 | |
| 346 | // Parse a W3C traceparent string into a SpanContext. |
| 347 | // Format: "{version}-{trace-id}-{parent-id}-{flags}" |
| 348 | static kj::Maybe<SpanContext> tryFromTraceparent(kj::StringPtr traceparent); |
| 349 | |
| 350 | private: |
| 351 | TraceId traceId; |
| 352 | kj::Maybe<SpanId> spanId; |
| 353 | kj::Maybe<TraceFlags> traceFlags; |
| 354 | }; |
| 355 | |
| 356 | kj::String KJ_STRINGIFY(const SpanId& id); |
| 357 | kj::String KJ_STRINGIFY(const TraceId& id); |
| 358 | kj::String KJ_STRINGIFY(const InvocationSpanContext& context); |
| 359 | kj::String KJ_STRINGIFY(const SpanContext& context); |
| 360 | |
| 361 | // The various structs defined below are used in both buffered tail workers |
| 362 | // and streaming tail workers to report tail events. |
| 363 | |
| 364 | // Describes a fetch request |
| 365 | struct FetchEventInfo final { |
| 366 | struct Header; |
| 367 | |
| 368 | explicit FetchEventInfo( |
| 369 | kj::HttpMethod method, kj::String url, kj::String cfJson, kj::Array<Header> headers); |
| 370 | FetchEventInfo(rpc::Trace::FetchEventInfo::Reader reader); |
| 371 | FetchEventInfo(FetchEventInfo&&) noexcept = default; |
| 372 | FetchEventInfo& operator=(FetchEventInfo&&) = default; |
| 373 | KJ_DISALLOW_COPY(FetchEventInfo); |
| 374 | |
| 375 | struct Header final { |
| 376 | explicit Header(kj::String name, kj::String value); |
| 377 | Header(rpc::Trace::FetchEventInfo::Header::Reader reader); |
| 378 | Header(Header&&) noexcept = default; |
| 379 | Header& operator=(Header&&) = default; |
| 380 | KJ_DISALLOW_COPY(Header); |
| 381 | |
| 382 | kj::String name; |
| 383 | kj::String value; |
| 384 | |
| 385 | void copyTo(rpc::Trace::FetchEventInfo::Header::Builder builder) const; |
| 386 | Header clone() const; |
| 387 | kj::String toString() const; |
| 388 | |
| 389 | JSG_MEMORY_INFO(Header) { |
| 390 | tracker.trackField("name", name); |
| 391 | tracker.trackField("value", value); |
| 392 | } |
| 393 | }; |
| 394 | |
| 395 | kj::HttpMethod method; |
| 396 | kj::String url; |
| 397 | // TODO(perf): It might be more efficient to store some sort of parsed JSON result instead? |
| 398 | kj::String cfJson; |
| 399 | kj::Array<Header> headers; |
| 400 | |
| 401 | void copyTo(rpc::Trace::FetchEventInfo::Builder builder) const; |
| 402 | FetchEventInfo clone() const; |
| 403 | kj::String toString() const; |
| 404 | }; |
| 405 | |
| 406 | // Describes a jsrpc request |
| 407 | struct JsRpcEventInfo final { |
| 408 | explicit JsRpcEventInfo(kj::String methodName); |
| 409 | JsRpcEventInfo(rpc::Trace::JsRpcEventInfo::Reader reader); |
| 410 | JsRpcEventInfo(JsRpcEventInfo&&) noexcept = default; |
| 411 | JsRpcEventInfo& operator=(JsRpcEventInfo&&) = default; |
| 412 | KJ_DISALLOW_COPY(JsRpcEventInfo); |
| 413 | |
| 414 | kj::String methodName; |
| 415 | |
| 416 | void copyTo(rpc::Trace::JsRpcEventInfo::Builder builder) const; |
| 417 | JsRpcEventInfo clone() const; |
| 418 | kj::String toString() const; |
| 419 | }; |
| 420 | |
| 421 | class ConnectEventInfo { |
| 422 | public: |
| 423 | explicit ConnectEventInfo(); |
| 424 | explicit ConnectEventInfo(rpc::Trace::ConnectEventInfo::Reader reader); |
| 425 | |
| 426 | void copyTo(rpc::Trace::ConnectEventInfo::Builder builder) const; |
| 427 | ConnectEventInfo clone() const; |
| 428 | }; |
| 429 | |
| 430 | // Describes a scheduled request |
| 431 | struct ScheduledEventInfo final { |
| 432 | explicit ScheduledEventInfo(double scheduledTime, kj::String cron); |
| 433 | ScheduledEventInfo(rpc::Trace::ScheduledEventInfo::Reader reader); |
| 434 | ScheduledEventInfo(ScheduledEventInfo&&) noexcept = default; |
| 435 | ScheduledEventInfo& operator=(ScheduledEventInfo&&) = default; |
| 436 | KJ_DISALLOW_COPY(ScheduledEventInfo); |
| 437 | |
| 438 | double scheduledTime; |
| 439 | kj::String cron; |
| 440 | |
| 441 | void copyTo(rpc::Trace::ScheduledEventInfo::Builder builder) const; |
| 442 | ScheduledEventInfo clone() const; |
| 443 | }; |
| 444 | |
| 445 | // Describes a Durable Object alarm request |
| 446 | struct AlarmEventInfo final { |
| 447 | explicit AlarmEventInfo(kj::Date scheduledTime); |
| 448 | AlarmEventInfo(rpc::Trace::AlarmEventInfo::Reader reader); |
| 449 | AlarmEventInfo(AlarmEventInfo&&) noexcept = default; |
| 450 | AlarmEventInfo& operator=(AlarmEventInfo&&) = default; |
| 451 | KJ_DISALLOW_COPY(AlarmEventInfo); |
| 452 | |
| 453 | kj::Date scheduledTime; |
| 454 | |
| 455 | void copyTo(rpc::Trace::AlarmEventInfo::Builder builder) const; |
| 456 | AlarmEventInfo clone() const; |
| 457 | }; |
| 458 | |
| 459 | // Describes a queue worker request |
| 460 | struct QueueEventInfo final { |
| 461 | explicit QueueEventInfo(kj::String queueName, uint32_t batchSize); |
| 462 | QueueEventInfo(rpc::Trace::QueueEventInfo::Reader reader); |
| 463 | QueueEventInfo(QueueEventInfo&&) noexcept = default; |
| 464 | QueueEventInfo& operator=(QueueEventInfo&&) = default; |
| 465 | KJ_DISALLOW_COPY(QueueEventInfo); |
| 466 | |
| 467 | kj::String queueName; |
| 468 | uint32_t batchSize; |
| 469 | |
| 470 | void copyTo(rpc::Trace::QueueEventInfo::Builder builder) const; |
| 471 | QueueEventInfo clone() const; |
| 472 | }; |
| 473 | |
| 474 | // Describes an email request |
| 475 | struct EmailEventInfo final { |
| 476 | explicit EmailEventInfo(kj::String mailFrom, kj::String rcptTo, uint32_t rawSize); |
| 477 | EmailEventInfo(rpc::Trace::EmailEventInfo::Reader reader); |
| 478 | EmailEventInfo(EmailEventInfo&&) noexcept = default; |
| 479 | EmailEventInfo& operator=(EmailEventInfo&&) = default; |
| 480 | KJ_DISALLOW_COPY(EmailEventInfo); |
| 481 | |
| 482 | kj::String mailFrom; |
| 483 | kj::String rcptTo; |
| 484 | uint32_t rawSize; |
| 485 | |
| 486 | void copyTo(rpc::Trace::EmailEventInfo::Builder builder) const; |
| 487 | EmailEventInfo clone() const; |
| 488 | }; |
| 489 | |
| 490 | // Describes a buffered tail worker request |
| 491 | struct TracePreview final { |
| 492 | explicit TracePreview(kj::String id, kj::String slug, kj::String name); |
| 493 | TracePreview(rpc::Trace::TracePreviewInfo::Reader reader); |
| 494 | TracePreview(TracePreview&&) noexcept = default; |
| 495 | TracePreview& operator=(TracePreview&&) = default; |
| 496 | KJ_DISALLOW_COPY(TracePreview); |
| 497 | |
| 498 | kj::String id; |
| 499 | kj::String slug; |
| 500 | kj::String name; |
| 501 | |
| 502 | void copyTo(rpc::Trace::TracePreviewInfo::Builder builder) const; |
| 503 | TracePreview clone() const; |
| 504 | }; |
| 505 | |
| 506 | struct TraceEventInfo final { |
| 507 | struct TraceItem; |
| 508 | |
| 509 | explicit TraceEventInfo(kj::ArrayPtr<const kj::Own<Trace>> traces); |
| 510 | TraceEventInfo(kj::Array<TraceItem> traces): traces(kj::mv(traces)) {} |
| 511 | TraceEventInfo(rpc::Trace::TraceEventInfo::Reader reader); |
| 512 | TraceEventInfo(TraceEventInfo&&) noexcept = default; |
| 513 | TraceEventInfo& operator=(TraceEventInfo&&) = default; |
| 514 | KJ_DISALLOW_COPY(TraceEventInfo); |
| 515 | |
| 516 | struct TraceItem final { |
| 517 | explicit TraceItem(kj::Maybe<kj::String> scriptName); |
| 518 | TraceItem(rpc::Trace::TraceEventInfo::TraceItem::Reader reader); |
| 519 | TraceItem(TraceItem&&) noexcept = default; |
| 520 | TraceItem& operator=(TraceItem&&) = default; |
| 521 | KJ_DISALLOW_COPY(TraceItem); |
| 522 | |
| 523 | kj::Maybe<kj::String> scriptName; |
| 524 | |
| 525 | void copyTo(rpc::Trace::TraceEventInfo::TraceItem::Builder builder) const; |
| 526 | TraceItem clone() const; |
| 527 | }; |
| 528 | |
| 529 | kj::Vector<TraceItem> traces; |
| 530 | |
| 531 | void copyTo(rpc::Trace::TraceEventInfo::Builder builder) const; |
| 532 | TraceEventInfo clone() const; |
| 533 | }; |
| 534 | |
| 535 | // Describes a hibernatable web socket event |
| 536 | struct HibernatableWebSocketEventInfo final { |
| 537 | struct Message final {}; |
| 538 | struct Close final { |
| 539 | uint16_t code; |
| 540 | bool wasClean; |
| 541 | }; |
| 542 | struct Error final {}; |
| 543 | |
| 544 | using Type = kj::OneOf<Message, Close, Error>; |
| 545 | |
| 546 | explicit HibernatableWebSocketEventInfo(Type type); |
| 547 | HibernatableWebSocketEventInfo(rpc::Trace::HibernatableWebSocketEventInfo::Reader reader); |
| 548 | HibernatableWebSocketEventInfo(HibernatableWebSocketEventInfo&&) noexcept = default; |
| 549 | HibernatableWebSocketEventInfo& operator=(HibernatableWebSocketEventInfo&&) = default; |
| 550 | KJ_DISALLOW_COPY(HibernatableWebSocketEventInfo); |
| 551 | |
| 552 | Type type; |
| 553 | |
| 554 | void copyTo(rpc::Trace::HibernatableWebSocketEventInfo::Builder builder) const; |
| 555 | HibernatableWebSocketEventInfo clone() const; |
| 556 | static Type readFrom(rpc::Trace::HibernatableWebSocketEventInfo::Reader reader); |
| 557 | }; |
| 558 | |
| 559 | // Describes a custom event |
| 560 | struct CustomEventInfo final { |
| 561 | explicit CustomEventInfo() {}; |
| 562 | CustomEventInfo(rpc::Trace::CustomEventInfo::Reader reader) {}; |
| 563 | }; |
| 564 | |
| 565 | // Describes a fetch response |
| 566 | struct FetchResponseInfo final { |
| 567 | explicit FetchResponseInfo(uint16_t statusCode); |
| 568 | FetchResponseInfo(rpc::Trace::FetchResponseInfo::Reader reader); |
| 569 | FetchResponseInfo(FetchResponseInfo&&) noexcept = default; |
| 570 | FetchResponseInfo& operator=(FetchResponseInfo&&) = default; |
| 571 | KJ_DISALLOW_COPY(FetchResponseInfo); |
| 572 | |
| 573 | uint16_t statusCode; |
| 574 | |
| 575 | void copyTo(rpc::Trace::FetchResponseInfo::Builder builder) const; |
| 576 | FetchResponseInfo clone() const; |
| 577 | }; |
| 578 | |
| 579 | // Describes an event published using the node:diagnostics_channel API |
| 580 | struct DiagnosticChannelEvent final { |
| 581 | explicit DiagnosticChannelEvent( |
| 582 | kj::Date timestamp, kj::String channel, kj::Array<kj::byte> message); |
| 583 | DiagnosticChannelEvent(rpc::Trace::DiagnosticChannelEvent::Reader reader); |
| 584 | DiagnosticChannelEvent(DiagnosticChannelEvent&&) noexcept = default; |
| 585 | KJ_DISALLOW_COPY(DiagnosticChannelEvent); |
| 586 | |
| 587 | kj::Date timestamp; |
| 588 | kj::String channel; |
| 589 | kj::Array<kj::byte> message; |
| 590 | |
| 591 | void copyTo(rpc::Trace::DiagnosticChannelEvent::Builder builder) const; |
| 592 | DiagnosticChannelEvent clone() const; |
| 593 | }; |
| 594 | |
| 595 | // Describes a stream diagnostics event. Currently only droppedEvents is supported. |
| 596 | struct StreamDiagnosticsEvent final { |
| 597 | explicit StreamDiagnosticsEvent(uint32_t droppedEventsCount); |
| 598 | StreamDiagnosticsEvent(rpc::Trace::StreamDiagnosticsEvent::Reader reader); |
| 599 | StreamDiagnosticsEvent(StreamDiagnosticsEvent&&) noexcept = default; |
| 600 | KJ_DISALLOW_COPY(StreamDiagnosticsEvent); |
| 601 | |
| 602 | // The count of dropped events for the "droppedEvents" diagnostic. When we support other event |
| 603 | // types, this should be replaced with a kj::OneOf<> of all the different types. |
| 604 | uint32_t droppedEventsCount; |
| 605 | |
| 606 | void copyTo(rpc::Trace::StreamDiagnosticsEvent::Builder builder) const; |
| 607 | StreamDiagnosticsEvent clone() const; |
| 608 | }; |
| 609 | |
| 610 | // Describes a log event |
| 611 | struct Log final { |
| 612 | explicit Log(kj::Date timestamp, LogLevel logLevel, kj::String message); |
| 613 | Log(rpc::Trace::Log::Reader reader); |
| 614 | Log(Log&&) noexcept = default; |
| 615 | KJ_DISALLOW_COPY(Log); |
| 616 | ~Log() noexcept(false) = default; |
| 617 | |
| 618 | // Only as accurate as Worker's Date.now(), for Spectre mitigation. |
| 619 | kj::Date timestamp; |
| 620 | |
| 621 | LogLevel logLevel; |
| 622 | // TODO(soon): Just string for now. Eventually, capture serialized JS objects. |
| 623 | kj::String message; |
| 624 | |
| 625 | void copyTo(rpc::Trace::Log::Builder builder) const; |
| 626 | Log clone() const; |
| 627 | }; |
| 628 | |
| 629 | // Describes an exception event |
| 630 | struct Exception final { |
| 631 | explicit Exception( |
| 632 | kj::Date timestamp, kj::String name, kj::String message, kj::Maybe<kj::String> stack); |
| 633 | Exception(rpc::Trace::Exception::Reader reader); |
| 634 | Exception(Exception&&) noexcept = default; |
| 635 | KJ_DISALLOW_COPY(Exception); |
| 636 | ~Exception() noexcept(false) = default; |
| 637 | |
| 638 | // Only as accurate as Worker's Date.now(), for Spectre mitigation. |
| 639 | kj::Date timestamp; |
| 640 | |
| 641 | kj::String name; |
| 642 | kj::String message; |
| 643 | |
| 644 | kj::Maybe<kj::String> stack; |
| 645 | |
| 646 | void copyTo(rpc::Trace::Exception::Builder builder) const; |
| 647 | Exception clone() const; |
| 648 | }; |
| 649 | |
| 650 | // EventInfo types are used to describe the onset of an invocation. The FetchEventInfo |
| 651 | // can also be used to describe the start of a fetch subrequest. |
| 652 | using EventInfo = kj::OneOf<FetchEventInfo, |
| 653 | JsRpcEventInfo, |
| 654 | ScheduledEventInfo, |
| 655 | AlarmEventInfo, |
| 656 | QueueEventInfo, |
| 657 | EmailEventInfo, |
| 658 | TraceEventInfo, |
| 659 | HibernatableWebSocketEventInfo, |
| 660 | ConnectEventInfo, |
| 661 | CustomEventInfo>; |
| 662 | |
| 663 | EventInfo cloneEventInfo(const EventInfo& info); |
| 664 | |
| 665 | template <typename T> |
| 666 | concept AttributeValue = kj::isSameType<kj::ConstString, T>() || kj::isSameType<bool, T>() || |
| 667 | kj::isSameType<double, T>() || kj::isSameType<int64_t, T>(); |
| 668 | |
| 669 | // An Attribute mark is used to add detail to a span over its lifetime. |
| 670 | // The Attribute struct can also be used to provide arbitrary additional |
| 671 | // properties for some other structs. |
| 672 | // Modeled after https://opentelemetry.io/docs/concepts/signals/traces/#attributes |
| 673 | struct Attribute final { |
| 674 | using Value = kj::OneOf<kj::ConstString, bool, double, int64_t>; |
| 675 | using Values = kj::Array<Value>; |
| 676 | |
| 677 | explicit Attribute(kj::ConstString name, Value&& value); |
| 678 | explicit Attribute(kj::ConstString name, Values&& values); |
| 679 | |
| 680 | template <AttributeValue V> |
| 681 | explicit Attribute(kj::ConstString name, kj::Array<V> vals) |
| 682 | : Attribute(kj::mv(name), KJ_MAP(v, vals) { return Value(kj::mv(v)); }) {} |
| 683 | |
| 684 | template <AttributeValue V> |
| 685 | explicit Attribute(kj::ConstString name, std::initializer_list<V> list) |
| 686 | : Attribute(kj::mv(name), kj::heapArray<V>(list)) {} |
| 687 | |
| 688 | Attribute(rpc::Trace::Attribute::Reader reader); |
| 689 | Attribute(Attribute&&) noexcept = default; |
| 690 | Attribute& operator=(Attribute&&) = default; |
| 691 | KJ_DISALLOW_COPY(Attribute); |
| 692 | |
| 693 | kj::ConstString name; |
| 694 | Values value; |
| 695 | |
| 696 | void copyTo(rpc::Trace::Attribute::Builder builder) const; |
| 697 | Attribute clone() const; |
| 698 | kj::String toString() const; |
| 699 | }; |
| 700 | using CustomInfo = kj::Array<Attribute>; |
| 701 | kj::String KJ_STRINGIFY(const CustomInfo& customInfo); |
| 702 | |
| 703 | struct SpanOpenData { |
| 704 | // Represents the data needed for a SpanOpen event |
| 705 | tracing::SpanId spanId; |
| 706 | tracing::SpanId parentSpanId; |
| 707 | |
| 708 | kj::ConstString operationName; |
| 709 | kj::Date startTime; |
| 710 | |
| 711 | SpanOpenData(rpc::SpanOpenData::Reader reader); |
| 712 | void copyTo(rpc::SpanOpenData::Builder builder) const; |
| 713 | explicit SpanOpenData(tracing::SpanId spanId, |
| 714 | tracing::SpanId parentSpanId, |
| 715 | kj::ConstString operationName, |
| 716 | kj::Date startTime) |
| 717 | : spanId(spanId), |
| 718 | parentSpanId(parentSpanId), |
| 719 | operationName(kj::mv(operationName)), |
| 720 | startTime(startTime) {} |
| 721 | }; |
| 722 | |
| 723 | struct SpanEndData { |
| 724 | // Represents the data needed when closing a span, including the Attributes and SpanClose events. |
| 725 | tracing::SpanId spanId; |
| 726 | |
| 727 | kj::Date endTime; |
| 728 | // Should be Span::TagMap, but we can't forward-declare that. |
| 729 | kj::HashMap<kj::ConstString, tracing::Attribute::Value> tags; |
| 730 | |
| 731 | SpanEndData(rpc::SpanEndData::Reader reader); |
| 732 | void copyTo(rpc::SpanEndData::Builder builder) const; |
| 733 | explicit SpanEndData(tracing::SpanId spanId, |
| 734 | kj::Date endTime, |
| 735 | kj::HashMap<kj::ConstString, tracing::Attribute::Value> tags = |
| 736 | kj::HashMap<kj::ConstString, tracing::Attribute::Value>()) |
| 737 | : spanId(spanId), |
| 738 | endTime(endTime), |
| 739 | tags(kj::mv(tags)) {} |
| 740 | }; |
| 741 | |
| 742 | // A Return mark is used to mark the point at which a span operation returned |
| 743 | // a value. For instance, when a fetch subrequest response is received, or when |
| 744 | // the fetch handler returns a Response. Importantly, it does not signal that the |
| 745 | // span has closed, which may not happen for some period of time after the return |
| 746 | // mark is recorded (e.g. due to things like waitUntils or waiting to fully ready |
| 747 | // the response body payload, etc). |
| 748 | struct Return final { |
| 749 | explicit Return(kj::Maybe<FetchResponseInfo> info = kj::none); |
| 750 | Return(rpc::Trace::Return::Reader reader); |
| 751 | Return(Return&&) noexcept = default; |
| 752 | Return& operator=(Return&&) = default; |
| 753 | KJ_DISALLOW_COPY(Return); |
| 754 | |
| 755 | kj::Maybe<FetchResponseInfo> info = kj::none; |
| 756 | |
| 757 | void copyTo(rpc::Trace::Return::Builder builder) const; |
| 758 | Return clone() const; |
| 759 | }; |
| 760 | |
| 761 | // Mark events no longer have a corresponding type, but the term generally refers to DiagnosticChannelEvent, Exception, Log, Return, and CustomInfo events. |
| 762 | |
| 763 | // Marks the opening of a child span within the streaming tail session. |
| 764 | struct SpanOpen final { |
| 765 | // If the span represents a subrequest, then the info describes the |
| 766 | // details of that subrequest. |
| 767 | using Info = kj::OneOf<FetchEventInfo, JsRpcEventInfo, CustomInfo>; |
| 768 | |
| 769 | explicit SpanOpen(SpanId spanId, kj::ConstString operationName, kj::Maybe<Info> info = kj::none); |
| 770 | SpanOpen(rpc::Trace::SpanOpen::Reader reader); |
| 771 | SpanOpen(SpanOpen&&) noexcept = default; |
| 772 | SpanOpen& operator=(SpanOpen&&) = default; |
| 773 | KJ_DISALLOW_COPY(SpanOpen); |
| 774 | |
| 775 | kj::ConstString operationName; |
| 776 | kj::Maybe<Info> info = kj::none; |
| 777 | SpanId spanId; |
| 778 | |
| 779 | void copyTo(rpc::Trace::SpanOpen::Builder builder) const; |
| 780 | SpanOpen clone() const; |
| 781 | kj::String toString() const; |
| 782 | }; |
| 783 | |
| 784 | // Marks the closing of a child span within the streaming tail session. |
| 785 | // Once emitted, no further mark events should occur within the closed |
| 786 | // span. |
| 787 | struct SpanClose final { |
| 788 | explicit SpanClose(EventOutcome outcome = EventOutcome::OK); |
| 789 | SpanClose(rpc::Trace::SpanClose::Reader reader); |
| 790 | SpanClose(SpanClose&&) noexcept = default; |
| 791 | SpanClose& operator=(SpanClose&&) = default; |
| 792 | KJ_DISALLOW_COPY(SpanClose); |
| 793 | |
| 794 | EventOutcome outcome = EventOutcome::OK; |
| 795 | |
| 796 | void copyTo(rpc::Trace::SpanClose::Builder builder) const; |
| 797 | SpanClose clone() const; |
| 798 | kj::String toString() const; |
| 799 | }; |
| 800 | |
| 801 | // The Onset and Outcome event types are special forms of SpanOpen and |
| 802 | // SpanClose that explicitly mark the start and end of the root span. |
| 803 | // A streaming tail session will always begin with an Onset event, and |
| 804 | // always end with an Outcome event. |
| 805 | struct Onset final { |
| 806 | using Info = EventInfo; |
| 807 | |
| 808 | // Information about the worker that is being tailed. |
| 809 | struct WorkerInfo final { |
| 810 | ExecutionModel executionModel = ExecutionModel::STATELESS; |
| 811 | kj::Maybe<kj::String> scriptName; |
| 812 | kj::Maybe<kj::Own<ScriptVersion::Reader>> scriptVersion; |
| 813 | kj::Maybe<TracePreview> preview; |
| 814 | kj::Maybe<kj::String> dispatchNamespace; |
| 815 | kj::Maybe<kj::String> scriptId; |
| 816 | kj::Maybe<kj::Array<kj::String>> scriptTags; |
| 817 | kj::Maybe<kj::String> entrypoint; |
| 818 | |
| 819 | WorkerInfo clone() const; |
| 820 | }; |
| 821 | |
| 822 | explicit Onset( |
| 823 | tracing::SpanId spanId, Info&& info, WorkerInfo&& workerInfo, CustomInfo attributes); |
| 824 | |
| 825 | Onset(rpc::Trace::Onset::Reader reader); |
| 826 | Onset(Onset&&) noexcept = default; |
| 827 | Onset& operator=(Onset&&) = default; |
| 828 | KJ_DISALLOW_COPY(Onset); |
| 829 | |
| 830 | tracing::SpanId spanId; |
| 831 | Info info; |
| 832 | WorkerInfo workerInfo; |
| 833 | CustomInfo attributes; |
| 834 | |
| 835 | void copyTo(rpc::Trace::Onset::Builder builder) const; |
| 836 | Onset clone() const; |
| 837 | }; |
| 838 | |
| 839 | // Helper functions to copy onset info to/from rpc reader |
| 840 | Onset::Info readOnsetInfo(const rpc::Trace::Onset::Info::Reader& info); |
| 841 | void writeOnsetInfo(const tracing::Onset::Info& info, rpc::Trace::Onset::Info::Builder& builder); |
| 842 | |
| 843 | struct Outcome final { |
| 844 | explicit Outcome(EventOutcome outcome, kj::Duration cpuTime, kj::Duration wallTime); |
| 845 | Outcome(rpc::Trace::Outcome::Reader reader); |
| 846 | Outcome(Outcome&&) noexcept = default; |
| 847 | Outcome& operator=(Outcome&&) = default; |
| 848 | KJ_DISALLOW_COPY(Outcome); |
| 849 | |
| 850 | EventOutcome outcome = EventOutcome::OK; |
| 851 | kj::Duration cpuTime; |
| 852 | kj::Duration wallTime; |
| 853 | |
| 854 | void copyTo(rpc::Trace::Outcome::Builder builder) const; |
| 855 | Outcome clone() const; |
| 856 | kj::String toString() const; |
| 857 | }; |
| 858 | |
| 859 | // A streaming tail worker receives a series of Tail Events. Tail events always |
| 860 | // occur within an InvocationSpanContext. The first TailEvent delivered to a |
| 861 | // streaming tail session is always an Onset. The final TailEvent delivered is |
| 862 | // always an Outcome. Between those can be any number of SpanOpen, SpanClose, |
| 863 | // and Mark events. Every SpanOpen *must* be associated with a SpanClose unless |
| 864 | // the stream was abruptly terminated. |
| 865 | // A future version may add support for Link events again. |
| 866 | struct TailEvent final { |
| 867 | using Event = kj::OneOf<Onset, |
| 868 | Outcome, |
| 869 | SpanOpen, |
| 870 | SpanClose, |
| 871 | DiagnosticChannelEvent, |
| 872 | Exception, |
| 873 | Log, |
| 874 | StreamDiagnosticsEvent, |
| 875 | Return, |
| 876 | CustomInfo>; |
| 877 | |
| 878 | explicit TailEvent(SpanContext context, |
| 879 | TraceId invocationId, |
| 880 | kj::Date timestamp, |
| 881 | kj::uint sequence, |
| 882 | Event&& event); |
| 883 | TailEvent(TraceId traceId, |
| 884 | TraceId invocationId, |
| 885 | kj::Maybe<SpanId> spanId, |
| 886 | kj::Date timestamp, |
| 887 | kj::uint sequence, |
| 888 | Event&& event, |
| 889 | kj::Maybe<TraceFlags> traceFlags = kj::none); |
| 890 | TailEvent(rpc::Trace::TailEvent::Reader reader); |
| 891 | TailEvent(TailEvent&&) = default; |
| 892 | TailEvent& operator=(TailEvent&&) = default; |
| 893 | KJ_DISALLOW_COPY(TailEvent); |
| 894 | |
| 895 | // The span context this event is associated with. |
| 896 | SpanContext spanContext; |
| 897 | TraceId invocationId; |
| 898 | |
| 899 | kj::Date timestamp; // Unix epoch, Spectre-mitigated resolution |
| 900 | kj::uint sequence; |
| 901 | |
| 902 | Event event; |
| 903 | |
| 904 | void copyTo(rpc::Trace::TailEvent::Builder builder) const; |
| 905 | TailEvent clone() const; |
| 906 | }; |
| 907 | |
| 908 | kj::String KJ_STRINGIFY(const tracing::TailEvent::Event& event); |
| 909 | |
| 910 | } // namespace tracing |
| 911 | |
| 912 | enum class PipelineLogLevel { |
| 913 | // WARNING: This must be kept in sync with PipelineDef::LogLevel (which is not in the OSS |
| 914 | // release). |
| 915 | NONE, |
| 916 | FULL |
| 917 | }; |
| 918 | |
| 919 | // TODO(someday): See if we can merge similar code concepts... Trace fills a role similar to |
| 920 | // MetricsCollector::Reporter::StageEvent, and Tracer fills a role similar to |
| 921 | // MetricsCollector::Request. Currently, the major differences are: |
| 922 | // |
| 923 | // - MetricsCollector::Request uses its destructor to measure a IoContext's wall time, so |
| 924 | // it needs to live exactly as long as its IoContext. Tracer currently needs to live as |
| 925 | // long as both the IoContext and those of any subrequests. |
| 926 | // - Due to the difference in lifetimes, results of each become available in a different order, |
| 927 | // and intermediate values can be freed at different times. |
| 928 | // - Request builds a vector of results, while Tracer builds a tree. |
| 929 | |
| 930 | // TODO(cleanup) - worth separating into immutable Trace vs. mutable TraceBuilder? |
| 931 | |
| 932 | // Collects trace information about the handling of a worker/pipeline fetch event. |
| 933 | class Trace final: public kj::Refcounted { |
| 934 | public: |
| 935 | explicit Trace(kj::Maybe<kj::String> stableId, |
| 936 | kj::Maybe<kj::String> scriptName, |
| 937 | kj::Maybe<kj::Own<ScriptVersion::Reader>> scriptVersion, |
| 938 | kj::Maybe<kj::String> dispatchNamespace, |
| 939 | kj::Maybe<kj::String> scriptId, |
| 940 | kj::Array<kj::String> scriptTags, |
| 941 | kj::Maybe<kj::String> entrypoint, |
| 942 | ExecutionModel executionModel, |
| 943 | kj::Maybe<kj::String> durableObjectId = kj::none, |
| 944 | kj::Maybe<tracing::TracePreview> preview = kj::none); |
| 945 | Trace(rpc::Trace::Reader reader); |
| 946 | ~Trace() noexcept(false); |
| 947 | KJ_DISALLOW_COPY_AND_MOVE(Trace); |
| 948 | |
| 949 | // Empty for toplevel worker. |
| 950 | kj::Maybe<kj::String> stableId; |
| 951 | |
| 952 | // We treat the origin value as "unset". |
| 953 | kj::Date eventTimestamp = kj::UNIX_EPOCH; |
| 954 | |
| 955 | kj::Maybe<tracing::EventInfo> eventInfo; |
| 956 | |
| 957 | // TODO(someday): Work out what sort of information we may want to convey about the parent |
| 958 | // trace, if any. |
| 959 | |
| 960 | kj::Maybe<kj::String> scriptName; |
| 961 | kj::Maybe<kj::Own<ScriptVersion::Reader>> scriptVersion; |
| 962 | kj::Maybe<kj::String> dispatchNamespace; |
| 963 | kj::Maybe<kj::String> scriptId; |
| 964 | kj::Array<kj::String> scriptTags; |
| 965 | kj::Maybe<kj::Array<tracing::Attribute>> tailAttributes; |
| 966 | kj::Maybe<kj::String> entrypoint; |
| 967 | kj::Maybe<tracing::TracePreview> preview; |
| 968 | kj::Maybe<kj::String> durableObjectId; |
| 969 | |
| 970 | kj::Vector<tracing::Log> logs; |
| 971 | // A request's trace can have multiple exceptions due to separate request/waitUntil tasks. |
| 972 | kj::Vector<tracing::Exception> exceptions; |
| 973 | |
| 974 | kj::Vector<tracing::DiagnosticChannelEvent> diagnosticChannelEvents; |
| 975 | |
| 976 | ExecutionModel executionModel; |
| 977 | EventOutcome outcome = EventOutcome::UNKNOWN; |
| 978 | |
| 979 | kj::Maybe<tracing::FetchResponseInfo> fetchResponseInfo; |
| 980 | |
| 981 | kj::Duration cpuTime; |
| 982 | kj::Duration wallTime; |
| 983 | |
| 984 | bool truncated = false; |
| 985 | bool exceededLogLimit = false; |
| 986 | bool exceededExceptionLimit = false; |
| 987 | bool exceededDiagnosticChannelEventLimit = false; |
| 988 | // Trace data is recorded outside of the JS heap. To avoid DoS, we keep an estimate of trace |
| 989 | // data size, and we stop recording if too much is used. |
| 990 | size_t bytesUsed = 0; |
| 991 | |
| 992 | // Copy content from this trace into `builder`. |
| 993 | void copyTo(rpc::Trace::Builder builder) const; |
| 994 | |
| 995 | // Adds all content from `reader` to this `Trace`. (Typically this trace is empty before the |
| 996 | // call.) Also applies filtering to the trace as if it were recorded with the given |
| 997 | // pipelineLogLevel. |
| 998 | void mergeFrom(rpc::Trace::Reader reader, PipelineLogLevel pipelineLogLevel); |
| 999 | }; |
| 1000 | |
| 1001 | // ======================================================================================= |
| 1002 | |
| 1003 | // Helper function used when setting "truncated_script_id" tags. Truncates the scriptId to 10 |
| 1004 | // characters. |
| 1005 | inline kj::String truncateScriptId(kj::StringPtr id) { |
| 1006 | auto truncatedId = id.first(kj::min(id.size(), 10)); |
| 1007 | return kj::str(truncatedId); |
| 1008 | } |
| 1009 | |
| 1010 | // ======================================================================================= |
| 1011 | // Span tracing |
| 1012 | // |
| 1013 | // TODO(cleanup): (Streaming) tail workers now have access to span tracing as well, but with that |
| 1014 | // the trace worker and span interfaces are still mostly independent of each other; separate span |
| 1015 | // tracing into a separate header. |
| 1016 | |
| 1017 | class SpanBuilder; |
| 1018 | class SpanObserver; |
| 1019 | |
| 1020 | struct Span { |
| 1021 | // Represents a trace span. `Span` objects are delivered to `SpanObserver`s for recording. To |
| 1022 | // create a `Span`, use a `SpanBuilder`. |
| 1023 | |
| 1024 | public: |
| 1025 | using TagValue = tracing::Attribute::Value; |
| 1026 | // TODO(someday): Support binary bytes, too. |
| 1027 | using TagMap = kj::HashMap<kj::ConstString, TagValue>; |
| 1028 | using Tag = TagMap::Entry; |
| 1029 | |
| 1030 | struct Log { |
| 1031 | kj::Date timestamp; |
| 1032 | Tag tag; |
| 1033 | }; |
| 1034 | |
| 1035 | kj::ConstString operationName; |
| 1036 | kj::Date startTime; |
| 1037 | kj::Date endTime; |
| 1038 | TagMap tags; |
| 1039 | kj::Vector<Log> logs; |
| 1040 | |
| 1041 | // We set an arbitrary (-ish) cap on log messages for safety. If we drop logs because of this, |
| 1042 | // we report how many in a final "dropped_logs" log. |
| 1043 | // |
| 1044 | // At the risk of being too clever, I chose a limit that is one below a power of two so that |
| 1045 | // we'll typically have space for one last element available for the "dropped_logs" log without |
| 1046 | // needing to grow the vector. |
| 1047 | static constexpr auto MAX_LOGS = 1023; |
| 1048 | uint droppedLogs = 0; |
| 1049 | |
| 1050 | explicit Span(kj::ConstString operationName, kj::Date startTime) |
| 1051 | : operationName(kj::mv(operationName)), |
| 1052 | startTime(startTime), |
| 1053 | endTime(startTime) {} |
| 1054 | }; |
| 1055 | |
| 1056 | // Utility functions for handling span tags. |
| 1057 | void serializeTagValue(rpc::TagValue::Builder builder, const Span::TagValue& value); |
| 1058 | Span::TagValue deserializeTagValue(rpc::TagValue::Reader value); |
| 1059 | |
| 1060 | // Clone function for span tags, avoids memory allocation for string literals and non-string values. |
| 1061 | Span::TagValue spanTagClone(const Span::TagValue& tag); |
| 1062 | |
| 1063 | // An opaque token which can be used to create child spans of some parent. This is typically |
| 1064 | // passed down from a caller to a callee when the caller wants to allow the callee to create |
| 1065 | // spans for itself that show up as children of the caller's span, but the caller does not |
| 1066 | // want to give the callee any other ability to modify the parent span. |
| 1067 | class SpanParent { |
| 1068 | public: |
| 1069 | SpanParent(SpanBuilder& builder); |
| 1070 | |
| 1071 | // Make a SpanParent that causes children not to be reported anywhere. |
| 1072 | SpanParent(decltype(nullptr)) {} |
| 1073 | |
| 1074 | SpanParent(kj::Maybe<kj::Own<SpanObserver>> observer): observer(kj::mv(observer)) {} |
| 1075 | |
| 1076 | SpanParent(SpanParent&& other) = default; |
| 1077 | SpanParent& operator=(SpanParent&& other) = default; |
| 1078 | KJ_DISALLOW_COPY(SpanParent); |
| 1079 | |
| 1080 | SpanParent addRef(); |
| 1081 | |
| 1082 | // Create a new child span. |
| 1083 | // |
| 1084 | // `operationName` should be a string literal with infinite lifetime. |
| 1085 | [[nodiscard]] SpanBuilder newChild( |
| 1086 | kj::ConstString operationName, kj::Maybe<kj::Date> startTime = kj::none); |
| 1087 | |
| 1088 | // Useful to skip unnecessary code when not observed. |
| 1089 | bool isObserved() { |
| 1090 | return observer != kj::none; |
| 1091 | } |
| 1092 | |
| 1093 | // Get the underlying SpanObserver representing the parent span. |
| 1094 | // |
| 1095 | // This is needed in particular when making outbound network requests that must be annotated with |
| 1096 | // trace IDs in a way that is specific to the trace back-end being used. The caller must downcast |
| 1097 | // the `SpanObserver` to the expected observer type in order to extract the trace ID. |
| 1098 | kj::Maybe<SpanObserver&> getObserver() { |
| 1099 | return observer; |
| 1100 | } |
| 1101 | |
| 1102 | // Return the serializable identity of this span for cross-boundary propagation. |
| 1103 | kj::Maybe<tracing::SpanContext> toSpanContext(); |
| 1104 | |
| 1105 | // Returns the observer's spanId, or SpanId::nullId if there is none. |
| 1106 | tracing::SpanId getSpanId(); |
| 1107 | |
| 1108 | private: |
| 1109 | kj::Maybe<kj::Own<SpanObserver>> observer; |
| 1110 | }; |
| 1111 | |
| 1112 | // Interface for writing a span. Essentially, this is a mutable interface to a `Span` object, |
| 1113 | // given only to the code which is meant to create the span, whereas code that merely collects |
| 1114 | // and reports spans gets the `Span` type. |
| 1115 | // |
| 1116 | // The reason we use a separate builder type rather than rely on constness is so that the methods |
| 1117 | // can be no-ops when there is no observer, avoiding unnecessary allocations. To allow for this, |
| 1118 | // SpanBuilder is designed to be write-only -- you cannot read back the content. Only the |
| 1119 | // observer (if there is one) receives the content. |
| 1120 | class SpanBuilder { |
| 1121 | public: |
| 1122 | // Create a new top-level span that will report to the given observer. If the observer is null, |
| 1123 | // no data is collected. |
| 1124 | // |
| 1125 | // `operationName` should be a string literal with infinite lifetime, or somehow otherwise be |
| 1126 | // attached to the observer observing this span. |
| 1127 | explicit SpanBuilder(kj::Maybe<kj::Own<SpanObserver>> observer, |
| 1128 | kj::ConstString operationName, |
| 1129 | kj::Maybe<kj::Date> startTime = kj::none); |
| 1130 | |
| 1131 | // Make a SpanBuilder that ignores all calls. (Useful if you want to assign it later.) |
| 1132 | SpanBuilder(decltype(nullptr)) {} |
| 1133 | |
| 1134 | SpanBuilder(SpanBuilder&& other) = default; |
| 1135 | SpanBuilder& operator=(SpanBuilder&& other); // ends the existing span and starts a new one |
| 1136 | KJ_DISALLOW_COPY(SpanBuilder); |
| 1137 | |
| 1138 | ~SpanBuilder() noexcept(false); |
| 1139 | |
| 1140 | // Finishes and submits the span. This is done implicitly by the destructor, but sometimes it's |
| 1141 | // useful to be able to submit early. The SpanBuilder ignores all further method calls after this |
| 1142 | // is invoked. |
| 1143 | void end(); |
| 1144 | |
| 1145 | // Useful to skip unnecessary code when not observed. |
| 1146 | bool isObserved() { |
| 1147 | return observer != kj::none; |
| 1148 | } |
| 1149 | |
| 1150 | // Get the underlying SpanObserver representing the span. |
| 1151 | // |
| 1152 | // This is needed in particular when making outbound network requests that must be annotated with |
| 1153 | // trace IDs in a way that is specific to the trace back-end being used. The caller must downcast |
| 1154 | // the `SpanObserver` to the expected observer type in order to extract the trace ID. |
| 1155 | kj::Maybe<SpanObserver&> getObserver() { |
| 1156 | return observer; |
| 1157 | } |
| 1158 | |
| 1159 | // Create a new child span. |
| 1160 | // |
| 1161 | // `operationName` should be a string literal with infinite lifetime. |
| 1162 | [[nodiscard]] SpanBuilder newChild( |
| 1163 | kj::ConstString operationName, kj::Maybe<kj::Date> startTime = kj::none); |
| 1164 | |
| 1165 | // Change the operation name from what was specified at span creation. |
| 1166 | // |
| 1167 | // `operationName` should be a string literal with infinite lifetime. |
| 1168 | void setOperationName(kj::ConstString operationName); |
| 1169 | |
| 1170 | using TagValue = Span::TagValue; |
| 1171 | // `key` must point to memory that will remain valid all the way until this span's data is |
| 1172 | // serialized. |
| 1173 | // Allow setting tags with an extended set of types to elide string allocations when we have a |
| 1174 | // string literal or are not being observed. We include String/LiteralStringConst here to avoid |
| 1175 | // having to manually cast them to ConstString each time. |
| 1176 | using TagInitValue = kj::OneOf<kj::StringPtr, |
| 1177 | kj::String, |
| 1178 | kj::LiteralStringConst, |
| 1179 | kj::ConstString, |
| 1180 | bool, |
| 1181 | double, |
| 1182 | int64_t>; |
| 1183 | |
| 1184 | void setTag(kj::ConstString key, TagInitValue value); |
| 1185 | |
| 1186 | // `key` must point to memory that will remain valid all the way until this span's data is |
| 1187 | // serialized. |
| 1188 | // |
| 1189 | // The differences between this and `setTag()` is that logs are timestamped and may have |
| 1190 | // duplicate keys. |
| 1191 | void addLog(kj::Date timestamp, kj::ConstString key, TagValue value); |
| 1192 | |
| 1193 | private: |
| 1194 | kj::Maybe<kj::Own<SpanObserver>> observer; |
| 1195 | // The under-construction span, or null if the span has ended. |
| 1196 | kj::Maybe<Span> span; |
| 1197 | |
| 1198 | friend class SpanParent; |
| 1199 | }; |
| 1200 | |
| 1201 | // Abstract interface for observing trace spans reported by the runtime. Different |
| 1202 | // implementations might support different tracing back-ends, e.g. Trace Workers, Jaeger, or |
| 1203 | // whatever infrastructure you prefer to use for this. |
| 1204 | // |
| 1205 | // A new SpanObserver is created at the start of each Span. The SpanBuilder drives the observer |
| 1206 | // through its lifecycle: onOpen() is called when the span is created, onClose() when the span |
| 1207 | // ends, and onUpdateName() if the operation name changes between open and close. |
| 1208 | class SpanObserver: public kj::Refcounted { |
| 1209 | public: |
| 1210 | // Allocate a new child span. |
| 1211 | // |
| 1212 | // Note that children can be created long after a span has completed. |
| 1213 | [[nodiscard]] virtual kj::Own<SpanObserver> newChild() = 0; |
| 1214 | |
| 1215 | // Allocate a child for a span initiated directly by user JavaScript (via |
| 1216 | // `ctx.tracing.enterSpan`). Allows implementations to apply different policies than for |
| 1217 | // runtime-issued spans (notably, edgeworker bypasses its operation-name allowlist here). |
| 1218 | [[nodiscard]] virtual kj::Own<SpanObserver> newChildFromUserCode() { |
| 1219 | return newChild(); |
| 1220 | } |
| 1221 | |
| 1222 | // Called when the span is opened. Delivers the initial operation name and start time. |
| 1223 | // Called exactly once, before any other lifecycle method. |
| 1224 | virtual void onOpen(kj::ConstString operationName, kj::Date startTime) = 0; |
| 1225 | |
| 1226 | // Called when the span is closed. Delivers the end time, tags, and logs. |
| 1227 | // Called exactly once per observer, after onOpen(). Tags and logs are moved from the span; |
| 1228 | // the observer takes ownership. |
| 1229 | virtual void onClose(kj::Date endTime, Span::TagMap&& tags, kj::Vector<Span::Log>&& logs) = 0; |
| 1230 | |
| 1231 | // Called when the operation name is changed after the span was opened (via |
| 1232 | // SpanBuilder::setOperationName()). Observers that eagerly stream the open event should handle |
| 1233 | // this; others may simply update their buffered state. Default implementation is a no-op. |
| 1234 | virtual void onUpdateName(kj::ConstString operationName) {} |
| 1235 | |
| 1236 | // The current time to be provided for the span. For user tracing, we will override this to |
| 1237 | // provide I/O time. This *requires* that spans are only created when an IOContext is available |
| 1238 | // (usually it is difficult to violate this assumption, but care must be taken that the observer |
| 1239 | // isn't used directly to create a span before the IoContext has been constructed (previously this |
| 1240 | // was a case with a top-level span owned by the WorkerTracer itself). |
| 1241 | virtual kj::Date getTime() { |
| 1242 | return kj::systemPreciseCalendarClock().now(); |
| 1243 | } |
| 1244 | |
| 1245 | // Return the serializable identity of this span for cross-boundary propagation. |
| 1246 | // Returns kj::none if this observer doesn't carry identity. |
| 1247 | virtual kj::Maybe<tracing::SpanContext> toSpanContext() { |
| 1248 | return kj::none; |
| 1249 | } |
| 1250 | |
| 1251 | // Returns this observer's spanId, or SpanId::nullId if it has no identity. |
| 1252 | virtual tracing::SpanId getSpanId() { |
| 1253 | return tracing::SpanId::nullId; |
| 1254 | } |
| 1255 | }; |
| 1256 | |
| 1257 | inline kj::Maybe<tracing::SpanContext> SpanParent::toSpanContext() { |
| 1258 | KJ_IF_SOME(obs, observer) { |
| 1259 | return obs->toSpanContext(); |
| 1260 | } |
| 1261 | return kj::none; |
| 1262 | } |
| 1263 | |
| 1264 | inline tracing::SpanId SpanParent::getSpanId() { |
| 1265 | KJ_IF_SOME(obs, observer) { |
| 1266 | return obs->getSpanId(); |
| 1267 | } |
| 1268 | return tracing::SpanId::nullId; |
| 1269 | } |
| 1270 | |
| 1271 | inline SpanParent::SpanParent(SpanBuilder& builder): observer(mapAddRef(builder.observer)) {} |
| 1272 | |
| 1273 | inline SpanParent SpanParent::addRef() { |
| 1274 | return SpanParent(mapAddRef(observer)); |
| 1275 | } |
| 1276 | |
| 1277 | inline SpanBuilder SpanParent::newChild( |
| 1278 | kj::ConstString operationName, kj::Maybe<kj::Date> startTime) { |
| 1279 | return SpanBuilder(observer.map([](kj::Own<SpanObserver>& obs) { return obs->newChild(); }), |
| 1280 | kj::mv(operationName), startTime); |
| 1281 | } |
| 1282 | |
| 1283 | inline SpanBuilder SpanBuilder::newChild( |
| 1284 | kj::ConstString operationName, kj::Maybe<kj::Date> startTime) { |
| 1285 | return SpanBuilder(observer.map([](kj::Own<SpanObserver>& obs) { return obs->newChild(); }), |
| 1286 | kj::mv(operationName), startTime); |
| 1287 | } |
| 1288 | |
| 1289 | // TraceContext to keep track of user tracing/existing tracing better |
| 1290 | class TraceContext { |
| 1291 | public: |
| 1292 | TraceContext(): span(nullptr), userSpan(nullptr) {} |
| 1293 | TraceContext(SpanBuilder span, SpanBuilder userSpan) |
| 1294 | : span(kj::mv(span)), |
| 1295 | userSpan(kj::mv(userSpan)) {} |
| 1296 | TraceContext(TraceContext&& other) = default; |
| 1297 | TraceContext& operator=(TraceContext&& other) = default; |
| 1298 | KJ_DISALLOW_COPY(TraceContext); |
| 1299 | |
| 1300 | // Set a tag on both the internal span and user span. |
| 1301 | void setTag(kj::ConstString key, SpanBuilder::TagInitValue value); |
| 1302 | bool isObserved() { |
| 1303 | return span.isObserved() || userSpan.isObserved(); |
| 1304 | } |
| 1305 | SpanParent getInternalSpanParent() { |
| 1306 | return SpanParent(span); |
| 1307 | } |
| 1308 | |
| 1309 | SpanParent getUserSpanParent() { |
| 1310 | return SpanParent(userSpan); |
| 1311 | } |
| 1312 | |
| 1313 | private: |
| 1314 | SpanBuilder span; |
| 1315 | SpanBuilder userSpan; |
| 1316 | }; |
| 1317 | |
| 1318 | // RAII object that measures the time duration over its lifetime. It tags this duration onto a |
| 1319 | // given request span using a specified tag name. Ideal for automatically tracking and logging |
| 1320 | // execution times within a scoped block. |
| 1321 | class ScopedDurationTagger { |
| 1322 | public: |
| 1323 | explicit ScopedDurationTagger( |
| 1324 | SpanBuilder& span, kj::ConstString key, const kj::MonotonicClock& timer); |
| 1325 | ~ScopedDurationTagger() noexcept(false); |
| 1326 | KJ_DISALLOW_COPY_AND_MOVE(ScopedDurationTagger); |
| 1327 | |
| 1328 | private: |
| 1329 | SpanBuilder& span; |
| 1330 | kj::ConstString key; |
| 1331 | const kj::MonotonicClock& timer; |
| 1332 | const kj::TimePoint startTime; |
| 1333 | }; |
| 1334 | |
| 1335 | } // namespace workerd |