Module: Bitfab::Replay

Defined in:
lib/bitfab/replay.rb

Overview

Replay historical traces through a traced method and create a test run.

Constant Summary collapse

CODE_CHANGE_UNSET = 
Object.new.freeze
TRUNK_CANDIDATES =

Read the code-change payload the replay wrapper captured and pointed us at via BITFAB_CODE_CHANGE_PATH. Returns [description, files] or nil. Best-effort: any error (no env var, missing file, bad JSON) yields nil so the replay proceeds with no code change rather than raising.

["origin/HEAD", "origin/main", "origin/master", "main", "master"].freeze
CC_MAX_FILES = 
60
CC_MAX_FILE_BYTES = 
500_000
CC_MAX_TOTAL_BYTES = 
2_000_000

Class Method Summary collapse

Class Method Details

.build_mock_tree(root) ⇒ Object

Walk the children of a root span tree node depth-first and build a lookup keyed by "#trace_function_key:#span_name:#call_index".

The root node itself is excluded: at replay time the runtime root span never queries the mock tree.

The compound (key, name) match disambiguates same-key spans that come from the fluent client.get_function(key).wrap(...) pattern: every wrapped method shares trace_function_key but differs in span_name. The counter is per-(key, name) pair so repeated same-name calls (including recursion) still order by occurrence. Mirrors the Python and TypeScript SDKs after HVT-2078: keying by trace_function_key alone caused the wrong historical output for fluent-API span sets.



798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
 
# File 'lib/bitfab/replay.rb', line 798

def build_mock_tree(root)
  spans = {}
  counters = {}

  walk = lambda do |node|
    key = node["traceFunctionKey"]
    if key && !key.empty?
      name = node["spanName"]
      name = key if name.nil? || name.empty?
      counter_key = "#{key}:#{name}"
      index = counters[counter_key] || 0
      counters[counter_key] = index + 1
      # externalSpanId is what the lazy path fetches the recorded output by
      # when the tree came back payload-free (includeOutputs=false). output
      # / outputMeta are copied ONLY when the node actually carries them, so
      # the presence of the :output key distinguishes an inline nil output
      # from a payload-free entry (mirrors TS's undefined-vs-null check).
      entry = {
        source_span_id: node["sourceSpanId"],
        external_span_id: node["externalSpanId"]
      }
      entry[:output] = node["output"] if node.key?("output")
      entry[:output_meta] = node["outputMeta"] if node.key?("outputMeta")
      spans["#{counter_key}:#{index}"] = entry
    end
    (node["children"] || []).each { |child| walk.call(child) }
  end

  (root["children"] || []).each { |child| walk.call(child) }

  spans
end

.build_span_output_fetcher(http_client) ⇒ Object

Build a memoized lazy fetcher for a single replay item. The returned lambda takes an externalSpanId and returns that span's deserialized recorded output, fetching it via get_external_span at most once per id (a span read twice - as a base "marked" mock and via an override's get_original_output - fetches once). The cache is per item: each item replays synchronously on its own thread, so no lock is needed.



837
838
839
840
841
842
843
844
845
846
847
848
849
850
 
# File 'lib/bitfab/replay.rb', line 837

def build_span_output_fetcher(http_client)
  cache = {}
  lambda do |external_span_id|
    return cache[external_span_id] if cache.key?(external_span_id)

    span = http_client.get_external_span(external_span_id)
    span_data = (span["rawData"] || {})["span_data"] || {}
    # Prefer the Ruby Marshal payload (output_serialized) written by this
    # SDK; fall back to another SDK's output_meta, then the raw JSON output.
    meta = span_data["output_serialized"]
    meta = span_data["output_meta"] if meta.nil? || meta == ""
    cache[external_span_id] = Serialize.deserialize_output(span_data["output"], meta)
  end
end

.capture_code_change_from_git(cwd, label = nil) ⇒ Object 



460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
 
# File 'lib/bitfab/replay.rb', line 460

def capture_code_change_from_git(cwd, label = nil)
  root = git(cwd, ["rev-parse", "--show-toplevel"])
  return nil if root.nil? || root.strip.empty?

  root = root.strip
  resolved = cc_resolve_base(root)
  return nil if resolved.nil?

  base, from_trunk = resolved

  tracked = git(root, ["diff", "--name-status", "--no-renames", "-z", base, "--", ":!.bitfab"]) || ""
  untracked = git(root, ["ls-files", "--others", "--exclude-standard", "-z", "--", ":!.bitfab"]) || ""

  entries = cc_parse_name_status_z(tracked)
  untracked.split("\0").reject(&:empty?).each { |p| entries << ["A", p] }
  return nil if entries.empty?

  files = []
  total = 0
  entries.each do |status, path|
    break if files.length >= CC_MAX_FILES

    # Skip oversized files by size BEFORE reading their full contents, so a
    # huge changed file can't OOM/stall replay just to be discarded.
    before_bytes = (status == "A") ? 0 : cc_blob_bytes(root, base, path)
    after_bytes = (status == "D") ? 0 : cc_working_bytes(root, path)
    next if before_bytes > CC_MAX_FILE_BYTES || after_bytes > CC_MAX_FILE_BYTES

    # Normalize CRLF -> LF on both sides: git's blob is already LF-normalized
    # (autocrlf clean), so a raw CRLF working file would otherwise skew every
    # line as changed. Comparing/storing LF keeps the diff meaningful.
    before = (status == "A") ? "" : (git(root, ["show", "#{base}:#{path}"]) || "").gsub("\r\n", "\n")
    after = (status == "D") ? "" : cc_read_working_file(root, path).gsub("\r\n", "\n")
    next if before == after

    size = before.bytesize + after.bytesize
    next if total + size > CC_MAX_TOTAL_BYTES ||
      cc_looks_binary?(before) || cc_looks_binary?(after)

    total += size
    files << {"path" => path, "before" => before, "after" => after}
  end
  return nil if files.empty?

  subject = git(root, ["log", "-1", "--format=%s", "HEAD"])
  subject = subject ? subject.strip : ""
  head = ((label && !label.strip.empty?) ? label.strip : nil) || (subject.empty? ? nil : subject) || "Working-tree change"
  word = (files.length == 1) ? "file" : "files"
  against = from_trunk ? "vs trunk" : "uncommitted (vs HEAD)"
  ["#{head} (#{files.length} #{word} changed #{against})", files]
rescue
  nil
end

.cc_blob_bytes(root, base, path) ⇒ Object 



447
448
449
450
451
452
 
# File 'lib/bitfab/replay.rb', line 447

def cc_blob_bytes(root, base, path)
  out = git(root, ["cat-file", "-s", "#{base}:#{path}"])
  (out && !out.strip.empty?) ? out.strip.to_i : 0
rescue
  0
end

.cc_looks_binary?(str) ⇒ Boolean

Returns:

  • (Boolean) 


531
532
533
 
# File 'lib/bitfab/replay.rb', line 531

def cc_looks_binary?(str)
  str[0, 8000].include?("\0")
end

.cc_parse_name_status_z(raw) ⇒ Object 



514
515
516
517
518
519
520
521
522
523
 
# File 'lib/bitfab/replay.rb', line 514

def cc_parse_name_status_z(raw)
  parts = raw.split("\0").reject(&:empty?)
  out = []
  i = 0
  while i + 1 < parts.length
    out << [parts[i][0], parts[i + 1]]
    i += 2
  end
  out
end

.cc_read_working_file(root, path) ⇒ Object 



525
526
527
528
529
 
# File 'lib/bitfab/replay.rb', line 525

def cc_read_working_file(root, path)
  File.read(File.join(root, path), encoding: "UTF-8")
rescue
  ""
end

.cc_ref_exists?(root, ref) ⇒ Boolean

Returns:

  • (Boolean) 


424
425
426
 
# File 'lib/bitfab/replay.rb', line 424

def cc_ref_exists?(root, ref)
  !git(root, ["rev-parse", "--verify", "#{ref}^{object}"]).nil?
end

.cc_resolve_base(root) ⇒ Object

Returns [base, from_trunk]; from_trunk is false on the HEAD fallback.



429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
 
# File 'lib/bitfab/replay.rb', line 429

def cc_resolve_base(root)
  forced = ENV["BITFAB_CODE_CHANGE_BASE"]
  if forced && cc_ref_exists?(root, forced)
    mb = git(root, ["merge-base", "HEAD", forced])
    return [mb.strip, true] if mb && !mb.strip.empty?

    rp = git(root, ["rev-parse", "--verify", forced])
    return (rp && !rp.strip.empty?) ? [rp.strip, true] : nil
  end
  TRUNK_CANDIDATES.each do |candidate|
    next unless cc_ref_exists?(root, candidate)

    mb = git(root, ["merge-base", "HEAD", candidate])
    return [mb.strip, true] if mb && !mb.strip.empty?
  end
  cc_ref_exists?(root, "HEAD") ? ["HEAD", false] : nil
end

.cc_working_bytes(root, path) ⇒ Object 



454
455
456
457
458
 
# File 'lib/bitfab/replay.rb', line 454

def cc_working_bytes(root, path)
  File.size(File.join(root, path))
rescue
  0
end

.db_branch_enabled?(db_branch) ⇒ Boolean

Whether the caller asked for database branching. true and a Hash both turn it on; false and nil leave it off. Kept separate from db_branch_settings because db_branch: true enables branching while contributing no settings, so the presence of settings can't stand in for the switch.

Returns:

  • (Boolean) 


343
344
345
 
# File 'lib/bitfab/replay.rb', line 343

def db_branch_enabled?(db_branch)
  !db_branch.nil? && db_branch != false
end

.db_branch_settings(db_branch) ⇒ Object

Convert the caller's db_branch options to the wire shape, or nil when nothing was set. Dropping the key entirely keeps the request identical to what older SDKs send. Booleans carry no settings: they only move the switch.

Keys: :min_cu and :max_cu are the branch compute's autoscaling floor and ceiling in Neon Compute Units (equal values pin the size, keeping items comparable); :warmup_sql is appended to the branch's readiness check so it warms the cache before the replayed function sees the lease.



356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
 
# File 'lib/bitfab/replay.rb', line 356

def db_branch_settings(db_branch)
  return nil if db_branch.nil? || db_branch == true || db_branch == false || db_branch.empty?

  # Symbol or string keys, matching normalize_code_change_files: a hash
  # assembled from JSON or YAML would otherwise be read as empty, leaving
  # the branch on the mirror's defaults with no error to notice.
  min_cu = db_branch[:min_cu] || db_branch["min_cu"]
  max_cu = db_branch[:max_cu] || db_branch["max_cu"]
  warmup_sql = db_branch[:warmup_sql] || db_branch["warmup_sql"]

  settings = {}
  settings["minCu"] = min_cu unless min_cu.nil?
  settings["maxCu"] = max_cu unless max_cu.nil?
  settings["warmupSql"] = warmup_sql unless warmup_sql.nil?
  settings.empty? ? nil : settings
end

.execute_item(item, receiver, method_name, test_run_id, input_source_span_id = nil, metrics = {}, input_source_trace_id: nil, mock_strategy: "marked", mock_tree: nil, mock_overrides: nil, fetch_span_output: nil, adapt_inputs: nil, adapt_ctx: nil, db_branch_lease: nil, source_bitfab_trace_id: nil, db_snapshot_ref: nil) ⇒ Object

Execute a single replay item: deserialize inputs, call method with replay context.



894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
 
# File 'lib/bitfab/replay.rb', line 894

def execute_item(item, receiver, method_name, test_run_id, input_source_span_id = nil, metrics = {},
  input_source_trace_id: nil, mock_strategy: "marked", mock_tree: nil, mock_overrides: nil,
  fetch_span_output: nil, adapt_inputs: nil, adapt_ctx: nil, db_branch_lease: nil, source_bitfab_trace_id: nil,
  db_snapshot_ref: nil)
  args, kwargs = Serialize.deserialize_inputs(item)

  fn_result = nil
  fn_error = nil
  # Client-side correlation id that tags this item's replay spans so the
  # server can echo back the row id it minted (resolved in run()'s
  # complete-replay loop). Carried on the item under :_sdk_trace_id, never
  # surfaced as the public :trace_id.
  sdk_trace_id = SecureRandom.uuid
  # Collects the root span's persistence threads (span uploads + trace
  # completion). Joined below so this item's trace is on the server
  # before run() calls complete_replay: otherwise the server's trace-ID
  # mapping races the uploads and the item's trace_id comes back nil.
  pending_persistence = []

  ReplayContext.with_context(
    test_run_id:,
    input_source_span_id:,
    input_source_trace_id:,
    trace_id: sdk_trace_id,
    mock_tree:,
    mock_strategy:,
    mock_overrides:,
    fetch_span_output:,
    pending_persistence:,
    db_branch_lease:,
    source_bitfab_trace_id:
  ) do
    # Reshape recorded inputs onto the current signature when an adapter is
    # supplied. Inside the rescue so a raising adapter surfaces on this
    # item's :error instead of crashing the run; args is reported on :input.
    if adapt_inputs
      ctx = adapt_ctx || {
        original_trace_id: nil,
        original_span_id: input_source_span_id,
        # Deprecated aliases for original_trace_id/original_span_id.
        source_trace_id: nil,
        source_span_id: input_source_span_id
      }
      args, kwargs = adapt_inputs.call(args, kwargs, ctx)
    end
    fn_result = if kwargs.empty?
      receiver.send(method_name, *args)
    else
      receiver.send(method_name, *args, **kwargs)
    end
  rescue => e
    fn_error = e.message
  end

  # Wait for this item's trace (spans + completion) to be fully persisted
  # before the item resolves. Runs on the error path too: a raising
  # method still emits a root span whose trace must land before
  # complete_replay. Joins are bounded by the HTTP layer's own timeouts.
  pending_persistence.each(&:join)

  {
    input: args,
    result: fn_result,
    original_output: item["output"],
    error: fn_error,
    duration_ms: metrics[:duration_ms],
    tokens: metrics[:tokens],
    model: metrics[:model],
    # Written in by run() from the complete-replay response once the server
    # has minted this replay trace's row. Nil until then: the client-side
    # correlation id (below) is never surfaced as the public :trace_id.
    trace_id: nil,
    _sdk_trace_id: sdk_trace_id,
    original_trace_id: source_bitfab_trace_id,
    original_span_id: input_source_span_id,
    # Deprecated aliases for original_trace_id/original_span_id.
    source_trace_id: source_bitfab_trace_id,
    source_span_id: input_source_span_id,
    db_snapshot_ref:
  }
end

.extract_server_item_metrics(server_item) ⇒ Object

Pull durationMs / model from the start-replay server item. Nil-safe defaults so older servers without these fields still produce a consistent shape. Tokens are intentionally NOT read from the start item (it carries the ORIGINAL trace's tokens); the replayed run's tokens are filled in by run() from the complete-replay response once spans are aggregated server-side, and stay nil here and on older servers.



871
872
873
874
875
876
877
 
# File 'lib/bitfab/replay.rb', line 871

def extract_server_item_metrics(server_item)
  {
    duration_ms: server_item["durationMs"],
    tokens: nil,
    model: server_item["model"]
  }
end

.extract_span_data(span) ⇒ Object

Extract input/output data from an external span's rawData.



853
854
855
856
857
858
859
860
861
862
863
 
# File 'lib/bitfab/replay.rb', line 853

def extract_span_data(span)
  raw_data = span["rawData"] || {}
  span_data = raw_data["span_data"] || {}

  {
    "input" => span_data["input"],
    "output" => span_data["output"],
    "inputSerialized" => span_data["input_serialized"],
    "outputSerialized" => span_data["output_serialized"]
  }
end

.git(cwd, args) ⇒ Object 



407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
 
# File 'lib/bitfab/replay.rb', line 407

def git(cwd, args)
  # capture3 so git's stderr (e.g. "not a git repository") is swallowed, not
  # leaked to the host process's console.
  out, _err, status = Timeout.timeout(30) do
    Open3.capture3("git", *args, chdir: cwd)
  end
  return nil unless status.success?

  # Tag as UTF-8 and scrub invalid bytes so blob contents compare cleanly
  # against the UTF-8 working-file reads and never break JSON.generate later
  # (a non-UTF-8 file that slips past the null-byte check would otherwise
  # abort the whole start-replay request).
  out.force_encoding("UTF-8").scrub
rescue
  nil
end

.normalize_tokens(raw_tokens) ⇒ Object

Normalize a complete-replay tokens hash (string-keyed JSON) into the symbol-keyed shape the replay item exposes. Nil when the server reported no token data for this trace.



882
883
884
885
886
887
888
889
890
891
 
# File 'lib/bitfab/replay.rb', line 882

def normalize_tokens(raw_tokens)
  return nil unless raw_tokens.is_a?(Hash)

  {
    input: raw_tokens["input"],
    output: raw_tokens["output"],
    cached: raw_tokens["cached"],
    total: raw_tokens["total"]
  }
end

.original_span_id_of(server_item) ⇒ Object 



770
771
772
 
# File 'lib/bitfab/replay.rb', line 770

def original_span_id_of(server_item)
  server_item["originalSpanId"] || server_item["sourceSpanId"]
end

.original_trace_id_of(server_item) ⇒ Object

The ORIGINAL (historical) trace/span an item replays. Canonical server keys are originalTraceId/originalSpanId; older servers send them under the deprecated sourceTraceId/sourceSpanId aliases.



766
767
768
 
# File 'lib/bitfab/replay.rb', line 766

def original_trace_id_of(server_item)
  server_item["originalTraceId"] || server_item["sourceTraceId"]
end

.process_items(http_client, server_items, receiver, method_name, test_run_id, max_concurrency, mock_strategy, adapt_inputs = nil, include_db_branch_lease = false, db_branch_settings = nil, on_progress: nil, mock_overrides: []) ⇒ Object

Process all replay items, optionally in parallel using threads.



549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
 
# File 'lib/bitfab/replay.rb', line 549

def process_items(http_client, server_items, receiver, method_name, test_run_id, max_concurrency, mock_strategy,
  adapt_inputs = nil, include_db_branch_lease = false, db_branch_settings = nil, on_progress: nil,
  mock_overrides: [])
  concurrency = max_concurrency || server_items.length

  # Reports running totals once per item as it settles. In the parallel
  # path it runs from worker threads, so the mutex both makes the counter
  # updates safe and serializes the user's callback (never called
  # concurrently). A raising callback is swallowed: progress UI must never
  # crash the run.
  total = server_items.length
  progress_mutex = Mutex.new
  completed = 0
  succeeded = 0
  errored = 0
  # Each event carries the single item that just settled so a progress UI
  # can render per-trace pass/fail as the run streams. The item's :trace_id
  # is nil at this stage: the server replay trace id isn't known until run()
  # writes it in after complete_replay, and the client correlation id is
  # never surfaced. original_trace_id (the historical trace being replayed,
  # taken from the server item) is what a UI keys on to identify what just
  # settled. source_trace_id/source_span_id are kept as deprecated aliases.
  report = lambda do |result, original_trace_id, original_span_id, test_run_id|
    return unless on_progress

    progress_mutex.synchronize do
      completed += 1
      error = result[:error]
      error.nil? ? (succeeded += 1) : (errored += 1)
      begin
        on_progress.call({
          test_run_id:, completed:, total:, succeeded:, errored:,
          item: {
            trace_id: result[:trace_id],
            original_trace_id:,
            original_span_id:,
            # Deprecated aliases for original_trace_id/original_span_id.
            source_trace_id: original_trace_id,
            source_span_id: original_span_id,
            input: result[:input],
            result: result[:result],
            original_output: result[:original_output],
            error:,
            duration_ms: result[:duration_ms],
            tokens: result[:tokens],
            model: result[:model],
            db_snapshot_ref: result[:db_snapshot_ref]
          }
        })
      rescue => e
        warn "Bitfab: replay on_progress callback raised: #{e.message}"
      end
    end
  end

  if concurrency <= 1
    server_items.map do |item|
      result = process_single_item(http_client, item, receiver, method_name, test_run_id, mock_strategy,
        adapt_inputs, include_db_branch_lease, db_branch_settings, mock_overrides:)
      report.call(result, original_trace_id_of(item), original_span_id_of(item), test_run_id)
      result
    end
  else
    results_mutex = Mutex.new
    results = []
    work_queue = server_items.each_with_index.to_a
    work_mutex = Mutex.new

    workers = [concurrency, server_items.length].min.times.map do
      Thread.new do
        loop do
          item, idx = work_mutex.synchronize { work_queue.shift }
          break unless item

          result = process_single_item(http_client, item, receiver, method_name, test_run_id, mock_strategy,
            adapt_inputs, include_db_branch_lease, db_branch_settings, mock_overrides:)
          results_mutex.synchronize { results[idx] = result }
          report.call(result, original_trace_id_of(item), original_span_id_of(item), test_run_id)
        end
      end
    end

    workers.each(&:join)
    results.compact
  end
end

.process_single_item(http_client, server_item, receiver, method_name, test_run_id, mock_strategy, adapt_inputs = nil, include_db_branch_lease = false, db_branch_settings = nil, mock_overrides: []) ⇒ Object

Fetch span data and execute a single replay item.

Any error while fetching the span, building the mock tree, or deserializing inputs is captured on the returned item's :error rather than propagated, so one bad trace never aborts the whole replay run (mirrors the TypeScript and Python SDKs' per-item rescue).



642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
 
# File 'lib/bitfab/replay.rb', line 642

def process_single_item(http_client, server_item, receiver, method_name, test_run_id, mock_strategy,
  adapt_inputs = nil, include_db_branch_lease = false, db_branch_settings = nil, mock_overrides: [])
  metrics = extract_server_item_metrics(server_item)
  # The ORIGINAL (historical) trace/span this item replays. Canonical
  # keys are originalTraceId/originalSpanId; older servers send them under
  # the deprecated sourceTraceId/sourceSpanId aliases (see *_of helpers).
  original_trace_id = original_trace_id_of(server_item)
  original_span_id = original_span_id_of(server_item)
  # The server resolves a Neon preview branch per item during /replay/start
  # (only when include_db_branch_lease was sent). Release it in the +ensure+
  # below so any raise (span fetch, mock-tree build, or the replayed
  # method) frees the Neon resource. Items whose source trace had no
  # snapshot ref arrive without a lease (env.active? is false), and
  # replaying those against the live database is correct.
  lease = include_db_branch_lease ? server_item["dbBranchLease"] : nil
  # A resolve that was ATTEMPTED and failed is different: the caller asked
  # for a branch, so running their method against live data would produce a
  # result that looks valid and is not. Fail the item instead, loudly.
  lease_error = include_db_branch_lease ? server_item["dbBranchLeaseError"] : nil
  db_snapshot_ref = server_item["dbSnapshotRef"]
  if include_db_branch_lease && lease.nil? && lease_error.nil?
    resolved = http_client.resolve_db_branch_lease(test_run_id, original_trace_id, db_branch_settings)
    lease = resolved["lease"]
    lease_error = resolved["leaseError"]
    db_snapshot_ref = resolved["dbSnapshotRef"] || db_snapshot_ref
  end
  if lease_error
    raise "Replay requested a database branch for trace #{original_trace_id} but it " \
      "could not be resolved (#{lease_error["code"]}): #{lease_error["message"]}. " \
      "The method was not run, because replaying it against the live database would " \
      "produce a result that looks valid but did not use the historical data you asked for."
  end

  span = http_client.get_external_span(original_span_id)
  item_data = extract_span_data(span)

  # Fetch the span tree when the base strategy needs recorded outputs
  # ("all"/"marked") OR when selective mock overrides are present. Overrides
  # gate on the same non-root call-counter machinery the tree drives, so
  # the tree must be fetched for them to fire even under mock: "none".
  #
  # Only mock: "all" needs every span's recorded output inline (it mocks
  # every child), so it fetches an eager tree. "marked" and overrides mock
  # only a few spans, so they fetch a payload-free tree (includeOutputs=
  # false) and pull each mocked span's output lazily by externalSpanId,
  # never dragging down outputs that no span consumes.
  overrides_present = !mock_overrides.nil? && !mock_overrides.empty?
  include_outputs = mock_strategy == "all"
  mock_tree = nil
  if mock_strategy == "all" || mock_strategy == "marked" || overrides_present
    begin
      tree = http_client.get_span_tree(original_span_id, include_outputs:)
      mock_tree = build_mock_tree(tree["root"] || {})
    rescue Exception => e # rubocop:disable Lint/RescueException
      raise if e.is_a?(SystemExit) || e.is_a?(SignalException)
      # "all" and overrides both depend on the tree: "all" mocks every span
      # from it, and overrides gate on its call-counter machinery. If the
      # fetch fails, surface the error on the item rather than silently
      # running everything real with the overrides dropped. Only bare
      # "marked" can fall back to real execution (its marked spans just
      # re-run). Mirrors the Python SDK's has_overrides re-raise.
      raise if mock_strategy == "all" || overrides_present
      mock_tree = nil
    end
  end

  # Memoized per-item lazy fetcher, present only on the payload-free path
  # (not eager "all", whose outputs are already inline). Passed to the
  # replay context so resolve_recorded_output can pull a mocked span's
  # output on demand, once per span even if read twice (base mock + an
  # override's get_original_output).
  fetch_span_output = (mock_tree && !include_outputs) ? build_span_output_fetcher(http_client) : nil

  adapt_ctx = {
    original_trace_id:,
    original_span_id:,
    # Deprecated aliases for original_trace_id/original_span_id.
    source_trace_id: original_trace_id,
    source_span_id: original_span_id
  }

  execute_item(
    item_data,
    receiver,
    method_name,
    test_run_id,
    span["id"],
    metrics,
    input_source_trace_id: span["externalTraceId"],
    mock_strategy:,
    mock_tree:,
    mock_overrides:,
    fetch_span_output:,
    adapt_inputs:,
    adapt_ctx:,
    db_branch_lease: lease,
    source_bitfab_trace_id: original_trace_id,
    db_snapshot_ref:
  )
rescue => e
  warn "Bitfab: replay item for span #{original_span_id} failed before execution: #{e.message}"
  {
    input: [],
    result: nil,
    original_output: nil,
    error: e.message,
    duration_ms: metrics&.dig(:duration_ms),
    tokens: metrics&.dig(:tokens),
    model: metrics&.dig(:model),
    trace_id: nil,
    original_trace_id:,
    original_span_id:,
    # Deprecated aliases for original_trace_id/original_span_id.
    source_trace_id: original_trace_id,
    source_span_id: original_span_id,
    db_snapshot_ref:
  }
ensure
  release_db_branch_lease(http_client, lease) if lease
end

.read_code_change_file ⇒ Object 



387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
 
# File 'lib/bitfab/replay.rb', line 387

def read_code_change_file
  path = ENV["BITFAB_CODE_CHANGE_PATH"]
  return nil if path.nil? || path.empty?

  begin
    parsed = JSON.parse(File.read(path))
    files = parsed["files"]
    # A malformed payload (non-array, or entries that aren't hashes) must
    # yield no code change, never crash normalize_code_change_files.
    files = nil unless files.is_a?(Array) && files.all?(Hash)
    description = parsed["description"]
    description = nil unless description.is_a?(String)
    return nil if files.nil? && description.nil?

    [description, files]
  rescue
    nil
  end
end

.release_db_branch_lease(http_client, lease) ⇒ Object

Delete the per-item Neon preview branch. Best-effort: a failure is warned but never raised: the server-side TTL janitor reaps orphans.



776
777
778
779
780
781
782
783
 
# File 'lib/bitfab/replay.rb', line 776

def release_db_branch_lease(http_client, lease)
  neon_branch_id = lease["neonBranchId"]
  return unless neon_branch_id

  http_client.release_db_branch_lease(neon_branch_id)
rescue => e
  warn "Bitfab: failed to release DB branch #{neon_branch_id} (TTL janitor will catch it): #{e.message}"
end

.resolve_auto_code_change(label = nil) ⇒ Object

Auto-capture the code change to attach when the caller passed none. Lives in replay() (the one call guaranteed to run on every replay) so it works no matter how the replay was launched. Precedence: a BITFAB_CODE_CHANGE_PATH override, else the git diff vs trunk. Best-effort: any failure yields nil. An explicit code_change_files always wins over this.



378
379
380
381
382
383
384
385
 
# File 'lib/bitfab/replay.rb', line 378

def resolve_auto_code_change(label = nil)
  return nil if ENV["BITFAB_DISABLE_CODE_CHANGE_CAPTURE"]

  from_env = read_code_change_file
  return from_env unless from_env.nil?

  capture_code_change_from_git(Dir.pwd, label)
end

.run(client, receiver, method_name, trace_function_key:, limit: nil, trace_ids: nil, name: nil, max_concurrency: 10, code_change_description: CODE_CHANGE_UNSET, code_change_files: CODE_CHANGE_UNSET, experiment_group_id: nil, dataset_id: nil, grader_ids: nil, mock: "marked", adapt_inputs: nil, mock_override: nil, db_branch: nil, on_progress: nil) ⇒ Hash

Replay historical traces through a method and create a test run.

Fetches the last N traces for the given trace function key, re-runs each through the provided receiver and method, and returns comparison data.

Parameters:

  • client (Bitfab::Client) —

    the Bitfab client instance

  • receiver (Object, Class) —

    an instance for instance methods, or a Class for class methods

  • method_name (Symbol) —

    the method to replay

  • trace_function_key (String) —

    the trace function key for this method

  • limit (Integer, nil) (defaults to: nil) —

    maximum number of traces to replay (default: 5). Ignored when trace_ids is passed (with a warning): an explicit ID list already determines how many traces replay.

  • trace_ids (Array<String>, nil) (defaults to: nil) —

    optional list of trace IDs to replay (max 100)

  • name (String, nil) (defaults to: nil) —

    optional display name for the resulting experiment/test run

  • max_concurrency (Integer, nil) (defaults to: 10) —

    max threads for parallel replay (default: 10)

  • code_change_description (String, nil) (defaults to: CODE_CHANGE_UNSET) —

    optional rationale for the code change being tested in this replay (stored on the experiment). When supplied without code_change_files, this text is preserved while the SDK automatically captures the files. Pass nil when the replay should have no description; use code_change_files: nil to suppress automatic file capture.

  • code_change_files (Array<Hash>, nil) (defaults to: CODE_CHANGE_UNSET) —

    optional list of edited files, each as { path:, before:, after: } (empty string for new/deleted files). Omit to capture automatically, or pass nil to suppress file capture.

  • experiment_group_id (String, nil) (defaults to: nil) —

    optional UUID grouping multiple replay runs into a single experiment batch

  • dataset_id (String, nil) (defaults to: nil) —

    optional UUID of the dataset this replay runs against, stored on the resulting experiment for durable attribution

  • grader_ids (Array<String>, nil) (defaults to: nil) —

    optional UUIDs of graders attached directly to this experiment, graded as the union with the dataset's runnable graders at completion; each must be an active/live grader in the same org and trace function or the server rejects the replay

  • mock (String) (defaults to: "marked") —

    mock strategy for child spans: "marked" (default), "none", or "all". "marked" only mocks spans declared with mock_on_replay: true; "all" mocks every child span.

  • adapt_inputs (#call, nil) (defaults to: nil) —

    optional hook to reshape recorded inputs onto the method's current signature when its shape changed after the traces were captured. Receives (args, kwargs, ctx) where ctx is { original_trace_id:, original_span_id: } (with deprecated source_trace_id/source_span_id aliases), and returns [new_args, new_kwargs]. Runs per item inside the same rescue as the method, so a raising adapter sets that item's :error rather than crashing the run.

  • db_branch (Boolean, Hash, nil) (defaults to: nil) —

    true branches with the mirror project's own sizing; a Hash tunes it and +false+/+nil+ leave branching off. Keys +:min_cu+/+:max_cu+ are the branch compute's autoscaling floor and ceiling in Neon Compute Units, and :warmup_sql warms the branch's cache before the replayed method sees it. Read the resolved branch inside the method with Bitfab.current_replay_branch.

  • on_progress (#call, nil) (defaults to: nil) —

    optional callback invoked once per item as it finishes, with a running-totals hash { completed:, total:, succeeded:, errored:, item: } where item is { trace_id:, original_trace_id:, original_span_id:, error:, duration_ms: } for the single item that just settled (source_trace_id/source_span_id remain as deprecated aliases). trace_id is the new server replay trace id, written in after the run completes (nil during progress callbacks); original_trace_id is the ORIGINAL (historical) trace that was replayed, error is that item's replay error or nil, and duration_ms is how long that one trace took to replay. Use it to render replay progress (e.g. a per-trace log). A raising callback never crashes the run.

Returns:

  • (Hash) —

    with :items, :test_run_id, :test_run_url



149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
 
# File 'lib/bitfab/replay.rb', line 149

def run(client, receiver, method_name, trace_function_key:, limit: nil, trace_ids: nil, name: nil,
  max_concurrency: 10, code_change_description: CODE_CHANGE_UNSET, code_change_files: CODE_CHANGE_UNSET, experiment_group_id: nil,
  dataset_id: nil, grader_ids: nil, mock: "marked",
  adapt_inputs: nil, mock_override: nil, db_branch: nil, on_progress: nil)
  unless MOCK_STRATEGIES.include?(mock.to_s)
    raise ArgumentError, "Invalid mock strategy '#{mock}'. Must be one of: #{MOCK_STRATEGIES.join(", ")}"
  end
  if trace_ids
    raise ArgumentError, "trace_ids must contain at least one trace ID." if trace_ids.empty?
    if trace_ids.length > 100
      raise ArgumentError, "trace_ids supports at most 100 trace IDs per replay (got #{trace_ids.length})."
    end
  end
  if limit && trace_ids
    warn "Bitfab: limit is ignored when trace_ids is passed: the explicit trace ID list already " \
      "determines how many traces replay."
  end

  # Reject a trace_function_key that contradicts the method's declared key.
  # replay() fetches historical traces by trace_function_key but records the
  # replayed spans under the method's own bitfab_span key (via send below),
  # so a mismatch produces an incoherent test run (traces fetched for one
  # function, recorded under another). Only fires when the method's key is
  # introspectable; an untraced method falls through to the persistence
  # check in complete_replay. Mirrors the TypeScript/Python SDKs.
  declared_key = Traceable.trace_function_key_for(receiver, method_name)
  if declared_key && declared_key != trace_function_key
    raise ArgumentError,
      "Method #{method_name} is traced under trace function key '#{declared_key}' but replay was " \
      "called with '#{trace_function_key}'. Pass trace_function_key: '#{declared_key}' to replay it, " \
      "or point at the method traced under '#{trace_function_key}'."
  end

  http_client = client.instance_variable_get(:@http_client)

  # Resolved override list: per-call overrides FIRST (they win), then the
  # client's registered overrides. First matching override wins downstream,
  # so this ordering makes per-call take precedence over registered, and
  # both take precedence over the base mock strategy.
  registered_overrides = client.instance_variable_get(:@mock_overrides) || []
  resolved_overrides = MockOverride.normalize(mock_override) + registered_overrides

  # limit is meaningless with explicit trace_ids (the ID list determines
  # the count), so it's omitted from the request entirely.
  effective_limit = trace_ids ? nil : (limit || 5)

  include_db_branch_lease = db_branch_enabled?(db_branch)

  # code_change_files controls capture: omitted auto-captures, an array
  # wins, and explicit nil opts out. Preserve a caller-supplied description
  # while filling only the omitted files.
  if code_change_files.equal?(CODE_CHANGE_UNSET)
    captured = resolve_auto_code_change(name)
    unless captured.nil?
      code_change_files = captured[1]
      code_change_description = captured[0] if code_change_description.equal?(CODE_CHANGE_UNSET)
    end
  end
  code_change_description = nil if code_change_description.equal?(CODE_CHANGE_UNSET)
  code_change_files = nil if code_change_files.equal?(CODE_CHANGE_UNSET)

  resolved_db_branch_settings = db_branch_settings(db_branch)
  replay_data = http_client.start_replay(
    trace_function_key,
    effective_limit,
    trace_ids:,
    name:,
    code_change_description:,
    code_change_files:,
    experiment_group_id:,
    include_db_branch_lease:,
    dataset_id:,
    grader_ids:,
    db_branch_settings: resolved_db_branch_settings
  )
  test_run_id = replay_data["testRunId"]
  test_run_url = replay_data["testRunUrl"]
  server_items = replay_data["items"] || []

  result_items = if server_items.any?
    process_items(http_client, server_items, receiver, method_name, test_run_id, max_concurrency, mock.to_s,
      adapt_inputs, include_db_branch_lease, resolved_db_branch_settings, on_progress:,
      mock_overrides: resolved_overrides)
  else
    []
  end

  # Every item joined its own trace-persistence threads (span uploads +
  # completion) in execute_item, so all replay traces are on the server
  # by now: no flush needed. complete_replay finalizes the run and returns
  # the token/diagnostic mapping; its failures propagate loudly because a
  # run that never completed can't be finalized.
  complete_response = http_client.complete_replay(test_run_id)
  trace_id_map = complete_response&.dig("traceIds")
  # Per-replay-trace token usage keyed by server trace id: the REPLAYED
  # run's tokens (span-aggregated server-side), used below to fill each
  # item's :tokens.
  replay_tokens = complete_response&.dig("tokens") || {}

  # trace_id_map maps each item's client-side correlation id (:_sdk_trace_id,
  # which tagged that item's spans during the run) to the server's trace row
  # id. We use it to write the real server replay trace id into item[:trace_id]
  # now that the row exists, attach each item's server-aggregated token usage,
  # and detect a systemic upload failure early. Verdict persistence does NOT
  # use this map: it is keyed by the original-trace lineage (:original_trace_id +
  # test_run_id), which needs no client-held server id. Older servers that omit
  # the map yield no replay tokens and leave item[:trace_id] nil.
  unless trace_id_map.nil?
    missing = []
    completed_count = 0
    result_items.each do |item|
      local_id = item[:_sdk_trace_id]
      mapped = local_id && trace_id_map[local_id]
      # Write the real server replay trace id in as it comes back; the item
      # held nil until now (the client correlation id is never surfaced).
      item[:trace_id] = mapped
      if item[:error].nil?
        completed_count += 1
        missing << local_id if mapped.nil?
      end
      item[:tokens] = normalize_tokens(replay_tokens[mapped]) if mapped
    end
    # ALL completed items missing: systemic (the replayed method is not
    # traced, or uploads are wholesale broken). Raise; the run's traces
    # never persisted, so nothing can be labeled.
    if completed_count.positive? && missing.length == completed_count
      trace_count = complete_response["traceCount"]
      server_count = trace_count.nil? ? "" : " The server persisted #{trace_count} trace(s) for this run."
      raise "Replay completed but the server has no persisted trace for " \
        "any of the #{completed_count} completed item(s) " \
        "(test_run_id #{test_run_id}).#{server_count} Trace uploads were " \
        "joined, so either the uploads failed or the replayed method is " \
        "not traced (no root span was emitted)."
    end
    # SOME completed items missing: per-item upload failure. Warn, but
    # return the run; the items that landed can still be labeled.
    if missing.any?
      warn "Bitfab: server has no persisted trace for #{missing.length} of " \
        "#{completed_count} completed replay item(s) " \
        "(test_run_id #{test_run_id}). Their replay token usage is " \
        "unavailable and they cannot be labeled."
    end
  end

  # Strip the internal correlation handle so returned items expose only the
  # public shape (runs against older servers that omit the map too).
  result_items.each { |item| item.delete(:_sdk_trace_id) }

  result = {
    items: result_items,
    test_run_id:,
    test_run_url: "#{client.service_url}#{test_run_url}"
  }
  # Persist the enriched result two ways so the Bitfab plugin never has to
  # parse the replay's stdout (which a dependency's logging can corrupt):
  # write it to BITFAB_REPLAY_RESULT_PATH when the plugin set that env var,
  # and stream a terminal "complete" progress event. The plugin prefers the
  # streamed event and falls back to the file. The event routes through
  # on_progress so only progress-reporting runs emit it.
  write_replay_result_file(result)
  if on_progress
    errored = result_items.count { |item| !item[:error].nil? }
    total = result_items.length
    begin
      on_progress.call({
        type: "complete",
        test_run_id:,
        completed: total,
        total:,
        succeeded: total - errored,
        errored:,
        result:
      })
    rescue => e
      warn "Bitfab: replay on_progress callback raised: #{e.message}"
    end
  end
  result
end

.write_replay_result_file(result) ⇒ Object 



535
536
537
538
539
540
541
542
543
544
545
546
 
# File 'lib/bitfab/replay.rb', line 535

def write_replay_result_file(result)
  result_path = ENV["BITFAB_REPLAY_RESULT_PATH"]
  return if result_path.nil? || result_path.empty?

  begin
    dir = File.dirname(result_path)
    FileUtils.mkdir_p(dir) unless dir.nil? || dir.empty? || dir == "."
    File.write(result_path, "#{JSON.pretty_generate(result)}\n")
  rescue => e
    warn "Bitfab: failed to write replay result to BITFAB_REPLAY_RESULT_PATH (#{result_path}): #{e.message}"
  end
end