Nobody knows the trace id

1# a budget decision, not a default:
2INDEXED_FIELDS = intrinsics + nominated_attrs
3#   intrinsics = service, operation, status, duration
4 
5def write_span(span):
6    blocks.append(span)
7    for field in INDEXED_FIELDS:
8        value = span.get(field)
9        if value is None: continue
10        # one posting per distinct value of field
11        index.add(field, value, span.trace_id)
12    # index compacts and is backed up with the blocks

1def by_attribute(question, window):
2    if not question.fields <= INDEXED_FIELDS:
3        return CANNOT_ANSWER
4    postings = index.lookup(
5        question.fields, question.values,
6    )
7    # window narrows postings; it reads nothing more
8    ids = [p.trace_id for p in postings
9           if p.time in window]
10    return [get_by_id(tid) for tid in ids]

1def scan_blocks(question, window):
2    candidates = []
3    for block in blocklist:
4        # every block records its min/max time
5        if block.max_time < window.start: continue
6        if block.min_time > window.end: continue
7        if not block.stats.may_contain(question):
8            continue
9        candidates.append(block)
10    # thousands of short-lived jobs, a slice each
11    return fan_out(candidates, read_and_match)

1# wired up on the metrics side, months ago
2def observe(duration, ctx):
3    bucket = histogram.bucket_for(duration)
4    bucket.count += 1   # the value counts here too
5    bucket.exemplars.append(Exemplar(
6        trace_id=ctx.trace_id,
7        span_id=ctx.span_id,
8        value=duration,
9    ))   # up to 5 per data point
10 
11def open_from_p99(bucket):   # no search runs
12    return get_by_id(bucket.exemplars[0].trace_id)