Bad performance when query with trace_id only

[jiekun]

Is your feature request related to a problem? Please describe

VictoriaTraces organizes data by streams. When querying by trace_id only, it needs to:

1. Query within ALL streams.
2. For each stream, query all the data from new to old by step (let's say step=3h)
3. Break when it finds the first result.

For the worst case (query with an "oldest" trace_id), VictoriaTraces needs to scan all the data (full time range * all streams), hence poor performance.

Describe the solution you'd like

PoC of the following:

• create a helper stream trace_id_idx when ingesting data.
  • this stream contains 2 (or 3) fields: trace_id and time (or start_time and end_time).
  • insert row to this helper stream when ingesting the root span (whose parent_span_id is empty).
• query helper stream for the time (or time range) and use it as the filter when querying by trace_id only.

Describe alternatives you've considered

No response

Additional information

No response