-
[OPIK-6899] [BE] feat: apply benchmark-driven codec refinements to traces_local_v2 (#7492)
frostbyte_neo 发布于
2026-07-17 08:19:51 +00:00 - [OPIK-6899] [BE] feat: apply benchmark-driven codec refinements to traces_local_v2
Add migration 000103 applying the six best-guess codec refinements the
OPIK-6899 benchmark identified over the provisional 000101 codecs
(metadata-only on the empty pre-cutover table):- end_time / last_updated_at: drop Delta (not monotonic in storage order;
plain ZSTD(1) ~11% smaller) - visibility_mode / source / environment: server default -> ZSTD(1)
(~2x smaller at equal decode) - output_keys: ZSTD(1) -> ZSTD(3) (structured text, ~16% smaller)
Best guesses on a synthetic slice, to re-validate at the staging cutover;
none is ever worse than the 000101 codec, so adopting now is low-regret.- fix(backend): use empty rollback for the in-place codec migration (000105)
Per the migrations guideline, in-place column changes (default/codec) take
--rollback empty — reverting a codec is not a clean inverse once data
exists, and restoring the provisional 000101 codecs is never a wanted
recovery step. Replaces the explicit revert-to-000101 rollback.- fix(backend): renumber codec migration to 000106 after rebase on main
main gained 000105_add_id_at_to_spans, so bump the refinements migration to
000106 (keeps it last) and update the changeset id and the pin-map/whole-row
references in the test accordingly.- fix(backend): apply 000106 changeset id and test references
Fill in the content the rename-only commit missed: set the migration's
changeset id to 000106 and update the pin-map/whole-row references in the
test from 000105 to 000106.
下载附件