发布

RSS 订阅源

2.0.61-5591 b20f9e1806

比较
[OPIK-6482] [BE] fix: fetch dashboards page in two steps to avoid filesort OOM on JSON config (#7015)

frostbyte_neo 发布于 2026-06-10 13:51:45 +00:00
- [OPIK-6482] [BE] fix: fetch dashboards page in two steps to avoid filesort OOM on JSON config
The dashboards list query did SELECT * ... ORDER BY id and, when the optimizer could
not satisfy the ordering from an index, fell back to a filesort. On MySQL 8.0.20+ the
filesort packs the selected columns (the addon-fields algorithm) — including the large
JSON config column — into the sort buffer. Once a single row's config exceeds
sort_buffer_size, MySQL raises ER_OUT_OF_SORTMEMORY and the list endpoint returns 500.
max_length_for_sort_data no longer forces the rowid algorithm in 8.0.20+, so the only
reliable control is to keep config out of the sorted projection.

Fetch the page in two steps: findPageIdsSorted orders and paginates selecting only id
(a narrow sort that cannot exceed sort_buffer_size), then findByIds loads the row bodies
with no ORDER BY and the service restores ordering in memory. This removes config from
every sort buffer regardless of config size or sort field, and works for any sort column.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6482] [BE] test: cover two-step reorder under custom sort + pagination
Adds a DashboardsResourceTest case that creates dashboards with random names
(name order != id order), then sorts by name and traverses every page, asserting
each page's full content (recursive comparison, not just ids) matches the expected
sorted slice. This verifies the two-step fetch's in-memory reorder under a custom
sorting_field across all pages, which the prior sort tests (single page, id-only
assertions) did not cover. Dashboards are created in parallel to speed up the run.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6482] [BE] refactor: drop redundant null filter in two-step dashboards fetch
Both findPageIdsSorted and findByIds run in the same READ_ONLY transaction
(consistent snapshot), so every pageId resolves in findByIds and the
filter(Objects::nonNull) can never drop an element. Trust findByIds' results
per review feedback.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Co-authored-by: Claude Opus 4.8 (1M context) noreply@anthropic.com
下载附件
- 源代码 (ZIP)
- 源代码 (TAR.GZ)

0 发布 8,340 Git 标签

[OPIK-6482] [BE] fix: fetch dashboards page in two steps to avoid filesort OOM on JSON config (#7015)