Changelog

Covers all DatasetStore operations from 0% to near-full coverage:

• CRUD: create_empty, list, get, delete, rename, set_category
• Mutations: insert_row, update_cell, delete_rows, add_column,

drop_column, rename_column, alter_column_type, duplicate,

import_csv_content, execute_mutation (committed + dry-run)

• Queries: query_dataset (pagination, sort, filter), get_columns,

execute_raw_query, aggregate_query, aggregate_query_grouped,

execute_sql_v2 (with source: alias resolution)

• Relations: create, list, delete
• Row-page linking: link, get_page_for_row, get_row_for_page, unlink
• Saved queries: create, update, delete, list, is_view flag
• Views: create, update, delete, list, config roundtrip
• Types: DatasetId format, DatasetColumnType mapping, serde roundtrips,

RelationType, StatementType, FilterOperator, SortDirection

1. DatasetColumnType::from_duckdb("DATETIME") mapped to Date instead of

Timestamp — the DATE branch matched before the TIMESTAMP/DATETIME

branch. Reorder match arms so Timestamp is checked first.

2. DatasetStore::delete() violated FK constraints by deleting the

_datasets_meta row before the _dataset_relations, _dataset_row_pages,

and _dataset_views rows that reference it. Reorder to delete

dependent rows first.

3. query_mutation_preview() panicked on DuckDB 1.4.4 because

column_count()/column_name() crash on un-executed prepared

statements. Use DESCRIBE to get column info instead.

• cross_verify_integration.rs: list prefix was missing `users/` and

`workspaces/` path segments, so it matched 0 keys instead of 3

• build.sh: before starting test containers, kill any process already

bound to ports 9000 (MinIO) or 3307 (MySQL) to prevent docker

compose bind failures

Update test files to match current struct definitions:

• StsCredentials: add missing `prefix` and `endpoint` fields
• CloudSyncStatus: add `is_rate_limited`, `rate_limit_remaining_secs`,

`synced_entity_count`, `total_entity_count` fields

• PendingChanges: rename `entities` → `pending` field
• PendingEntity: rename `current_cursor` → `latest_cursor`, remove `batches`

Add CloudError::is_transient() to detect 503s, connection resets, dispatch

failures, and timeouts. The sync engine now logs these at debug level with

a "(transient, will retry next cycle)" note instead of error/warn, reducing

log noise from normal S3 hiccups. Genuine failures remain at error/warn.