Exact

Exact Union RegExp

Gerhard Schlager b8d389cb1b MT: Only use INSERT OR IGNORE for tables that want it in the shard merge ... The sharded write path (#41262) merges every shard table with `INSERT OR IGNORE`. That's right for `uploads`, but for every other table it silently swallows a genuine cross-step/cross-shard duplicate row that the single-writer path would have raised on — so a concurrency bug that doubles a key becomes silent data loss with no error and no baseline to diff against. This scopes the `OR IGNORE` back to just the tables that actually want it, derived from the models rather than a hardcoded table name: - `IntermediateDB.conflict_strategy_for(table)` resolves a table's model and reads its declared `conflict_strategy`, defaulting to `:raise`. - Only `upload` declares `:ignore` (its rows are content-identical for a given id-hash, so first-writer-wins is intended). A future `OR IGNORE` model flips its own merge clause with no change to the merge code. - `Connection#merge_database` takes `dedupe_tables:` and picks the clause per table; a plain-table conflict raises, wrapped with the table name, and the error surfaces through `Consolidator#drain` as a run failure. This restores the earlier behaviour where a genuine duplicate on a plain table is an error, not a dropped row.		2026-07-03 17:05:25 +02:00
..
intermediate_db
connection_spec.rb	MT: Only use INSERT OR IGNORE for tables that want it in the shard merge	2026-07-03 17:05:25 +02:00
intermediate_db_spec.rb	MT: Only use INSERT OR IGNORE for tables that want it in the shard merge	2026-07-03 17:05:25 +02:00
migrator_spec.rb	MT: Run conversion steps concurrently and split the heavy ones across cores	2026-07-03 11:48:25 +02:00
prepared_statement_cache_spec.rb