Serge Petrenko

Backports: 3.1https://github.com/tarantool/tarantool/pull/10065 2.11 https://github.com/tarantool/tarantool/pull/10066

Backports: 3.1 https://github.com/tarantool/tarantool/pull/10255 2.11 https://github.com/tarantool/tarantool/pull/10254

> Signed integers (`MP_INT`) often are treated in the code as strictly negative, which results in assertions and/or incorrect behaviour in most or all of these places: > * `datetime`,...

@Gerold103, thanks for the investigation! > Proposed solution > > Use 3.2 (re-decode without MP header) and re-build Vinyl bloom filters. Sounds good to me. Are you going to handle...

> For Memtx there is no need, because it simply won't start anyway, if there are such duplicates in unique indexes. It seems the same script that would run for...

Good news, everyone! We've reconsidered the decision to put this into EE. The patch will be put into CE. Thank you for your feedback and thank you for caring.

Some quick perf test results. Current master: ```shell Tarantool 3.2.0-entrypoint-114-g7db4de75f8 Target: Linux-x86_64-RelWithDebInfo ## 150 operations per txn ./src/tarantool perf/lua/1mops_write.lua --nodes 2 # making 10000000 REPLACE operations, # 150 operations per...

> Such a huge impact with such a small patch, huh? Yeah, the difference is crazy.

> @sergepetrenko Don't we have the same situation without mvcc, like described in https://github.com/tarantool/tarantool/issues/10142? When we're waiting for disk write to happen, read-only can become true and we'll successfully commit...

> Am I right, that we need wal_sync in https://github.com/tarantool/tarantool/issues/10142, because we may start writing to the disk in wal_write_to_disk (and consequently we passed read-only check), but as wal didn't...