Change Restore method to be a "Safe Restore"

[DantasB]

It's just a suggestion to the clickhouse-backup CLI.

When you apply a restore, passing the --schema and --data, the steps are basically:

1. Drop the table if exists
2. Create table
3. Attach partitions to this existing table.

But, lets supose that the attach partitions, for some how, just breaks? For that, i would suggest to do a:

1. Drop the table_tmp if exists
2. Create table_tmp
3. Attach partitions to this existing table_tmp
4. Drop the table if exists
5. Rename table_tmp to table.

With that approach, the system would be almost always up and could be safer. And if any problem occurs in the tmp table, it would be just a tmp table and could not bring ANY data loss, or am I wrong?

[Slach]

this approach have side effect and need a lot of additional logic related to Materialized View, Replicated* and distributed engines

[DantasB]

OOOF, what kind of side effect? Just curious about it

[Slach]

lot of corner cases with replication_path parameter
need lot of work with dependencies (for example dictionary with SOURCE(CLICKHOUSE(...)))