Allow backup+restore mutation with "in progress" status, disallow backup tables and parts have inconsistent structure

[Slach]

• make hard links /var/lib/clickhouse/backup/backup_name/shadow/db/table/disk/part_name -> /var/lib/clickhouse/(data/db/table|store/uuid)/detached/part_name
• maybe need SYSTEM STOP MERGES, SYSTEM RESTORE REPLICA for ReplicatedMergeTree

[Slach]

Yet another idea:

• save SELECT * FROM system.mutations WHERE is_done=0 into metadata/db/table_name.json
• restore ReplicatedMergeTree as MergeTree, and create mutations_X.txt with ATTACH TABLE
• poll SELECT * system.mutations WHERE is_done=0
• ALTER TABLE tmp_table ATTACH PART FROM TABLE orig_table to allow replicate mutated parts

[Slach]

need create properly test case to test behavior

to avoid race-condition state, I tried something like

CREATE TABLE t1 (v UInt64, m UInt64) ENGINE=MergeTree() PARTITION BY v ORDER BY v;
INSERT INTO t1 SELECT number, number FROM numbers(99);

SYSTEM STOP MERGES;
ALTER TABLE t1 MODIFY COLUMN m String;

SELECT type, active, count() FROM system.parts_columns WHERE table='t1' AND column='m' GROUP BY type, active FORMAT Vertical;

but ALTER TABLE is stuck when MERGES stop ...
and executed synchronously when i run it before SYSTEM STOP MERGES...
if even we will try to run both ALTER + STOP MERGES parellel, i don't sure ALTER will stop or stuck ...

Need ideas how to make possible ALTER TABLE ... in progress state and make ALTER TABLE ... FREEZE ..
without generate gigabytes of data!

[Slach]

ATTACH PART works perfectly , ATTACH TABLE doesn't need

but need to add backup_mutations was added to allow us backup and restore in progress mutations

[Slach]

to disallow backup inconsistent types in data parts

need execute

SELECT database, table, column, uniqExact(type) AS column_types
FROM system.parts_columns
WHERE active
GROUP BY database, table, column
HAVING column_types>1 

and add separate config option clickhouse->check_parts_columns and CLI parameter --check-parts-columns to create and create_remote command