Previously we extracted attributes from DDL plan nodes and left
the nodes in arena in the inconsistent state. This approach is
error prone, so now we extract the whole node from the plan arena
before using its attributes.

Previous version of the SQL module hierarchy was overcomplicated.
Current commit simplifies it, so SQL has only two submodules:
router and storage.

Previously we created a very strange error sandwitch when failed
on the storage or router runtimes:
- wrapped traft error into a sbroad one
- wrapped the result again into traft error

Now we wrapp sbroad errors into traft or return traft errors as is.

Now it is possible to drop sharded tables with pico.sql() Lua
function.

Adds support for creation of the sharded memtx spaces via pico.sql().

The reason of this refactoring is that we need Picodata specific
API rather then the API provided by tarantool module. For example,
table creation needs access to the clusterwide API for sharded and
global spaces.

Georgy Moshkin authored 1 year ago and Georgy Moshkin committed 1 year ago

Closes #321

Denis Smirnov authored 1 year ago and Yaroslav Dynnikov committed 1 year ago

INSERT command in the SQL library (sbroad) was refactored. Now
we don't need a custom SQL function to calculate bucket ids. So,
remove redundant functions from picodata SQL module.

Denis Smirnov authored 1 year ago and Yaroslav Dynnikov committed 1 year ago

Check that picodata SQL uses `key_def` module from Tarantool to
calculate the tuple hash.