feat(search): Added Alias Commands to Search

[jonathanc-n]

references #2555

Added alias commands to search with unit tests in Go. (FT.ALIASADD, FT.ALIASDEL, FT.ALIASUPDATE)

Just wanted to see if this seemed fine before adding it to the actual alias functionality of the other search commands. Will create the alias functionality immediately if this PR gets merged.

Alias Subkey
ns | type | index | alias_name => index_name

[jonathanc-n]

Is somebody able to resolve the conflict, its just the previous command is just in the same place. When I try to retrieve from origin the previous commits aren't showing up right now, sorry about that

[PragmaTwice]

Thanks for your contribution.

However FIELD_ALIAS is not for this purpose. It's for alias of index fields.

Also we should see the encoding design in your proposal, but I think now it's not consistent with the implementation.

[jonathanc-n]

@PragmaTwice I decided to switch it up the serialization on the implementation, I can include it on the pr. As for the field_alias, you can use this implementation to use index fields to look and see if the alias matches when it receives a non matching index in a command (which will be implemented in a GetAlias function).

[PragmaTwice]

FT.ALIAS* is for alias of indexes.

FIELD_ALIAS is for alias of fields.

Indexes is not fields, and one index can includes several fields.

Your implementation is quite confusing, if you really want to add alias of indexes, the encoding should be something like

ns | INDEX_ALIAS | alias => index

If you put index name before alias, how do you quickly map from alias to the index name before you know the original index?

I'd suggest that figure out what you want to do and how to do before writing some random code.

[PragmaTwice]

Besides ADD, UPDATE and DELETE operations of alias, the most important part of aliasing is to find the corresponding index by the user-provided alias in query commands (e.g. FT.SEARCH, FT.SEARCHSQL).

So we need to consider how to efficiently identify the alias and find the corresponding index in the query (in a very little runtime overhead). e.g. for index a, b and their aliases (c -> index a, d -> index b), we can put a key like this:

// alias
ns | INDEX_ALIAS | c -> a
ns | INDEX_ALIAS | d ->b

// index metadata (already exists)
ns | INDEX_META | a -> ...
ns | INDEX_META | b -> ...

And when we saw a query like select * from <idx>, we need to check if <idx> is a real index name or it's an alias, e.g.

• select * from a: a is a real index name.
• select * from d: d is an alias and the real index is b.
• select * from e: e doesn't exist.

For index names and its metadata, we will load all of them into the memory from start instead of getting it from rocksdb every time we need it to achieve better performance.

So I think for alias we should also load from start.

In your design:

Alias Subkey
ns | type | index | alias_name => index_name

The index name seems repeated in both keys and values, which is not so useful.

[jonathanc-n]

@PragmaTwice While the load commands happen do you think it'd be better to create a global alias map in index_manager?
There would be two places for aliases to be stored:

1. A global alias map where it is 'alias -> index' for quick alias to index lookups.
2. A alias array in IndexInfo for deleting all aliases that are associated with the index quickly if ft.dropindex is called.