get_index_for_condition should take into account the partial index

[R-omk]

Since exclude_null is implemented choosing the wrong index may affect the result.

[DifferentialOrange]

To be honest, I don't get what issue is described here, especially considering exclude_null comment. How exactly we should take partial index into account? We already pick multipart index if there is a first field in condition specified. In other cases, there is fullscan either way so it's not like multipart index is more preferable than any other one.

It seems that this issue is likely to be closed if we support specifying scan_index for issue #346 . If it is so, I'll close this issue in favor of #346

[R-omk]

pick multipart index if there is a first field in condition specified

that's exactly the problem, it cannot use a partial index because if any other field is null, then the record will simply be lost.

Before closing this ticket it is better to make a draft with tests covering this case

[DifferentialOrange]

that's exactly the problem, it cannot use a partial index because if any other field is null, then the record will simply be lost.

Can you provide an example of the request and the description of expected and actual behavior? If relevant, with single instance request comparison.

[R-omk]

Ughhh...

let's run command from Readme -> Quickstart

$ git clone https://github.com/tarantool/crud.git
$ cd crud
$ tarantoolctl rocks make
$ ./doc/playground.lua
tarantool> crud.select('customers', {{'<=', 'age', 35}}, {first = 10})
tarantool> crud.select('developers', nil, {first = 6})

try to find Elizabeth

...

tarantool> crud.select('customers', {{'=', 'name', 'Elizabeth'}}, {first = 10})
---
- metadata: [{'name': 'id', 'type': 'unsigned'}, {'name': 'bucket_id', 'type': 'unsigned'},
    {'name': 'name', 'type': 'string'}, {'name': 'age', 'type': 'number'}, {'type': 'number',
      'name': 'nul', 'is_nullable': true}]
  rows:
  - [1, 477, 'Elizabeth', 12]
  - [7, 693, 'Elizabeth', 18]
- null
...

ok, the records are really exists

let's add nullable field and add an partial index, it shouldn't break anything, right?

box.space.customers:alter({name = 'customers', format = {
    {name = 'id', type = 'unsigned'},
    {name = 'bucket_id', type = 'unsigned'},
    {name = 'name', type = 'string'},
    {name = 'age', type = 'number'},
    {name = 'nul', type = 'number', is_nullable=true},
}})

box.space.customers:create_index('partial1', {parts = {{'name', 'string', is_nullable=true,exclude_null=true},{'nul', 'number', is_nullable=true,exclude_null=true}}})

look for Elizabeth again

tarantool> crud.select('customers', {{'=', 'name', 'Elizabeth'}}, {first = 10})
---
- metadata: [{'name': 'id', 'type': 'unsigned'}, {'name': 'bucket_id', 'type': 'unsigned'},
    {'name': 'name', 'type': 'string'}, {'name': 'age', 'type': 'number'}, {'type': 'number',
      'name': 'nul', 'is_nullable': true}]
  rows: []
- null
...

oops, what happened?

drop index and retry search

tarantool> box.space.customers.index.partial1:drop()
---
...

tarantool> crud.select('customers', {{'=', 'name', 'Elizabeth'}}, {first = 10})
---
- metadata: [{'name': 'id', 'type': 'unsigned'}, {'name': 'bucket_id', 'type': 'unsigned'},
    {'name': 'name', 'type': 'string'}, {'name': 'age', 'type': 'number'}, {'type': 'number',
      'name': 'nul', 'is_nullable': true}]
  rows:
  - [1, 477, 'Elizabeth', 12]
  - [7, 693, 'Elizabeth', 18]
- null
...

Sigh... looks like it's fixed...

[DifferentialOrange]

Thank you, now I'll get what this ticket is about.