Insert with sequence index

[aiibe]

I tried to insert new tuples using the sequence index, but I got this error: Invalid MsgPack - packet body

local res, err = tar:insert('projects', { 16, { name = 'Festival 2018' } }) -- Works fine
...

With a nil primary index , I expected Tarantool to auto generate it...

local res, err = tar:insert('projects', { nil, { name = 'Festival 2018' } })
if err return nil, err  -- Got above error
...

which works with tarantoolctl using this command:

localhost:3301 >> box.space.projects:insert{nil, 'other stuff'}
--- [17, 'other stuff']

Do I have to specify a primary index everytime to insert new tuple? Does lua-resty-tarantool support Sequence ?

[affair]

Hi @aiibe. Did you solve the problem?

[DarkWiiPlayer]

This seems to be a problem with lua-MessagePack, which turns a table like {nil, "foo", "bar"} into a messagepack map. To illustrate, this is the JSON representation of the returned MessagePack data:

{
  "2": "foo",
  "3": "bar"
}

This is what ticks tarantool off, because it expects an array and gets a map.

Sadly though, this is incredibly tricky to fix. lua-MessagePack provides no way (at least no documented way) of hinting that a table is a sparse array. There's a global option to enable holes in arrays, but this seems to completely break most other parts of the protocol.

Why the author of that library decided it would be a good idea to set such options globally and not allow any overrides in the function call is beyond me.

---

Okay, so much for my investigation; how could this be fixed?

The only thing that seems realistic to me is to move away from lua-MessagePack and switch to another, less poorly designed alternative. I tried this a while ago and ultimately dropped the idea after openresty segfaulted while trying to load that library (yay C-extensions!).

One way or another, fixing that problem will most likely be a rather large pull-request and take lots of work. I wouldn't count on it happening anytime soon.

---

Might there be a workaround?

I don't know, but it might work to just increment the sequence manually and use the returned value as the index. It's ugly, but it might at least work.