It's possible to have shards and unique indexes?

[vitorenesduarte]

Hi,
I was testing how the indexes behave with shards and noticed that is possible to have two records with the same unique property.
My schema:

ALTER DATABASE MINIMUMCLUSTERS 3

create class User extends V
create property User.name string
create index User.name unique
alter cluster user_0 name user_usa
alter cluster user_1 name user_eur
alter cluster user_2 name user_chi

• hazelcast.xml
• default-distributed-db-config.json

This code will reproduce it.
The output is:

0 : ok
1 : ok
2 : ok
3 : ko -> result : [mike, mike]
4 : ko -> result : [mike, mike]
5 : ko -> result : [mike, mike]

Start first node usa so that its port is 2424. Then eur -> 2425. Then chi -> 2426.

After the first insert (on chi node), I believe this is what happens:

• Node eur see that it does not have a record with name mike and asks node usa if it has, but it doesn't (only chi node has a record with name mike).
• As writeQuorum is 2, the record is inserted.

My question is: if it's an unique index, shouldn't the writeQuorum be disrespected, and check all nodes?

v2.0.7
Logs:

• usa node
• eur node
• chi node

[lvca]

This is not supported yet.

[lvca]

This will be supported in v3.0 with the autosharded index. In v2.2 we already have a local implementation, in v3.0 will be supported in distributed mode to have a regular DHT.

[hossein-md]

Hi @lvca
I have tried this issue to reproduce the result with the version 2.2.33. I can't!
But I could after I had changed my Orientdb version to 2.0.7.
I don't understand why it happens. ( This is my problem that I mention in #8234 )