When TiKV is selected as the metastore, why do single KV writes and deletions in the code require transaction commits?

[scuzyj]

When TiKV is selected as the metastore, why do single KV writes and deletions in the code require transaction commits?

[xiaofan-luan]

right now we don't support TiKV as the meta store.
ETCD is still the main solution.

[scuzyj]

Thanks, there is already support for TiKV in both the code and configuration.
Is TiKV used as a metadata store not completed testing? So, it is not recommended to use TiKV as the meta store.

[xiaofan-luan]

We did some test over tikv long time ago.

But it never put into production yet. ETCD should be good enough for most of the case unless you are running with huge amount of data.

[scuzyj]

We want to test vector data of billions, with a large number of segments, which is estimated to generate nearly millions of KV pairs.
In the implementation of the KV interface, using transactions for single KV save and remove interfaces in TiKV feels meaningless and inconsistent with the etcd implementation.
Additionally, why does TiKV use snapshots for all query interfaces?

[JackLCL]

We want to test vector data of billions, with a large number of segments, which is estimated to generate nearly millions of KV pairs. In the implementation of the KV interface, using transactions for single KV save and remove interfaces in TiKV feels meaningless and inconsistent with the etcd implementation. Additionally, why does TiKV use snapshots for all query interfaces?

@scuzyj We have had many users who have tested billions of vector before, and Milvus has been deployed in the production environment. Especially in the scenarios of autonomous driving and public opinion risk control, vector data of this scale is very common. Would you please add my WeChat (13772075230) and schedule an online meeting to discuss your test scenario? We can give you some suggestions based on our previous experience to help you complete this test.

[scuzyj]

Thanks, I'm exploring the possibility of using TiKV to manage the metadata of the openGemini distributed time-series database and would like to understand how TiKV is used for metadata storage.