In general, the latest release of TiKV Java Client is ready for production use. But it is not battle-tested as full featured client for TiKV in all use cases. This page will give you more details.
All RawKV APIs are covered by CI.
At this time, RawKV has been used in the production environment of some commercial customers in latency sensitive systems. But they only use part of the RawKV APIs (mainly including raw_put, raw_get, raw_compare_and_swap, and raw_batch_put).
All TxnKV APIs are covered by CI.
In addition, TxnKV has been used in the TiSpark and TiBigData project to integrate data from TiDB to Big Data ecosystem. TiSpark and TiBigData are used in the production system of some commercial customers and internet companies.
Similar to RawKV, only part of APIs are used in this scenario (mainly including prewrite/commit and coprocessor). And this use case doesn't care about latency but throughput and reliability.
Directly using TiKV is not possible on TiDB Cloud due to the fact that client has to access the whole cluster, which has security issues. And TiKV managed service is not coming soon as it's not contained in roadmap yet.