ClickHouse version: Altinity Stable 25.3.6.10034

Environment: 3-node replicated cluster with embedded ClickHouse Keeper

Summary

SELECT generateSerialID('some_sequence') blocks indefinitely when Keeper quorum is lost. The behavior is non-deterministic — sometimes it returns error code 999 (KEEPER_EXCEPTION / Operation timeout) after roughly operation_timeout_ms, but other times it hangs with no timeout at all. max_execution_time has no effect on the hang.

Reproduction

1. Set up a 3-node cluster with embedded ClickHouse Keeper
2. With all 3 Keeper nodes healthy, run SELECT generateSerialID('test') — works normally
3. Stop 2 of 3 Keeper nodes to break quorum
4. Run SELECT generateSerialID('test') — the query hangs indefinitely

Observed behavior

• Sometimes: query hangs, no error, no timeout (it fails after send_receive_timeout passes)
• Sometimes: returns Code: 999. Coordination::Exception: Operation timeout after ~10s

Expected behavior

generateSerialID should consistently respect operation_timeout_ms (or a similar setting) and return an error within a bounded time when Keeper quorum is unavailable.

Workaround

Client-side HTTP timeout.

Questions

1. Is there a server-side or query-level setting that can enforce a timeout on the internal Keeper call made by generateSerialID?
2. Is the inconsistency (sometimes timeout, sometimes hang) a known issue?