Problem

Lite connects to all configured servers sequentially. When servers are offline/unreachable, each connection attempt waits ~30s for the TCP timeout before moving on. With 7 offline servers, this eats ~8 minutes per collection cycle, making the effective refresh interval much longer than intended.

Expected behavior

Offline servers should be detected quickly (short timeout or async probing) so they don't block collection from healthy servers.

Possible approaches

• Reduce SQL connection timeout for Lite (e.g. 5s instead of 30s default)
• Skip servers that failed N consecutive times (exponential backoff)
• Collect from servers in parallel instead of sequentially
• Surface server names in collection log (currently only integer hash server_id, can't tell which server is which)

Additional context

During testing with 8 configured servers (only sql2022 reachable), Lite completed only 4 collection cycles in 40 minutes. The DuckDB collection_log showed 448 errors, all from unreachable servers.