Saved-response replay stalls with an artificially slow writer, even when the read returns immediately.

That pins down a logging delay; keep a separate delayed-read case, because fixing the writer will not make a slow read fresh.

Give the writer a bounded queue with a defined overflow and failure report. Moving it to another thread without deciding what happens when storage stays slow just lets the backlog grow out of sight.

What will the operator see when the log writer falls behind? Useful if that differs from the displayed reading becoming stale.

Writer backlog is separate from reading age. Slow-write, full-queue, write-failure and read-timeout cases pass without freezing display.

Was that the installed viewer too, including shutdown with queued output? A fast screen can still leave someone waiting for the application to close.

Installed viewer passed those cases and bounded shutdown. Any unflushed output is reported; no silent claim of a complete log.