tree a7c6b7c2de851fd1c6815782df020b54ca1bbf69
parent 09f8bcd78e3fb634ed69c66f46695b821a5b3822
author Georg Semmler <github@weiznich.de> 1642065785 +0100
committer Georg Semmler <github@weiznich.de> 1642065785 +0100
gpgsig -----BEGIN PGP SIGNATURE-----
 
 iQEzBAABCgAdFiEEt/EOFp+yWvA8Dk28qHvO5SBc5IkFAmHf8DkACgkQqHvO5SBc
 5ImbGQf+MPqf8+1j7irTRssmEiIV/W4aDxChC9mXD5oCVIjH2zhR6R9cJLkNLKWp
 HFPjABA/Ln+4H667bAFGS4jgGm/g49Pa1gbz5GXnsYa5qEGaAVuS4YQFvUdrp1yG
 RjvTAxM3nwB7eBFFqk+V8yrLvvbD2V7uXybe9EtxP8Lf740wKbG0dpyIKH1QYn/I
 TeuRj7slSxdD88IcpBmai7UyzMCflgxzkuqYE/YA2xvs56l3bUu097iRgePoHi/9
 +jtiM6eNgd3e5NX6etiXi+pu2HVMLVl61+IPG5SFYLr3gYCkng4bR6gWOqMb0Hmg
 DR9Uh0tGRGeAdbeodL04/WdJjulH8w==
 =EnhT
 -----END PGP SIGNATURE-----

Fix #3016

This demonstrate a potential fix for #3016 by using the async libpq
interface + busy polling. This allows us to abort polling as soon as we
reach the timeout.
I see multiple issues with this approach:
* Add's new methods to `Connection` (`get_timeout`/`set_timeout`). This
  could likely be avoided by exposing them directly on `PgConnection`.
On the other hand that would, at least for 1.4.x imply that we cannot
use those timeout only for the `is_valid` function. It would be all or
nothing then.
* This uses busy polling, which could be quite inefficient. I'm not sure
  what better posibilities there are without pulling in
tokio/mio/async_std.