Fix hang in full_mesh_connectivity

[andrewjstone]

[Properly use tokio::time::{pause,advance,resume};

Fixes #9377

[sunshowers]

Hmm, so in theory Tokio's pausable timer stuff has auto-advance built into it. Is it even necessary to manage timers manually?

[jgallagher]

Hmm, so in theory Tokio's pausable timer stuff has auto-advance built into it. Is it even necessary to manage timers manually?

IIUC, I think it is. The test wants to skip ahead by RECONNECT_TIME, hence the pause/advance/resume bit. But it can't only use auto-advancing timers, because that breaks the wait_for_condition checks that are willing to wait up to 10 seconds for various networking activity to complete. If tokio's clock is paused and it autoadvances, each wait_for_condition iteration auto-advances immediately, which doesn't give enough real clock time for the networking bits to do what they're supposed to do.

[andrewjstone]

If tokio's clock is paused and it autoadvances, each wait_for_condition iteration auto-advances immediately, which doesn't give enough real clock time for the networking bits to do what they're supposed to do.

At least on Linux, it's even worse than that. If I comment out all the existing stuff and just call tokio::time::pause() before the last wait_for_condition the test hangs. This is essentially what was happening before. Note that this is using std::time::Instant in wait_for_condition.

If I switch to using tokio::time::Instant in wait_for_condition, it does what John says and times out too quickly.

[sunshowers]

Gotcha, thanks!