Improve reliability of gossip convergence test.

[bdarnell]

Instead of increasing the cycle count to accommodate race-enabled
tests on slow hardware, increase the gossip interval so that
the test is no longer cpu-limited.

[spencerkimball]

LGTM

[bdarnell]

I tried to convert gossip to a manual clock but couldn't make it work (there's no good way to define a "cycle" with all the goroutines involved). Increasing the gossip interval is better than the cycle count because when the interval is too low, some goroutines just fall further and further behind. With this change it converges in 2-3 cycles pretty consistently on my laptop, race or non-race.

[spencerkimball]

Sounds good.

On Fri, Mar 6, 2015 at 11:21 AM, Ben Darnell [email protected]
wrote:

I tried to convert gossip to a manual clock but couldn't make it work
(there's no good way to define a "cycle" with all the goroutines involved).
Increasing the gossip interval is better than the cycle count because when
the interval is too low, some goroutines just fall further and further
behind. With this change it converges in 2-3 cycles pretty consistently on
my laptop, race or non-race.

—
Reply to this email directly or view it on GitHub
#383 (comment).