I can't remember the rule. I think it was 3gb and 15 minutes. In order to get under 3 gb, I think we need to run single threaded. Last I checked, running with 4 threads on dedicated hardware completes in ~12 minutes. However, the Travis instances used to be really slow virtual machines. I'm sure a solution can be found but I think we'd need some concerted effort on reducing the test footprint.
We talked before (Daniel's suggestion) about treating more of the tests as integration tests. This would help as much of the test time is spent starting and stopping Drillbits for each test class. If we only did this once for all those tests, the footprint would be much smaller. On Wed, Jun 24, 2015 at 8:37 AM, Ted Dunning <[email protected]> wrote: > On Wed, Jun 24, 2015 at 11:35 AM, Jacques Nadeau <[email protected]> > wrote: > > > We tried Travis before. The problem is that travis's nodes aren't > > substantial enough to complete our test suite within their timeout. > > > > Haven't the tests been substantially improved since then? > > Can the tests be segregated into pieces so Travis can still do some useful > work? >
