[Tarantool-patches] [PATCH v1] test: fix hang of vinyl/select_consistency.* test

[Nikita Pettik]

On 02 Dec 21:57, Alexander V. Tikhonov wrote:
> Hi Nikita, thanks for the review. I've made all of your suggestions,
> please review below.
> 
> On Mon, Nov 16, 2020 at 09:06:54PM +0000, Nikita Pettik wrote:
> > On 15 Nov 21:43, Alexander V. Tikhonov wrote:
> > > 
> > > It happened because on heavy loaded hosts may occure the situation
> > > when the previous snapshot was inprogress when the new snapshot came
> > > with the same file name *.snap.inprogress. It happens before the
> > > current snapshot completed and printed "dump completed" in log file.
> > > Also this file *.snap.inprogress was seen left on manual debug, when
> > > the test hanged before this patch. To resolve the test issue fiber
> > > sleep delay after it can be increased, but we want to save the issue
> > > reproducable. The current patch corrects the test to avoid of hang on
> > 
> > I guess increasing sleep and fileing exact repro with 'bug' label is
> > enough to deal with this test.
> >
> 
> Ok, I've added instructions how to reproduce the issue in the issue and
> increased the sleep from 0.1 to 0.5.
> 
> > > box snapshot, to be able to continue testing after it failed. Fiber
> > > sleep was even decreased after adding fiber for box.snapshot to be
> > > able to reproduce the issue.
> > > 
> > > Needed for #4385
> > > ---
> > > @@ -75,8 +80,13 @@ end;
> > >  ...
> > >  function snap_loop()
> > >      while not stop do
> > > -        box.snapshot()
> > > -        fiber.sleep(0.1)
> > > +        local ok, err = fiber.create(function() local ok, err = pcall(box.snapshot) return ok, err end)
> > 
> > Why not simply wrap box.snapshot() in pcall? Why do you need another
> > one separate fiber for it?
> >

Could you please file new issue since it is not longer 'flaky test', but
rather vinyl bug? I mean this situation with simultanious instance
dumps seems to core issue. Also it would be nice to have reproducer out
of test-run environemt, so owner of the bug can run Tarantool, copy-paste
reproducer to the console and get mentioned errors.