Thanks for the pointer - I've commented on that issue now, but for the benefit 
of anyone following this thread, I managed to track this down to a bug in 
libbacktrace's allocation strategy which has thankfully been fixed upstream 
already. I've created a PR on rust to update this code 
(https://github.com/rust-lang/rust/pull/24344), which causes servo to crash 
with much less drama on linux :)


On Friday, April 10, 2015 at 12:47:55 AM UTC+10, Josh Matthews wrote:
> Yes, we noted this recently too - 
> https://github.com/servo/servo/issues/5340 has zero information about 
> it, but in particular this only occurs when RUST_BACKTRACE=1 is enabled, 
> suggesting that it's a problem in the stackwalking code on linux in the 
> Rust runtime.
> 
> On 2015-04-09 7:46 AM, [email protected] wrote:
> > While having a look into a layout-related panic, I found that sometimes 
> > when servo crashes, it takes _all_ my memory with it.
> >
> >   - I'm using linux (fedora 21)
> >   - I see it maybe 80% of the time in a debug build (the other ~20% are 
> > clean crashes which just exit after stacktrace)
> >   - I've never seen it do this in a release build
> >
> > I don't know whether this is servo, or whether (perhaps more likely) rust 
> > has some builtin panic-diagnostic behavior which is going out of control, 
> > but I thought I'd post here in case anyone has seen it before, or has an 
> > idea what might be causing it.
> >
> > For brevity, I've put the console output and relevant strace snippet into a 
> > gist:
> >
> > https://gist.github.com/gfxmonk/4c3a67ad42a521f43c88
> >
> > Basically, on the console I get a panic! stacktrace from the failure in the 
> > layout code. I then get a stacktrace from the servo constellation in 
> > RecvError, which seems fairly reasonable.
> >
> > After that however, servo (along with my entire X server) becomes 
> > unresponsive, and system monitor shows servo rapidly consuming 100% of RAM 
> > until I kill it.
> >
> >
> > Perhaps more usefully, if I use `ulimit` to only allow it a couple GB of my 
> > memory and run it under `strace` (relevant output included in the gist), I 
> > see that after it write()s the stacktrace strings to the console, the next 
> > thing it does is just to `mmap` the world in increasingly large quantities, 
> > until it runs out of memory and dies.
> >
> > I guess that series of syscalls could probably fit any number of runaway 
> > allocation scenarios, but perhaps someone has an idea what might be causing 
> > it, or how to narrow down the possible culprits a little? Does servo have 
> > its own crash diagnostics, or is this more likely the rust runtime 
> > misbehaving?
> >
> > Cheers,
> >   - Tim
> >
_______________________________________________
dev-servo mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-servo

Reply via email to