varnish panic investigation information

[Balder VC]

Hi

I'm requesting some help or info for a panic we had.

First a small token of appreciation and some background.
We had our varnish-cache (varnish-4.0.3 revision b8c4a34 )  filled with
around 4.6 billion ( 4.6KK ) page entries on a 90GB RAM machine.  !
Great stuff, still responding fast and furious! Thank you varnish!

Varnish has 7 back ends configured. At some point they started failing
till they all broke down. As expected varnish does the job it's need to
do. It responds with an error page if the requested entry is not in the
cache, serves a cache paged ( that is either in grace mode or just
cached ) when it can.

After a certain time the varnish child process panicked and restarted
and we could not serve the cached pages any more since the were gone.

Assert error in VDI_GetFd(), cache/cache_dir.c line 111:

  Condition((d) != NULL) not true.


see http://pastebin.com/H1ZdTyPd for the full panic.show log

Now, there is one url that is served differently. Via a pipe, I show
this because it seems from the panic log that this has something to do
with it.

    if (req.url ~ "^/content/" ) {

      return (pipe);

    }





I perhaps thought that this pipe might have caused the panic due to no
backends being available to pipe too ? And that this could be a bug?
Or if not, perhaps we are doing something wrong?

Let me know if you want more information.


Kind Regards

Balder VC