[Varnish] #642: log buffer (VSL) can get corrupted

[Varnish]

#642: log buffer (VSL) can get corrupted
------------------------+---------------------------------------------------
 Reporter:  Tv          |        Owner:  phk  
     Type:  defect      |       Status:  new  
 Priority:  normal      |    Milestone:       
Component:  varnishlog  |      Version:  trunk
 Severity:  normal      |   Resolution:       
 Keywords:              |  
------------------------+---------------------------------------------------
Comment (by Tv):

 I don't think it was caused by two varnishds using the same file, but that
 is possible. As far as I'm concerned (and because I didn't save a copy of
 the corrupted file), you can consider the corruption to have been caused
 by cosmic rays, kernel bugs and a bad disk, all conspiring to ruin my day.
 What I'm really concerned about is, this effectively DoSed an important
 service. Are the benefits of persistence big enough to not zap out the old
 content when starting varnishd? I can even imagine a scheme that unlinks
 the underlying file, leaving existing varnishlog readers to finish
 flushing it to permanent storage (and switch to the new file once they're
 done). I absolutely expected a restart to fix something that has been
 described as a "shared RAM buffer", and was actually surprised to realize
 it has a persistent backing file.

-- 
Ticket URL: <http://www.varnish-cache.org/ticket/642#comment:2>
Varnish <http://varnish.projects.linpro.no/>
The Varnish HTTP Accelerator