On 09/14/2010 12:38 AM, Graham Leggett wrote: > On 13 Sep 2010, at 8:47 AM, Ruediger Pluem wrote: > >> Can't this lead to a situation where buckets that follow an EOS bucket >> (the only ones >> I can think of are Metabuckets) get swallowed forever by mod_disk_cache? >> These possible Metabuckets will be swallowed and added to dobj->bb, >> but never put >> in the out brigade. So shouldn't we just CONCAT the remaining part of >> in to out and >> leave? > > We do need to do this, yes, thanks for catching this. Can you verify > r996713 and confirm that it fixes it? After seeing the eos bucket, we > enter a simple "passthrough" mode for all remaining buckets (and any > further buckets that might arrive) and exit.
I don't have a test case at hand, but it looks fine from inspection. BTW: Can't we avoid the loop if dobj->done and just concat the in brigade on the out brigade and leave? Regards Rüdiger
