[ Re-send in plain text. ] On Tue, Aug 14, 2018 at 7:41 AM, Sage Weil <sweil@xxxxxxxxxx> wrote: > Currently we default to 0/5, which means we collect log messages on > messages sent and received and dump them on crash. The intent is that on > a crash we'll have some additional useful context. > > In practice, I can't remember any instances where this information has > been useful for debugging. And Radek's testing is suggesting we get quite > a bit of additional performance on the OSD side by turning this down. > > Has anybody else found this information to be useful in real world > debugging situations? I believe I've used it many times in the past, and I'm a bit surprised if you haven't. Seeing the actual messages between daemons is invaluable at establishing control flow! That said at level 5 we may collect several other things and we could perhaps turn the cost of the simple message prints a long way down. Or just keep it turned up in teuthology where we actually induce failures and let it be on production clusters. -Greg
