On Saturday 01 August 2015 03:59:28 Gene Heskett wrote: > On Saturday 01 August 2015 03:26:55 Jon LaBadie wrote: > > On Fri, Jul 31, 2015 at 10:38:54PM -0400, Gene Heskett wrote: > > > On Friday 31 July 2015 17:06:30 Jon LaBadie wrote: > > > > On Sun, Jul 19, 2015 at 04:21:29AM -0400, Gene Heskett wrote: > > > > > On Saturday 18 July 2015 10:40:38 Gene Heskett wrote: > > > > > > Only on THIS machine, the remotes being backed up over the > > > > > > cat5 always work. > > > > > > > > > > > > etimeout was 600, made it 1800 > > > > > > dtimeout was 1800, made it 2400 > > > > > > > > > > > > uptime is 7 days & small change, 135 megs into swap on an > > > > > > 8Gb equipt machine. > > > > > > > > > > > > A reboot fixes it, for a few days, then once its started, > > > > > > only a reboot seems to fix it. > > > > > > > > > > > > The only clue might be fromn a grep for a "code 1", in > > > > > > the /tmp/amanda-dbg/amdump############# files > > > > > > > > > > > > root@coyote:/tmp/amanda-dbg/server/Daily# grep "code 1" * > > > > > > amdump.20150717030105.debug:Fri Jul 17 09:01:06 2015: > > > > > > thd-0x962a400: amdump: planner finished with exit code 1 > > > > > > amdump.20150718030105.debug:Sat Jul 18 09:01:06 2015: > > > > > > thd-0x980c400: amdump: planner finished with exit code 1 > > > > > > > > > > > > Is that 10 minutes for the estimate per DLE, or a global for > > > > > > the whole thing. I thought it was per DLE? > > > > > > > > You have probably already checked, but the etimeout > > > > is "per estimate". There could be up to 3 estimates > > > > per DLE. > > > > > > > > dtimeout is listed as being per "disk". I'm pretty > > > > sure in this context "disk" == "DLE". > > > > > > > > For the estimates, are you using the default "client" > > > > type of estimate rather than "calcsize" or "server". > > > > If yes, is there a reason you must use "client" style? > > > > You might try the much faster "calcsize" or the almost > > > > instantaneous "server" estimates. > > > > > > > > Jon > > > > > > And where is that set, Jon? > > > > > > Cheers, Gene Heskett > > > > It is a dumptype parameter: > > > > estimate [ client | calcsize | server ]+ > > > > jon > > I just looked, I have no dumptype that specifies the estimate in any > fashion. Is the client "survey" the estimate default method? > > Cheers, Gene Heskett
I probably should explain that the amanda.conf in use is NOT an edited copy of the one in 3.3.7p1, its a good decade older than that one. So its comments in the dumptype header do not and never have included any estimate type instructions. I have known, from reading this list all these years since 1998 or 1999, that it was selectable somehow, but client, the default, was working well so I never played with it. The point I am trying to make is that when it works, it works plenty fast enough that it is not a problem. When it fails, no amount of estimate time up to 2400 seconds makes any difference, all amanda processes are simply hung, using zero cpu, until whatever etimeout has been set to, so a setting of 2400 seconds means I don't get the emailed report till noonish the same day, a 10+ hour lag. When it works, etimeout could be set as low as 5 minutes, or 300 seconds. So the problem is NOT etimeout related, it is whatever causes the client estimates, only on this machine, to be blocked from using any cpu at all. They are all sitting at the very last in a cpu sorted htop display, using zero cpu. And I'm now aware why I don't run this particular kernel, a 3.2.0-4amd64, it just froze kmail for about about 5 minutes but has now recovered. But I thought I'd boot it in place of the 3.4.something I normally run to see if it made any difference to amanda. I have a 3.14-somethng but it segfaults and hangs less than a second into the boot. Posting that to vger a year ago got a "deal with it" response. I've got better things to do than screw with a screwed up kernel configuration that precludes any semblance of maintaining a working config. A make oldconfig now has a mind of its own, even going so far as to build an intel only kernel while running on a quad core phenom! It makes zero use of an existing old config. Anyway, I'll run this one till something forces a reboot just to see if amanda hangs. Cheers, Gene Heskett -- "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Genes Web page <http://geneslinuxbox.net:6309/gene>
