On Wed, Sep 13, 2017 at 11:42 AM, Matthew Huff <[email protected]> wrote: > Both should have been similar. > > In the first case we lost power to all of our BGP border routers that are > peered with the upstream providers > In the second case, I did an explicit “shut” on the interface connected to > the upstream provider that appeared “stuck” after an hour after the outage. >
oh, I had thought when you broke the second time intentionally you might have shut the bgp session (and then maybe the interface too) ... causing a different semantic for the withdrawals in the isp network. it's possible that if the load of updates was large enough for the ISP(s) in question that things simply took a long while to process. In a recent similar situation we'd observed updates/convergence taking upwards of 30 minutes to trickle through the global system :( > > From: <[email protected]> on behalf of Christopher Morrow < > [email protected]> > Date: Wednesday, September 13, 2017 at 10:58 AM > To: Matthew Huff <[email protected]> > Cc: nanog2 <[email protected]> > Subject: Re: Reliability of looking glass sites / rviews > > > > On Wed, Sep 13, 2017 at 5:30 AM, Matthew Huff <[email protected]<mailto: > [email protected]>> wrote: > This weekend our uninterruptible power supply became interruptible and we > lost all circuits. While I was doing initial debugging of the problem while > I waited on site power verification, I noticed that there was still paths > being shown in rviews for the circuit that were down. This was over an hour > after we went hard down and it took hours before we were back up. > > explicit vs implicit withdrawals causing different handling of the problem > routes? > > I worked with our providers last night to verify there weren't any hanging > static routes, etc... We shut the upstream circuit down and watched the > convergence and saw that eventually all the paths disappeared. Given what > we saw on Saturday, what would cause route-views to cache the paths that > long? Some looking glass sites only show what they are peered with or at > most what their peers are peered with, that's why I've always used > route-views. > > What looking glass sites other than route-views would people recommend? > > ripe ris. >
