the real problem here is not that data will be lost if the ip/hostname pair changes. the real problem is that the data forks to two different places.
gmetad is currently not smart enough to know that a fork occurred. i don't see a simple way to make it smart. here's one way... gmetad saves a dbm(or xml) with the hostname/ip tuples. these files persist on disk even though gmetad is stopped/started. if there is a mismatch with previous tuples then gmetad will merge/move database files to deal with the changes. having this persistent database.. it really wouldn't matter if we used hostname or ip address in the RRDs filesystem. --- we have to have a way to know when the hostname/ip matched up changed really. if we simply key on ip address, it would cause other problems. when you remap ip/hostname pairs in a network overhaul.. it's very possible to have ip gets swapped (not just changed for a single host). in that case the data forks and you never know. -matt On Tue, 2004-02-24 at 08:29, Brooks Davis wrote: > On Tue, Feb 24, 2004 at 09:06:10AM +0100, Leif Nixon wrote: > > Brooks Davis <[EMAIL PROTECTED]> writes: > > > > > Since this will require a conversion script anyway, would it be > > > possiable to move away from using hostnames in the data store entierly? > > > It has the weird effect that if you don't have a hosts entry in > > > place the first time the host comes on, it gets stuck with an IP > > > address instead of a name. > > > > Hm, yes, I can see that is a problem. On the other hand, I think the > > hostname is a more permanent attribute of a machine than its IP > > address. It's nice being able to change a machine's IP address without > > losing its history in Ganglia. > > > > (We're in the middle of a major network reorganization here. You tend > > to appreciate little things like this after a while...) > > If ganglia were smart enough to switch to using the hostname once it > knows it, I would agree, but it's not and fixing it would be hard given > that name is used to make the directory the data is stored in. I'm > obviously biased in this, but I think that having to run a script to > remap your directories during a network reogranization isn't a big deal > since it had better be a rare event if you're ever going to get any work > done. > > -- Brooks -- mobius strippers never show you their backside pgp fingerprint A7C2 3C2F 8445 AD3C 135E F40B 242A 5984 ACBC 91D3
signature.asc
Description: This is a digitally signed message part
