(Illustration by Gaich Muramatsu)
i was trying to do the same thing as you, valerio, and had to give up as well (i was clearly missing something in the installation/configuration.) however, the conclusion that i came to was that: 1> both webservers would be file servers (to achieve real-time mirroring) 2> both webservers would have coda clients pointing to /themselves/. that way, both servers would automatically mirror each other (with only one of them serving as the SCM, of course,) but would be fault tolerant, in the event that one of them (even the SCM,) failed. my hopes were raised that this would work after i read about a company that actually built an LVS web server cluster appliance using this technique. however, i had to give up on installing coda, as well - i just couldn't get the clients to mount reliably, the servers to talk to each other, etc. obviously i was doing something wrong, but i too noticed that the documentation wasn't completely maintained - it seemed out of date, and in some cases clearly out of sync with recent releases (config files in different locations, different formats, etc.) i think adoption of coda would be *greatly* increased if this one area were improved. i also appreciate the fact that few people (one person?) maintain coda, and they seem to be pretty well loaded down. is there any discussion of the possibility of opening up maintenance of the documentation (via a coordinated effort with the development team, not anecdotal advice,) or is there currently a good source of documentation outside of the coda.cs.cmu.edu site? i didn't find any. i was (and am still,) curious about performance data when using the above technique. is there significant overhead or lag when a coda client accesses a local repository, assuming that the access is 99% read, and essentially random (as is the case with many web servers, especially in the asp space.) does this even work? ::ja Jason bond pratt lead infrastructure architect ________________________ ePrize 34405 w twelve mile rd, ste 123 farmington hills, mi 48331 p 248.848.1720x122 f 248.848.1905 jason_at_eprize.net -----Original Message----- From: Valerio Morettini [mailto:swan_at_multiplayer.it] Sent: Friday, November 16, 2001 4:14 PM To: Jan Harkes Cc: codalist_at_coda.cs.cmu.edu Subject: Coda and webserver mirroring? > I also believe you've got the whole idea of how the replication works > wrong. > > The way you've configured it, /usr/www (normally /vicepa) will store a > directory tree of anonymous file data objects. The server's RVM will > hold associated metadata and this data and metadata is combined by the > client and exported through the /coda mountpoint. Ok, i messed up the whole thing inside my poor head :) So the only way for doing replication is actually having the webservers as coda clients? If i've not misunderstood again, this means that the replication is not full, but happens only on the data hoarded by the clients... so if the server goes down and a web user requests a file which had not been hoarded, the webserver cannot serve it to him...and this is not very good for websites :( May i ask you an advice on a possible real-time solution? (no rsync or such...) Anyway, thanks for the answer, you were way too kind. I deserved to be flamed really :) ValerioReceived on 2001-11-16 16:35:50