(Illustration by Gaich Muramatsu)
Afternoon all First - great product, thanks. Please forgive the longish post. I have installed coda on a small(ish) cluster of 2 mail servers, using our own MTA, and I have been using the coda filesystem as a method of replicating the 'pending' and 'quarantine' mail archive to provide a failover capability in the cluster. I have a few questions regarding the right way to do this: I have read the Admin manual but I am not particularly convinced that it is up to date in certain repects. I have built Slackware packages for all the components - anyone is welcome to them, or instructions. They are not hard to build. Here is a summary of the setup: 2xDell 1750 dual xeon, dual GB NICs, 2Gig RAM. Slackware linux 9.0 with kernel 2.6.3. Each server is configured as both client and server. Mount points are defaults suggested in *-setup scripts. I have a replicated volume working successfully (E0000104). The files are stored on a single (/vicepa) partition. The messages are stored as individual text files, some compressed, in a structure like <codadir>/<billions>/<millions>/<thousands>/messages There can be a lot. A server running a million messages per day can often be holding 10M-20M files. The server I am testing with handles a lot less - maybe 500000 or so but this could increase. The servers are in different geographical locations (one in central london and the other on the edge) with 100MBit link. 1. I would like to know the maxima for the current build, ie: max number of files per server, max files per partition - anything in fact which is a hard limit which may bite us. I chose the '16M' files per server in the setup, but found a changelog note referring to a max of 512K files per partition. Not sure how up to date any of this is. if the max is 512K per partition, can I rebuild for more? Clearly I am using coda in a slightly unusual way - there is no 'real' client. I have some questions that follow on from that. 2. Authentication: all went relatively smoothly until I got to the first authentication. The manual is obviously largely pre-realm - this confused me. Added to the fact that there is a bug in clog which means that the -h <authhost> is broken - any use of the -h swicth results in a usage message. In the end, creating a realm in /etc/coda/realms solved the problems there. I would also have liked to know a bit more about the relationship between system users and coda users - ie not very much, except that the username (and uid?) is used as a default for programs like clog, cpasswd and au. 3. As the machines require the client to be running and authenticated at all times, I have adopted the following strategy: I have created a script containing basically 'echo <passwd> | clog' which is run at startup and in a cron job once per hour (I know token lasts 25 hours but you never know). This works fine on the non-SCM machine, but on the SCM it almost never boots up authenticated. Running the auth script fixes it. Here is the order I do stuff: - init coda server: (auth, rpc2, *mon etc) - authentication script - init venus. Is this the right order? Is there likely to be a pause before auth2 is ready to authenticate me? Is this the right way to do it? 4. Will there be any break in (read/write) service while auth2 is re-authenticating during the cron job? I take it that permissions and so on are only evaluated when an actual disk operation is performed so unless the token is deleted during re-authentication I don't see why there should be a problem. 5. I see there are 'from file' and 'to file' switches in clog. Should I be using these for auto-authentication? How? 6. In my situation there is always a server running on the same box at the client. I could really do without the client cache as the client will (hopefully) never be disconnected form the server. Is there any existing way to shortcut this and avoid the extra disk writes to the client cache? 7. Why would I want to use cfsmount? is there any advantage to using this over linking to dirs in the the standard /coda directory? That's all for now. Thanks again Jim Email has been scanned for viruses and SPAM by Email Systems *** Email the way you want it ***Received on 2004-05-21 08:30:50