Howdie people! I'm looking into a FOSS solution for a client and have several directions I'm looking at.
the situation at the moment is a bank of Win2003 webservers with FOSS ODBC drivers all pointing at a single backend GNU/Linux server running MySQL. the idea is to both upgrade it to a more robust system as well as add HA in some way. the two new machines will be Xeon4 class, possibly with shared storage (at the moment we are discussin SCSI but FC is also an option). Downtime of 1-2 minutes for takeover is acceptable. The ideas I was thinking about: 1. one machine active, the other standing by, heartbeats with Linux-HA, and when the first machine faults, takes over storage and IP address, then runs the MySQL daemon. Cons: taking over storage is tricky, especially (I was told) with SCSI compared to FC, and I'm not using the other machine's bandwidth during standby 2. both machines active, each with some of the DBs, when node1 fails, node2 takes down the daemon, takes over the storage and restarts the daemon with the extra databases. Cons: clients need to know server IP was changed (presumably unsupported in current client), or a PAT/LVS should be put in between for routing the sessions to the active server, a bitof a headache, still problems with storage takeover 3. both machines are on OCFS or similar system that does not require to take over the storage, but still care has to be taken with database files themselves not accidently open by both instances at the same time. Cons: still got locking problems, still need to balance sessions externally or use one only in standby btw, has anyone tried the GPL OCFS for general stuff? 4. I install a MySQL cluster, which I've no experiance in yet. their website says it's a "share nothing cluster" in one place, but then talks about a common storage node in another. Either way it's not pre-packaged by any distributions, so I wanted to see if anyone can clear things up for me. can I build a two node cluster with both nodes serving as redundant storage nodes with or without shared storage? can/should both machines serve as API nodes as well, or only one at a time with IP takeover in my case? I found me an article from October LJ but I'd appreciate someone with hands-on experiance. Also I heard in a few circles that 4.1 still has the odd bug or two, some avoidable but others may still be discovered... http://mail.wikipedia.org/pipermail/mediawiki-l/2004-November/002245.html your thoughts appreciated! -- Just say no to Ira Abramov http://ira.abramov.org/email/ ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
