Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
I'm about to set up a luke-warm standby server - with DRBD and/or another form of synchronisation. The server will be running a general-purpose web hosting environment (LAMP + mail) for up to 100 domains. A second server will be at another data centre, ready to take over, should the first fail - either in terms of network infrastructure or hardware. The switchover will be done manually with a script and a DNS update by a real person. The bandwidth is relatively expensive, so I will try to keep the synchronised data to a minimum (i.e. no syncing /tmp, /var/run ...). The data would be a partition of around 100Gb. Once the system is in place, we won't be able to economically do a full sync - ever. (For the price of the bandwidth of a full sync we could hire two drivers for a month to ferry the disk between data centres). I am considering running DRBD over openvpn with lzo compression. Is DRBD an overkill for this application? Will it work? Will it work well? I would be grateful to hear recommendations on best practices for this kind of arrangement. Some things I've wondered about ... - How it worked for you ... - Whether splitting the volume into sub-volumes would be sensible - The best filesystem (ext3?) - Which files would be better rsync'd (or that the entire job should be) - Excluding database files from drbd, and using mysql active-active replication for those - Whether temporarily turning off synchronisation (drbdadm disconnect all) would be a good idea when the load is high, since that would somewhat reduce latency. - Tuning for better bandwidth usage for the common case of disconnects due to network interuption.... while not killing the disk to too-regularly update its (internal) meta-data. I also wondered if it would be feasable to take lvm snapshots on the drbd secondary ...