Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Thu, Apr 21, 2016 at 05:41:40PM +0200, Dietmar Maurer wrote: > > - it accesses the drbdmanage control volume, which is a drbd resource > > itself, and where the cluster information is stored, via TCP/IP. The > > reason satellites exist is that currently a drbd volume is limited to > > 32 nodes. As the cluster information is stored on a drbd volume, that > > would limit the whole cluster to 32 nodes. That is avoided with > > satellite nodes that get their cluster configuration via TCP/IP, so > > you can keep adding a lot more than 32 nodes to you cluster. > > What happens to a satellite node if the corresponding control > node is offline? On startup the satellite node just waits for a connection. If the connection breaks, the satellite detects it and waits for a new connection. Can be the old control node or a new one (assign-satellite). > Why is there single a dedicated control node? Because it is the mode that is the easiest to implement. Which control nodes qualify to be the control node for a particular satellite? All? The ones in the same site? The ones with the best $insertyourfavoritenetworkingterm? Plus some minor, usual considerations about determinism of the heuristic and locking. Patches welcome... Regads, rck