Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello world, this is my first post here, please appreciate if I ask something trivial. Tried to dig into - drbd pages - Sybase knowledge base articles - google - this list's archive before this post, with no answer to the question. We envisioned a HA solution of a pair of Linux processors with one disk attached to each, running Sybase database software components. We call them 'Primary' and 'Replica', users are connected to Primary only. Primary is running Sybase ASE. The logfile of the database is mirrored with drbd (Protocol C) to the disk of the Replica. The logfile both on the Primary and the Replica is on a raw device. This way we mimique two-phase commit. Here the tricky part begins: although there is neither any filesystem on top of this raw device on the Replica, nor is it mounted in any way, we still read data from it while it is being written by drbd. Sybase has a component designed for it, the Mirror Replication Agent. We've put a lot of effort to prototype this solution and seems to work with both 0.6 and 0.7 releases. Also my feeling is that we don't do anything that is clearly not recommended by drbd, especially described here, (second question): http://www.drbd.org/316.html. Still this approach raised some concerns within the team by saying it's dangerous to do both writes and reads to the same log and that drbd is not designed for this. Partly just to be on the safe side, partly to collect arguments to be able to defend this architecture (the concerns bubbled up very-very late in the project, any deviation from the envisioned architecture would likely mean forgetting the deadline) I kindly ask you to share any concerns, success stories or whatever pieces of information you feel might be relevant. Any report of a successful read attempt would be much appreciated, don't filter out the non-Sybase ones, please. Here drbd is questioned in general. Thank you in advance, Peter Princz