Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Tue, Sep 04, 2012 at 12:58:48AM +0800, Mia Leung wrote: > Hi All: > > Think about the following scenario in two node cluster with drbd. > > > > 1. A is primary, B is secondary, drbd0 created on a LVM volume > > 2. During A doing quick sync to B , A is crashed > > 3. The data on B is inconsistent , to keep the service running, B will > merge the LVM snapshot before start service > > 4. When A rejoin the cluster, A will resync its data from node B and > discard its modified data. > > > > As we known, on step 4 , A should do a full resync from B, because A's > bitmap is modified in step 2. > > > > But A full sync will take us a very long time because the data size is up to > T level. Are there any solution to do a quick sync? By low-level manipulating the drbd meta data in the right way, sure. Or by merging that snapshot data through DRBD, instead of below it, so it will be able to track those changes in its bitmap. But, if that was a so catastrophic failure that you decide to go online with a "stale" snapshot, maybe you should rather do the full sync anyways, to increase confidence levels? -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.