tag:blogger.com,1999:blog-9149523927864751087.post2757449613285361112..comments2024-03-26T09:43:01.052-07:00Comments on Small Datum: Losing it?Mark Callaghanhttp://www.blogger.com/profile/09590445221922043181noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-9149523927864751087.post-51679152227433541072015-10-15T23:38:30.100-07:002015-10-15T23:38:30.100-07:00DRBD has three modes. C requires a sync to disk o...DRBD has three modes. C requires a sync to disk on the remote node, and an ACK over the network. This is the slowest mode. B mode is like mysql semi-sync. The write has to make it into the buffer on B, and B has to send an ACK. In C mode, writes are async. There is no ack, and writes proceed immediately. This is normally a DR node.<br /><br />If using A or B, you probably want a serial cable (or whatever) for heartbeat, and be sure to configure STONITH.<br /><br />Only one side can mount the DRBD device at once, unless you use a cluster filesystem. Crash recovery is required after the system comes back up, so it can take awhile, which can really bites into the 9sJustin Swanharthttps://www.blogger.com/profile/08193089637089861226noreply@blogger.comtag:blogger.com,1999:blog-9149523927864751087.post-775380138810418942015-10-12T08:46:38.826-07:002015-10-12T08:46:38.826-07:00I don't have any experience with DRBD. How far...I don't have any experience with DRBD. How far apart can the disk pair be so that I don't lose data when I lose a rack or consecutive racks?Mark Callaghanhttps://www.blogger.com/profile/09590445221922043181noreply@blogger.comtag:blogger.com,1999:blog-9149523927864751087.post-1854391964034472142015-10-12T04:24:20.120-07:002015-10-12T04:24:20.120-07:00Nice summary Mark.
I can see another implementatio...Nice summary Mark.<br />I can see another implementation of "cannot lose visible, cannot lose acknowledged": shared storage (DRBD or filer with high durability enabled). This one comes with the drawback of suffering crash recovery on MySQL startup. When comparing "lossless semisync" to shared storage, the network round-trip on commit does not look that bad.Jean-François Gagnéhttps://www.blogger.com/profile/12267071794432977020noreply@blogger.com