[DRBD-user] How efficient is DRBD during Sync?

Digimer lists at alteeve.ca
Thu Oct 22 17:12:17 CEST 2015

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On 22/10/15 10:50 AM, Lars Ellenberg wrote:
> On Thu, Oct 22, 2015 at 01:03:10AM -0400, Digimer wrote:
>> On 21/10/15 03:57 PM, Christian Völker wrote:
>>> Hi all,
>>>
>>> just a simple question for my own curiosity.
>>>
>>> Assume a two node master-slave setup. Disk on the master fails. All
>>> reads and writes go through the slave (master remains primary).
>>>
>>> Now the disk on the primary is replaced and the full sync is started.
>>>
>>> Now let's say the sync is at 40% and a write occurs which fits into the
>>> not yet synced 60% of the disk.
>>>
>>> Ist the block now written in parallel to both nodes and marked on the
>>> primary as "already synced" or does the write affect only the slave and
>>> during the remaining sync the block is copied back to the primary?
>>>
>>> Just curious....
>>>
>>> Thanks!
>>>
>>> Chrischan
>>
>> Once connected, all new writes are synchronous, yes. Out of sync blocks
>> replicate at the resync rate. This is why setting a high resync rate
>> will slow down applications writing to DRBD because the bandwidth
>> available for new writes is total capacity minus capacity used by resync.
> 
> But yes, any write during resync to no-yet synced, "dirty" blocks
> will bring those blocks in sync as well.
> 
> As an extreme example, if you'd just zero-out the full device during
> resync, you'd bring it in-sync by application writes.

This 'zero-out' trick is exactly what I do for all new DRBD builds. It's
a great way, pre-production, to sync DRBD without mucking around trying
to force the normal resync rate up (which I wouldn't want in production
anyway).

-- 
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without
access to education?



More information about the drbd-user mailing list