On 21/04/14 13:18, Michael Paquier wrote:
> When a standby connects for the first time to a primary, it is not yet
> synchronized, this is the "catchup" phase. Once the lag between the
> standby and the master is reduced to zero for the first time,
> replication state changes to "streaming".
Thanks.
I am seeing several standbys changing from streaming to catchup and
back. Sometimes they also get completely lost. This happens when the lag
becomes high, hundreds of MB or even GB. The standby servers are
relatively far away on the internet. And the operation to generate this
kind of lag is index creation on large tables.
Is there a difference in the protocol used in those phases? Maybe the
catchup phase is optimized for bulk throughput?
Torsten