Thread: VACUUM FREEZE and replication lag

Hello everyone,

We have a very big table in our pg92 database that requires a manual vacuum freeze in order to keep sane number of transaction ids available. Yesterday we did a vacuum freeze on this table that took roughly 9 hours. After performing the operation we got back roughly 0.5 billion transaction ids. I was wondering how much this operation can affect replication lag? If vacuum freeze needs to mark a lot of tuples to the xfrozenid, will this mean that a lot of informaton is going to be stremead in the way files? Just trying understand the relation. As soon as we started the operation, our standbys that have the lower io got significantly impacted for several hours. 

Thanks!