spurious walsender activity

[Andrei-Pozolotin]

1. say we have 2 separate databases:

• main_data - configured for replication, all tables are part of default replication set
• unit_test - not configured for replication at all

2. we observe the following spurious walsender activity:

• when a test client produces heavy insert activity into unit_test, say at 50% cpu load
• then all walsender processes representing main_data nodes show 5% cpu load each
• as expected, no data are transmitted, and no tables are updated on the main_data peer nodes

3. we realize that some process somewhere need to read wal and filter out proper events, but 5% cpu for each walsender processes, when insert happens into unrelated / non-replicated database seems like some omission in the replication design

[luss]

Hi Andrei,

Do you have a proposed patch that fixes the problem you have detailed. Sounds like, perhaps, we just need a high level if test somewhere to say "don't bother checking all of the details in this database if it is not part of a replication set".

--Luss

[Andrei-Pozolotin]

need a high level if test somewhere to say "don't bother checking

yes, that was exactly my thought as well