We are also getting similar warning messages in the log file, for Insert operation as it is blocking concurrent inserts on the same table. As per the online documents, I have come across, suggest is because the Postgres process takes time to search for the relevant buffer in the shared_buffer area if shared_buffer is too big.
In the highly transactional system, there may not be enough free buffers to allocate for incoming transactions. In our case allocated shared buffer is 24GB and has RAM 120GB, not sure whether we can call it too big but while querying pg_buffercache has always given indication that 12-13GB shared_buffers would be appropriate in our case. I have used the below URL to evaluate the shared buffer sizing.
On Mon, Oct 5, 2020 at 10:14 PM Michael Lewis <mlewis@entrata.com> wrote:
What is relation 266775 of database 196511? Is it cms_c207c1e2_0ce7_422c_aafb_77d43f61e563.cms_item or some system catalog table?
When I search google for "ExclusiveLock on extension of relation" I find one thread about shared_buffers being very high but not big enough to fit the entire data in the cluster. How much ram, what is shared buffers and what is the total size of the database(s) on that Postgres instance?
