On Tue, 2020-08-25 at 15:44 +0300, Achilleas Mantzios wrote: > Hello Keisuke > On 25/8/20 9:50 π.μ., Keisuke Kuroda wrote: > > Hi All, > > > > There was a similar problem in this discussion: > > Logical decoding CPU-bound w/ large number of tables > > https://www.postgresql.org/message-id/flat/CAHoiPjzea6N0zuCi%3D%2Bf9v_j94nfsy6y8SU7-%3Dbp4%3D7qw6_i%3DRg%40mail.gmail.com > > > > > RelfilenodeMapHash from 1024 entries to 64. > > The above changes reduced the performance impact. > > > > However, I think the problem that there are too > > many invalidations of RelfilenodeMapHash still remains. > > As you report, when many tables are created/dropped/truncated, > Right! IIRC also a massive drop created the very same problem during the latest attempts for reproduction but I did not focus on this scenario. > > The walsender process can get stuck. Thanks you both, that is indeed the same problem, and the linked thread helps understand the problem. Yours, Laurenz Albe
