Re: [CAUTION: SUSPECT SENDER] RE: [CAUTION: SUSPECT SENDER] RE: BUG #19029: Replication Slot size keeps increasing while logical subscription works fine
Nantha kumar.T. <nanthad@gmail.com>
From: "Nantha kumar.T." <nanthad@gmail.com>
To: Thadeus Anand <thadeus@rmkv.com>
Cc: Amit Kapila <amit.kapila16@gmail.com>, "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>,
pgsql-bugs@lists.postgresql.org, vignesh C <vignesh21@gmail.com>
Date: 2025-08-27T06:30:59Z
Lists: pgsql-bugs
Hi Thadeus Anand, This spill over files are held by slots thinking that it need for sequential apply of transaction. Manually deletion may have impact on the logical streaming. I too had that thought but ours high critical production system and so I didn't have courage to proceed further. Rather, alter publication/subscription would work/help. If any have technical insights on this manual intervention, kindly please share which will help for our case too. On Wed, 27 Aug, 2025, 11:01 Thadeus Anand, <thadeus@rmkv.com> wrote: > Hi, > > Thank you everyone for your wonderful guidance. Really appreciate it. > > I have a couple of doubts. > > Can these spill files in the pg_replslot folder be deleted safely, if all > the subscribers have caught up? (I noticed last night when I restarted my > publisher, the spill files were removed automatically) > > Can I use ALTER SUBSCRIPTION to set streaming on, with my existing > subscriptions? > > Thadeus. > > > > > On Wed, Aug 27, 2025 at 10:06 AM Amit Kapila <amit.kapila16@gmail.com> > wrote: > >> On Wed, Aug 27, 2025 at 9:09 AM Hayato Kuroda (Fujitsu) >> <kuroda.hayato@fujitsu.com> wrote: >> > >> > Dear Thadeus, >> > >> > > I do not know or understand what a "spill" is. >> > >> > As I wrote in a previous mail, logical decoding (and logical >> replication) >> > sometimes writes changes into the disk to avoid using too-much memory, >> and the >> > ".spill" file is the fragment. This can happen when the total amount of >> decoded >> > changes exceeds logical_decoding_work_mem. >> > >> > > I will look it up. But the tables that are part of the >> > > publication are not updated as part of any huge transaction. They may >> be part of some other >> > > long running procedures >> > >> > Actually, logical decoding decodes all changes even if tables are not >> published, >> > and suppress outputting changes at commit phase. >> > >> >> One can avoid the generation of so many spill files for large >> transactions by using streaming=on or streaming=parallel while >> creating subscriptions. >> >> -- >> With Regards, >> Amit Kapila. >> >