Thread

1. Re: Potential "AIO / io workers" inter-worker locking issue in PG18?

   Marco Boeringa <marco@boeringa.demon.nl> — 2025-10-10T21:26:38Z

   Just found out about the other thread, which seems to describe an issue 
   that may well be related to the problems I observe in my multi-threaded 
   Python implementation using PostgreSQL 18:
   
   https://www.postgresql.org/message-id/flat/CA%2BhUKG%2BHh7u3YpPLKQS%2BB7KJu0Bw99EjhuPuvBRVccDfgod95g%40mail.gmail.com#969d704180758e8dbd124604206175e0
   
   As I have zero experience compiling PostgreSQL from scratch and 
   PostgreSQL development in general, I will unfortunately avoid going down 
   the rabbit hole of such an attempt to be able to test the patch, but 
   will eagerly await PG18.1 which hopefully will have it included and 
   might fix the issues I observed.
   
   Marco
   
   
   
   
   

2. Re: Potential "AIO / io workers" inter-worker locking issue in PG18?

   Andres Freund <andres@anarazel.de> — 2025-10-10T21:30:20Z

   Hi, 
   
   On October 10, 2025 5:26:38 PM EDT, Marco Boeringa <marco@boeringa.demon.nl> wrote:
   >Just found out about the other thread, which seems to describe an issue that may well be related to the problems I observe in my multi-threaded Python implementation using PostgreSQL 18:
   >
   >https://www.postgresql.org/message-id/flat/CA%2BhUKG%2BHh7u3YpPLKQS%2BB7KJu0Bw99EjhuPuvBRVccDfgod95g%40mail.gmail.com#969d704180758e8dbd124604206175e0
   >
   >As I have zero experience compiling PostgreSQL from scratch and PostgreSQL development in general, I will unfortunately avoid going down the rabbit hole of such an attempt to be able to test the patch, but will eagerly await PG18.1 which hopefully will have it included and might fix the issues I observed.
   
   If that issue were causal you'd see something in pg_aios and the issue wouldn't be present with worker or sync io methods. Nor would you see high cpu utilization. This is a different issue, likely not aio related.
   
   Greetings,
   
   Andres 
   -- 
   Sent from my Android device with K-9 Mail. Please excuse my brevity.
   
   
   
   

3. Re: Potential "AIO / io workers" inter-worker locking issue in PG18?

   Marco Boeringa <marco@boeringa.demon.nl> — 2025-10-10T21:36:37Z

   Then as follow up on that: is there anything you still need from me, or 
   is the information I gave so far including the last requested answers 
   and data of perf, enough to give you at least a decent start to figure 
   out where the issue is?
   
   Op 10-10-2025 om 23:30 schreef Andres Freund:
   > Hi,
   >
   > On October 10, 2025 5:26:38 PM EDT, Marco Boeringa <marco@boeringa.demon.nl> wrote:
   >> Just found out about the other thread, which seems to describe an issue that may well be related to the problems I observe in my multi-threaded Python implementation using PostgreSQL 18:
   >>
   >> https://www.postgresql.org/message-id/flat/CA%2BhUKG%2BHh7u3YpPLKQS%2BB7KJu0Bw99EjhuPuvBRVccDfgod95g%40mail.gmail.com#969d704180758e8dbd124604206175e0
   >>
   >> As I have zero experience compiling PostgreSQL from scratch and PostgreSQL development in general, I will unfortunately avoid going down the rabbit hole of such an attempt to be able to test the patch, but will eagerly await PG18.1 which hopefully will have it included and might fix the issues I observed.
   > If that issue were causal you'd see something in pg_aios and the issue wouldn't be present with worker or sync io methods. Nor would you see high cpu utilization. This is a different issue, likely not aio related.
   >
   > Greetings,
   >
   > Andres