Re: 15 pg_upgrade with -j

Jeff Ross <jross@xxxxxxxxxxxxxx> · Tue, 23 May 2023 08:38:53 -0600



    On 5/22/23 5:42 PM, Tom Lane wrote:

    
      Jeff Ross <jross@xxxxxxxxxxxxxx> writes:

      
        On 5/22/23 5:24 PM, Adrian Klaver wrote:

        
          So is the 1400G mostly in one database in the cluster?

        
        Yes, one big database with about 80 schemas and several other smaller 
databases so -j should help, right?

      
      AFAICT from a quick look at the code, you won't get any meaningful
parallelism unless you have several large DBs and/or several large
tablespaces.  It looks like the assumption was that issuing link()
requests in parallel wouldn't help much but just swamp your disk
if they're all on the same filesystem.  Maybe that could use
rethinking, not sure.

			regards, tom lane


    Thanks Tom.  These are all smokingly fast SSDs so it would be
    interesting to see how well they'd hold up under that load.

    
    Jeff