4 ms·
The first real time GC I encountered accomplished its goal by amortizing deletes across creates so the max allocation time could be guaranteed. After all these
by hinkley 3y ago
The first real time GC I encountered accomplished its goal by amortizing deletes across creates so the max allocation time could be guaranteed.
After all these years, I wonder why vacuum is still such a PITA with Postgres. Vacuuming should be at least partially concurrent with CREATE.
- ses1984 3y agoHow often are you executing create, compared to how often do you need to vacuum?
- hinkley 3y agoIf the data set isn't growing, how big of a priority is vacuuming? You're at homeostasis if you aren't creating or updating records. If memory serves the GC implementation was freeing/defragging 10 chunks of memory for each one it allocated. In a real time system if you're allocating a bunch of memory, you're already on notice, so having it cost a few thousand instructions instead of a few hundred isn't that big of an inconvenience. And with SQL the network time is going to dominate.
- ses1984 3y agoDid you mean insert…?