Memory usage

[mazecrown]

I'm running some tests using the standalone server on a macbook with a simple server to try and determine memory usage:

use spacetimedb::{ReducerContext, Table, reducer, table};

#[table(name = my_row, public, )]
pub struct MyRow {
    #[primary_key]
    #[auto_inc]
    pub id: u32,
    pub i: i32,
    // pub i2: i32,
}

const NUM_ROWS: i32 = 10_000_000;

#[reducer(init)]
pub fn init(ctx: &ReducerContext) {
    for i in 0..NUM_ROWS {
        ctx.db.my_row().insert(MyRow {
            id: 0,
            i: i,
            // i2: 2 * i,
        });
    }
    log::info!("inserted {NUM_ROWS} rows");
}

Toggling between NUM_ROWS = 1_000_000 and 10_000_000 and also toggling i2 on and off I see the following results using activity monitor to monitor spacetime-standalone process:

Num rows	Num i32 in struct	Memory used
1m	1	197.5MB
1m	2	237.3MB
10m	1	1.02GB
10m	2	1.53GB

I know its not the most scientific of tests but it looks like it results in an extra 40-50 bytes per row just to add another i32. Is this to be expected? Are there any rules of thumb about how to calculate expected memory usage?

[bfops]

Hey @mazecrown, we don't have a "general" rule of thumb for estimating the total amount of RAM a SpacetimeDB instance will consume. It will definitely grow based on the amount of rows in your tables - are you running a server in a RAM-constrained environment?

[mazecrown]

It was mainly from the point of view of architecting and understanding how much RAM will be used up rather than any particular constraint at this time. It's my understanding that the max size of the DB is limited by the amount of RAM the server has and it cannot scale beyond that.