Schinckel.net

I happened to come across Speeding up Django unit tests with SQLite, keepdb and /dev/shm today. I can’t quite do that, because we use a bunch of Postgres specific things in our project, but it did make me think “Can we run Postgres on a RAM disk?”

Following up on this, it turns out that using a TABLESPACE in postgres that is on a RAM disk is a really bad idea, but it should be possible to run a seperate database cluster (on a different port) that could have the whole cluster on the database.

It’s possible to create a RAM disk on macOS, and init a cluster there:

#! /bin/sh

PG_CTL=/Applications/Postgres.app/Contents/Versions/10/bin/pg_ctl

new_disk=$(hdid -nomount ram://1280000)
newfs_hfs $new_disk
mkdir /tmp/ramdisk
mount -t hfs $new_disk /tmp/ramdisk
mkdir /tmp/ramdisk/pg

PG_CTL initdb -D /tmp/ramdisk/pg

This needs to be run once, after bootup (and probably before Postgres.app starts it’s servers). You’d also need to do the setup in Postgres.app to show where it is (alternatively, you could just start it on the port you want in the script above).

But, is it faster? Let’s look at the results. The “Initial run” is the first run of tests after deleting the test database.

So, it’s probably not actually worth it. I’m guessing that the variance is more to do with just how busy my machine was at each of the times I ran the test.