@cryptotaxi247 / infra / commits / 20a8da6b

hydra-scaler: stop autoscaling arm64

Cole Helbling committed Feb 17, 2024 at 12:18 UTC 20a8da6bc35c32f9fa449ffb59b717794175a857
1 file changed +11 -26
delft/hydra-scaler.nix
+11 -26
@@ -13,32 +13,17 @@
13 facilities = ["any"];
14 tags = ["hydra"];
15 categories = {
16 - aarch64-linux = {
17 - bigparallel = {
18 - divisor = 24;
19 - minimum = 1;
20 - maximum = 3;
21 - plans = [
22 - {
23 - bid = 2.0;
24 - plan = "c3.large.arm64";
25 - netboot_url = "${netboot_base}/c3-large-arm--big-parallel";
26 - }
27 - ];
28 - };
29 - small = {
30 - divisor = 4000;
31 - minimum = 0; # we have one static 80-core in addition
32 - maximum = 3;
33 - plans = [
34 - {
35 - bid = 2.0;
36 - plan = "c3.large.arm64";
37 - netboot_url = "${netboot_base}/c3-large-arm";
38 - }
39 - ];
40 - };
41 - };
16 + # NOTE(cole-h): We don't autoscale arm64 anymore because EM asked us not to: the arm64 spot
17 + # market appears to be a little funky as of this comment (we would commonly spin up a
18 + # machine, only for it to be reclaimed before it was even able to boot into NixOS and run
19 + # even 1 build for Hydra).
20 + # As of 17 Feb 2024, we have 2 dedicated arm64 machines on EM -- one `small`
21 + # (`small-c3.large.arm64`) and one `big-parallel` (`big-parallel-c3.large.arm64`). Hopefully
22 + # this will be an improvement over "maybe we have no arm64 machines at all because they spin
23 + # down before they can do any work".
24 + # The netboot URL for arm64 big-parallel is: https://netboot.nixos.org/dispatch/hydra/hydra.nixos.org/equinix-metal-builders/main/c3-large-arm--big-parallel ("Always PXE" enabled, "hydra" tag)
25 + # The netboot URL for arm64 small is: https://netboot.nixos.org/dispatch/hydra/hydra.nixos.org/equinix-metal-builders/main/c3-large-arm ("Always PXE" enabled, "hydra" tag)
26 +
27 x86_64-linux = rec {
28 bigparallel = {
29 divisor = 16;