chore: configure slurmcluster per-cpu memory (#974)

determined-ai · Apr 18, 2024 · 64aa59a · 64aa59a
1 parent 5951c01
commit 64aa59a
Show file tree

Hide file tree

Showing 4 changed files with 13 additions and 2 deletions.
diff --git a/tools/slurm/README.md b/tools/slurm/README.md
@@ -16,6 +16,11 @@ Once the VM is created and SSH connection is established, one can also directly
 gcloud compute ssh --zone <zone_name> <instance_name> --project <project_name>
 ```
 
+For example:
+```
+gcloud compute ssh --zone us-west1-b phillipgaisford-dev-box --project determined-ai
+```
+
 [1] It is fine to exit this and restart it with `make slurmcluster` again as you please.
 
 To see usage of the `make slurmcluster` target, run `make -C tools/slurm usage`.

diff --git a/tools/slurm/packer/README.md b/tools/slurm/packer/README.md
@@ -6,6 +6,12 @@ documentation for interacting with and building this code. To build, you will ne
 
 It was last built with `Packer v1.8.6`.
 
+## Packer Initialization
+
+This needs doing once:
+
+`packer init tools/slurm/packer`
+
 ## How to Build an Image
 
 After the pre-requisite software is installed, one can run `make build WORKLOAD_MANAGER=[type]` where `type` is either `slurm` or `pbs` (default value is `slurm`) to build a SLURM or PBS image, respectively. Upon succesful completion of the build, the image name will be placed in the appropriate value in `../terraform/images.conf` (either `slurm` or `pbs` depending on what was specified).

diff --git a/tools/slurm/packer/static/slurm.conf b/tools/slurm/packer/static/slurm.conf
@@ -82,7 +82,7 @@ Waittime=0
 #
 #
 # SCHEDULING
-#DefMemPerCPU=0
+DefMemPerCPU=3072
 #MaxMemPerCPU=0
 #SchedulerTimeSlice=30
 SchedulerType=sched/backfill

diff --git a/tools/slurm/terraform/images.conf b/tools/slurm/terraform/images.conf
@@ -1,2 +1,2 @@
-slurm: det-environments-slurm-ci-1689266288
+slurm: det-environments-slurm-ci-1690569603
 pbs: det-environments-pbs-ci-1689269947