Skip to content

Commit

Permalink
bug 1907211: use europe-west4 instances in all preemptible translatio…
Browse files Browse the repository at this point in the history
…ns GPU worker pools (#31)

Our small test worked out well, and the cost is only marginally more than the US instances. Let's open up some more capacity here.

We're getting the GPU quota raised for this region in https://mozilla-hub.atlassian.net/browse/RELOPS-1011, so let's hold off merging this until that's dealt with, on the off chance we don't get the new limit we'd like.
  • Loading branch information
bhearsum authored Jul 17, 2024
1 parent c21b995 commit 64956fa
Showing 1 changed file with 14 additions and 5 deletions.
19 changes: 14 additions & 5 deletions worker-pools.yml
Original file line number Diff line number Diff line change
Expand Up @@ -1584,7 +1584,10 @@ pools:
maxTaskRunTime: 2592000
enableInteractive: true
minCapacity: 0
maxCapacity: 97
# We use 4 GPUs per instance across 4 regions with a limit of 128
# per region at any given time. 4 regions * 4 GPUs = 512 total GPUs
# 512 GPUs / 4 per instance = 128 instances possibly running at once.
maxCapacity: 128
implementation: generic-worker/worker-runner-linux
regions: [us-central1, us-west1, us-east1, europe-west4]
image: monopacker-translations-worker
Expand Down Expand Up @@ -1615,9 +1618,12 @@ pools:
maxTaskRunTime: 2592000
enableInteractive: true
minCapacity: 0
maxCapacity: 96
# We use 4 GPUs per instance across 4 regions with a limit of 128
# per region at any given time. 4 regions * 4 GPUs = 512 total GPUs
# 512 GPUs / 4 per instance = 128 instances possibly running at once.
maxCapacity: 128
implementation: generic-worker/worker-runner-linux
regions: [us-central1, us-west1, us-east1]
regions: [us-central1, us-west1, us-east1, europe-west4]
image: monopacker-translations-worker
instance_types:
- minCpuPlatform: Intel Skylake
Expand Down Expand Up @@ -1713,9 +1719,12 @@ pools:
maxTaskRunTime: 2592000
enableInteractive: true
minCapacity: 0
maxCapacity: 96
# We use 4 GPUs per instance across 4 regions with a limit of 128
# per region at any given time. 4 regions * 4 GPUs = 512 total GPUs
# 512 GPUs / 4 per instance = 128 instances possibly running at once.
maxCapacity: 128
implementation: generic-worker/worker-runner-linux
regions: [us-central1, us-west1, us-east1]
regions: [us-central1, us-west1, us-east1, europe-west4]
image: monopacker-translations-worker
instance_types:
- minCpuPlatform: Intel Skylake
Expand Down

0 comments on commit 64956fa

Please sign in to comment.