README.md

SSDs

Using local SSDs dramatically speeds up many of Sourcegraph's services. Even if the cluster's default storage class uses SSDs, it's likely network-mounted rather than local. Read your cloud provider's documentation for mouting local SSDs.

• GCP
• AWS

If you mount local SSDs on your nodes:

1. Change the cache-ssd volume to point to ${SSD_NODE_PATH}/pod-tmp where ${SSD_NODE_PATH} is the absolute path of the SSD on each node.

   For example, GCP mounts the first SSD disk to /mnt/disks/ssd0, so the cache-ssd volume would be configured like this:

   volumes:
     - hostPath:
         path: /mnt/disks/ssd0/pod-tmp
       name: cache-ssd

2. Append the kubectl apply command for the provided pod-tmp-gc DaemonSet to kubectl-apply-all.sh to periodically clean up files in the SSD on each node. This is necessary because files on the SSDs are not automatically cleaned up if pods crash or are rescheduled which can cause the SSDs to fill up.

   echo kubectl apply --prune -l deploy=pod-tmp-gc -f configure/ssd --recursive >> kubectl-apply-all.sh

3. Apply your changes to pod-tmp-gc and the base deployments to the cluster.

   ./kubectl-apply-all.sh

Here is a convenience script:

# This script requires https://github.com/sourcegraph/jy and https://github.com/sourcegraph/yj

SSD_NODE_PATH=/mnt/disks/ssd0 # update this to reflect the absolute path where SSDs are mounted on each node

# Mount the SSDs path in each deployment
find . -name "*Deployment.yaml" -exec sh -c "cat {} | yj | jq '(.spec.template.spec.volumes | select(. != null) | .[] | select(.name == \"cache-ssd\")) |= (del(.emptyDir) + {hostPath: {path: \"$SSD_NODE_PATH/pod-tmp\"}})' | jy -o {}" \;

# Update pod-tmp-gc.DaemonSet.yaml with SSD_NODE_PATH
DS=configure/ssd/pod-tmp-gc.DaemonSet.yaml
cat $DS | yj | jq ".spec.template.spec.volumes = [{name: \"pod-tmp\", hostPath: {path: \"$SSD_NODE_PATH/pod-tmp\"}}]" | jy -o $DS

# Deploy everything
echo kubectl apply --prune -l deploy=pod-tmp-gc -f configure/ssd --recursive >> kubectl-apply-all.sh
./kubectl-apply-all.sh

Note: If you are deploying Sourcegraph to a non-default namespace, you'll have to change the namespace specified in configure/ssd/pod-tmp-gc.ClusterRoleBinding.yaml to the one that you created.