Huge pages support

[Remp69]

**What is the motivation or use case for the change? **
Improve performances for large databases

Describe the solution you'd like
Huge pages support by adding declaration of hugepages-2Mi resource limit for the postgresql container.
https://kubernetes.io/docs/tasks/manage-hugepages/scheduling-hugepages/

Please tell us about your environment:

• Where is this running ( Local, Cloud Provider) local
• Container Image Tag: 4.5.1
• PostgreSQL Version: 13.1
• Platform (Docker, Kubernetes, OpenShift): kubernetes
• Platform Version: 1.18

[jkatz]

Thanks for the suggestion.

In the interim, you can modify the cluster-deployment.json entry in the pgo-config ConfigMap and add the mount points for the huge pages. After restarting the Operator Pod, any new cluster you create will include the huge pages mount.

[ns-rsuvorov]

We are running into the same issue it seems. But we could not locate the cluster-deployment.json entry in the pgo-config ConfigMap. Has something changed since the above response?
We are using PGO 5.1.0.

[cbandy]

@ns-rsuvorov in v5 you can set any compute request or limit in the resources field of the PostgresCluster spec.

[jdambly-ns]

@cbandy yes while we can add that to the, we still need to be able to set volume mounts which is not allowed, here is a patch we created to work around the issue

{
	"spec": {
		"template": {
			"spec": {
				"containers": [{
					"name": "database",
					"resources": {
						"limits": {
							"hugepages-2Mi": "100Mi"
						}
					},
					"volumeMounts": [{
						"mountPath": "/hugepages-2Mi",
						"name": "hugepage-2mi"
					}]
				}],
				"volumes": [{
					"name": "hugepage-2mi",
					"emptyDir": {
						"meduim": "HugePages-2Mi"
					}
				}]
			}
		}
	}
}

what concerns me is that even if we set huge_pages: off in parameters, it's not making into postgres configMap, I can see it on the cluster configmap for patroni, but it doesn't seem to give that to the postgres process that is running the db

[benjaminjb]

Hi, thanks for the feedback!

We've been running some tests with huge_pages and have determined that you shouldn't have to mount additional volumes to use huge_pages. You should be able to request those through the postgrescluster spec directly in the resources field.

There are also some additional considerations when requesting huge_pages:

1. for one thing, your underlying Kubernetes nodes have to have huge_pages enabled;
2. for another thing, your PG instance needs to be trying to use huge_pages.

Point (2) shouldn't be an issue here because the PG default for huge_pages is try. (This actually causes some problems with users who are not requesting huge_pages through the postgrescluster but whose Kubernetes nodes have huge_pages enabled because PG will try to use those huge_pages anyway. As long as you're requesting huge_pages, it shouldn't concern you, but it's an interesting issue that came up and that we're addressing in the two PRs already mentioned in this thread.)

If you do continue to see a need to mount additional volumes, please reopen and provide additional information about your specific Kubernetes environment and operator Deployment (e.g. Kubernetes version, PGO version, etc.)?