Merge pull request #30370 from MaterializeInc/doy-cluster-sizes

make cluster sizes configurable in the orchestratord helm chart
MaterializeInc · Nov 7, 2024 · 9b6ef61 · 9b6ef61
2 parents 8c42cac + 7f3cad8
commit 9b6ef61
Show file tree

Hide file tree

Showing 5 changed files with 109 additions and 57 deletions.
diff --git a/misc/helm-charts/operator/README.md b/misc/helm-charts/operator/README.md
@@ -127,6 +127,21 @@ The following table lists the configurable parameters of the Materialize operato
 | `operator.args.localDevelopment` |  | ``true`` |
 | `operator.args.region` |  | ``"kind"`` |
 | `operator.args.startupLogFilter` |  | ``"INFO,mz_orchestratord=TRACE"`` |
+| `operator.clusters.defaultSizes.analytics` |  | ``"3xsmall"`` |
+| `operator.clusters.defaultSizes.catalogServer` |  | ``"2xsmall"`` |
+| `operator.clusters.defaultSizes.default` |  | ``"3xsmall"`` |
+| `operator.clusters.defaultSizes.probe` |  | ``"probe"`` |
+| `operator.clusters.defaultSizes.support` |  | ``"3xsmall"`` |
+| `operator.clusters.defaultSizes.system` |  | ``"3xsmall"`` |
+| `operator.clusters.sizes.2xsmall.credits_per_hour` |  | ``"0.5"`` |
+| `operator.clusters.sizes.2xsmall.scale` |  | ``1`` |
+| `operator.clusters.sizes.2xsmall.workers` |  | ``1`` |
+| `operator.clusters.sizes.3xsmall.credits_per_hour` |  | ``"0.25"`` |
+| `operator.clusters.sizes.3xsmall.scale` |  | ``1`` |
+| `operator.clusters.sizes.3xsmall.workers` |  | ``1`` |
+| `operator.clusters.sizes.probe.credits_per_hour` |  | ``"0"`` |
+| `operator.clusters.sizes.probe.scale` |  | ``1`` |
+| `operator.clusters.sizes.probe.workers` |  | ``1`` |
 | `operator.image.pullPolicy` |  | ``"IfNotPresent"`` |
 | `operator.image.repository` |  | ``"materialize/orchestratord"`` |
 | `operator.image.tag` |  | ``"v0.124.0-dev.0--pr.gd5e227d7d07dc4878d2a065c7c10a48c2555385a"`` |

diff --git a/misc/helm-charts/operator/templates/deployment.yaml b/misc/helm-charts/operator/templates/deployment.yaml
@@ -53,6 +53,29 @@ spec:
         {{- if .Values.operator.args.createBalancers }}
         - "--create-balancers"
         {{- end }}
+        {{- if .Values.operator.clusters }}
+        {{- if .Values.operator.clusters.sizes }}
+        - '--environmentd-cluster-replica-sizes={{ toJson .Values.operator.clusters.sizes }}'
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.default }}
+        - "--bootstrap-default-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.default }}"
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.system }}
+        - "--bootstrap-builtin-system-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.system }}"
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.probe }}
+        - "--bootstrap-builtin-probe-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.probe }}"
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.support }}
+        - "--bootstrap-builtin-support-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.support }}"
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.catalogServer }}
+        - "--bootstrap-builtin-catalog-server-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.catalogServer }}"
+        {{- end }}
+        {{ if .Values.operator.clusters.defaultSizes.analytics }}
+        - "--bootstrap-builtin-analytics-cluster-replica-size={{ .Values.operator.clusters.defaultSizes.analytics }}"
+        {{- end }}
+        {{- end }}
         - "--image-pull-policy={{ kebabcase .Values.operator.image.pullPolicy }}"
         {{- range $key, $value := .Values.environmentd.nodeSelector }}
         - "--environmentd-node-selector={{ $key }}={{ $value }}"

diff --git a/misc/helm-charts/operator/values.yaml b/misc/helm-charts/operator/values.yaml
@@ -33,6 +33,27 @@ operator:
     environmentdConnectionRoleARN: ""
     # awsAccountID is required when cloudProvider is "aws"
     awsAccountID: ""
+  clusters:
+    sizes:
+      3xsmall:
+        workers: 1
+        scale: 1
+        credits_per_hour: "0.25"
+      2xsmall:
+        workers: 1
+        scale: 1
+        credits_per_hour: "0.5"
+      probe:
+        workers: 1
+        scale: 1
+        credits_per_hour: "0"
+    defaultSizes:
+      default: 3xsmall
+      system: 3xsmall
+      probe: probe
+      support: 3xsmall
+      catalogServer: 2xsmall
+      analytics: 3xsmall
   # Node selector to use for the operator pod
   nodeSelector: {}
   resources:

diff --git a/src/orchestratord/src/controller/materialize.rs b/src/orchestratord/src/controller/materialize.rs
@@ -16,7 +16,6 @@ use std::{
 use http::HeaderValue;
 use k8s_openapi::apimachinery::pkg::apis::meta::v1::{Condition, Time};
 use kube::{api::PostParams, runtime::controller::Action, Api, Client, Resource, ResourceExt};
-use serde_json::json;
 use tracing::{debug, trace};
 
 use crate::metrics::Metrics;
@@ -61,20 +60,20 @@ pub struct Args {
     #[clap(flatten)]
     network_policies: NetworkPolicyConfig,
 
-    #[clap(long, default_value_t = default_cluster_replica_sizes())]
-    environmentd_cluster_replica_sizes: String,
-    #[clap(long, default_value = "25cc")]
-    bootstrap_default_cluster_replica_size: String,
-    #[clap(long, default_value = "25cc")]
-    bootstrap_builtin_system_cluster_replica_size: String,
-    #[clap(long, default_value = "mz_probe")]
-    bootstrap_builtin_probe_cluster_replica_size: String,
-    #[clap(long, default_value = "25cc")]
-    bootstrap_builtin_support_cluster_replica_size: String,
-    #[clap(long, default_value = "50cc")]
-    bootstrap_builtin_catalog_server_cluster_replica_size: String,
-    #[clap(long, default_value = "25cc")]
-    bootstrap_builtin_analytics_cluster_replica_size: String,
+    #[clap(long)]
+    environmentd_cluster_replica_sizes: Option<String>,
+    #[clap(long)]
+    bootstrap_default_cluster_replica_size: Option<String>,
+    #[clap(long)]
+    bootstrap_builtin_system_cluster_replica_size: Option<String>,
+    #[clap(long)]
+    bootstrap_builtin_probe_cluster_replica_size: Option<String>,
+    #[clap(long)]
+    bootstrap_builtin_support_cluster_replica_size: Option<String>,
+    #[clap(long)]
+    bootstrap_builtin_catalog_server_cluster_replica_size: Option<String>,
+    #[clap(long)]
+    bootstrap_builtin_analytics_cluster_replica_size: Option<String>,
 
     #[clap(
         long,
@@ -187,15 +186,6 @@ impl Display for Error {
     }
 }
 
-fn default_cluster_replica_sizes() -> String {
-    json!({
-        "25cc": {"workers": 1, "scale": 1, "credits_per_hour": "0.25"},
-        "50cc": {"workers": 1, "scale": 1, "credits_per_hour": "0.5"},
-        "mz_probe": {"workers": 1, "scale": 1, "credits_per_hour": "0.00"},
-    })
-    .to_string()
-}
-
 pub struct Context {
     config: Args,
     tracing: TracingCliArgs,

diff --git a/src/orchestratord/src/controller/materialize/resources.rs b/src/orchestratord/src/controller/materialize/resources.rs
@@ -801,16 +801,42 @@ fn create_environmentd_statefulset_object(
     ));
 
     // Add cluster and storage host size arguments.
-    args.extend([
-        format!(
-            "--cluster-replica-sizes={}",
-            &config.environmentd_cluster_replica_sizes
-        ),
-        format!(
-            "--bootstrap-default-cluster-replica-size={}",
-            &config.bootstrap_default_cluster_replica_size
-        ),
-    ]);
+    args.extend(
+        [
+            config
+                .environmentd_cluster_replica_sizes
+                .as_ref()
+                .map(|sizes| format!("--cluster-replica-sizes={sizes}")),
+            config
+                .bootstrap_default_cluster_replica_size
+                .as_ref()
+                .map(|size| format!("--bootstrap-default-cluster-replica-size={size}")),
+            config
+                .bootstrap_builtin_system_cluster_replica_size
+                .as_ref()
+                .map(|size| format!("--bootstrap-builtin-system-cluster-replica-size={size}")),
+            config
+                .bootstrap_builtin_probe_cluster_replica_size
+                .as_ref()
+                .map(|size| format!("--bootstrap-builtin-probe-cluster-replica-size={size}")),
+            config
+                .bootstrap_builtin_support_cluster_replica_size
+                .as_ref()
+                .map(|size| format!("--bootstrap-builtin-support-cluster-replica-size={size}")),
+            config
+                .bootstrap_builtin_catalog_server_cluster_replica_size
+                .as_ref()
+                .map(|size| {
+                    format!("--bootstrap-builtin-catalog-server-cluster-replica-size={size}")
+                }),
+            config
+                .bootstrap_builtin_analytics_cluster_replica_size
+                .as_ref()
+                .map(|size| format!("--bootstrap-builtin-analytics-cluster-replica-size={size}")),
+        ]
+        .into_iter()
+        .flatten(),
+    );
 
     // Add networking arguments.
     args.extend([
@@ -993,29 +1019,6 @@ fn create_environmentd_statefulset_object(
         args.push(format!("--sentry-tag=region={}", config.region));
     }
 
-    args.extend([
-        format!(
-            "--bootstrap-builtin-system-cluster-replica-size={}",
-            &config.bootstrap_builtin_system_cluster_replica_size
-        ),
-        format!(
-            "--bootstrap-builtin-probe-cluster-replica-size={}",
-            &config.bootstrap_builtin_probe_cluster_replica_size
-        ),
-        format!(
-            "--bootstrap-builtin-support-cluster-replica-size={}",
-            &config.bootstrap_builtin_support_cluster_replica_size
-        ),
-        format!(
-            "--bootstrap-builtin-catalog-server-cluster-replica-size={}",
-            &config.bootstrap_builtin_catalog_server_cluster_replica_size
-        ),
-        format!(
-            "--bootstrap-builtin-analytics-cluster-replica-size={}",
-            &config.bootstrap_builtin_analytics_cluster_replica_size
-        ),
-    ]);
-
     // Add Persist PubSub arguments
     args.push(format!(
         "--persist-pubsub-url=http://{}:{}",