add cpuaffinity module

proxmox/README.md

@@ -101,13 +101,13 @@ vpc_main_cidr = "172.16.0.0/24"
 # We will create one control-plane node on the Proxmox node `node1` (count = 1)
 controlplane = {
   "node1" = {
-    id    = 500
+    id    = 500,
     count = 1,
     cpu   = 2,
     mem   = 6144,
   },
   "node2" = {
-    id    = 520
+    id    = 520,
     count = 0,
     cpu   = 2,
     mem   = 6144,
@@ -117,21 +117,23 @@ controlplane = {
 # One web and worker node:
 instances = {
   "node1" = {
-    web_id       = 1000
+    enabled      = true,
+    web_id       = 1000,
     web_count    = 1,
     web_cpu      = 2,
     web_mem      = 4096,
-    worker_id    = 1050
+    worker_id    = 1050,
     worker_count = 1,
     worker_cpu   = 2,
     worker_mem   = 4096,
   },
   "node2" = {
-    web_id       = 2000
+    enabled      = true,
+    web_id       = 2000,
     web_count    = 0,
     web_cpu      = 2,
     web_mem      = 4096,
-    worker_id    = 2050
+    worker_id    = 2050,
     worker_count = 0,
     worker_cpu   = 2,
     worker_mem   = 4096,
@@ -159,9 +161,14 @@ Receive `kubeconfig` file
 make kubeconfig
 ```
 
+Test the cluster
+
 ```shell
+export KUBECONFIG=kubeconfig
+
 kubectl get nodes -o wide
 kubectl get pods -o wide -A
+kubectl get csistoragecapacities -ocustom-columns=CLASS:.storageClassName,AVAIL:.capacity,ZONE:.nodeTopology.matchLabels -A
 ```
 
 Resault:

proxmox/cpuaffinity/outputs.tf

@@ -0,0 +1,8 @@
+
+output "arch" {
+  value = { for k, v in local.vm_arch : k => {
+    cpus : v.cpus
+    numa : { for numa in range(length(var.cpu_affinity)) : numa => v.numa[numa] if length(v.numa[numa]) > 0 }
+    }
+  }
+}

proxmox/cpuaffinity/variables.tf

@@ -0,0 +1,41 @@
+
+variable "cpu_affinity" {
+  description = "CPU numa affinity list"
+  type        = list(string)
+  default     = ["0-15,64-79", "16-31,80-95", "32-47,96-111", "48-63,112-127"]
+}
+
+variable "vms" {
+  type    = number
+  default = 2
+}
+
+variable "cpus" {
+  type    = number
+  default = 16
+}
+
+variable "shift" {
+  type    = number
+  default = 0
+}
+
+locals {
+  server_cpus = [for i in var.cpu_affinity :
+    flatten([for r in split(",", i) : (strcontains(r, "-") ? range(split("-", r)[0], split("-", r)[1] + 1, 1) : [r])])
+  ]
+
+  cpus = [for k, v in local.server_cpus :
+    flatten([flatten([for r in range(length(v) / 2) : [v[r], v[r + length(v) / 2]]])])
+  ]
+
+  shift = var.shift * length(local.cpus[0])
+
+  vm_arch = { for k in flatten([
+    for inx in range(var.vms) : {
+      inx : inx
+      cpus : slice(flatten(local.cpus), inx * var.cpus + local.shift, (inx + 1) * var.cpus + local.shift)
+      numa : { for numa in range(length(var.cpu_affinity)) : numa => setintersection(local.cpus[numa], slice(flatten(local.cpus), inx * var.cpus + local.shift, (inx + 1) * var.cpus + local.shift)) }
+    }
+  ]) : k.inx => k }
+}

proxmox/instances-db.tf

@@ -6,20 +6,11 @@ locals {
   dbs = { for k in flatten([
     for zone in local.zones : [
       for inx in range(lookup(try(var.instances[zone], {}), "db_count", 0)) : {
+        inx : inx
         id : lookup(try(var.instances[zone], {}), "db_id", 9000) + inx
         name : "${local.db_prefix}-${format("%02d", index(local.zones, zone))}${format("%x", 10 + inx)}"
         zone : zone
         cpu : lookup(try(var.instances[zone], {}), "db_cpu", 1)
-        cpus : lookup(try(var.instances[zone], {}), "db_affinity", "") != "" ? lookup(var.instances[zone], "db_affinity") : join(",", slice(
-          flatten(local.cpus[zone]),
-          2 * data.proxmox_virtual_environment_node.node[zone].cpu_count - (inx + 1) * lookup(try(var.instances[zone], {}), "db_cpu", 1),
-          2 * data.proxmox_virtual_environment_node.node[zone].cpu_count - inx * lookup(try(var.instances[zone], {}), "db_cpu", 1)
-        ))
-        numas : [0]
-        # range(
-        #   length(local.cpu_numa[zone]) - (inx + 1) * lookup(try(var.instances[zone], {}), "db_numas", 1),
-        #   length(local.cpu_numa[zone]) - inx * lookup(try(var.instances[zone], {}), "db_numas", 1)
-        # )
         mem : lookup(try(var.instances[zone], {}), "db_mem", 2048)
 
         hvv4 = cidrhost(local.subnets[zone], 0)
@@ -34,6 +25,19 @@ locals {
   ]) : k.name => k }
 }
 
+module "db_affinity" {
+  for_each = { for zone in local.zones : zone => {
+    zone : zone
+    vms : lookup(try(var.instances[zone], {}), "db_count", 0)
+  } if lookup(try(var.instances[zone], {}), "db_count", 0) > 0 }
+
+  source       = "./cpuaffinity"
+  cpu_affinity = var.nodes[each.value.zone].cpu
+  vms          = each.value.vms
+  cpus         = lookup(try(var.instances[each.value.zone], {}), "db_cpu", 1)
+  # shift        = length(var.nodes[each.value.zone].cpu) - 1
+}
+
 resource "proxmox_virtual_environment_file" "db_machineconfig" {
   for_each     = local.dbs
   node_name    = each.value.zone
@@ -92,7 +96,7 @@ resource "proxmox_virtual_environment_vm" "db" {
   cpu {
     architecture = "x86_64"
     cores        = each.value.cpu
-    affinity     = each.value.cpus
+    affinity     = join(",", module.db_affinity[each.value.zone].arch[each.value.inx].cpus)
     sockets      = 1
     numa         = true
     type         = "host"
@@ -103,10 +107,10 @@ resource "proxmox_virtual_environment_vm" "db" {
     # keep_hugepages = true
   }
   dynamic "numa" {
-    for_each = { for idx, numa in each.value.numas : numa => {
-      device = "numa${idx}"
-      cpus   = "${idx * (each.value.cpu / length(each.value.numas))}-${(idx + 1) * (each.value.cpu / length(each.value.numas)) - 1}"
-      mem    = each.value.mem / length(each.value.numas)
+    for_each = { for idx, numa in module.db_affinity[each.value.zone].arch[each.value.inx].numa : idx => {
+      device = "numa${index(keys(module.db_affinity[each.value.zone].arch[each.value.inx].numa), idx)}"
+      cpus   = "${idx * (each.value.cpu / length(module.db_affinity[each.value.zone].arch[each.value.inx].numa))}-${(idx + 1) * (each.value.cpu / length(module.db_affinity[each.value.zone].arch[each.value.inx].numa)) - 1}"
+      mem    = each.value.mem / length(module.db_affinity[each.value.zone].arch[each.value.inx].numa)
     } }
     content {
       device    = numa.value.device

proxmox/instances-web.tf

@@ -6,15 +6,11 @@ locals {
   webs = { for k in flatten([
     for zone in local.zones : [
       for inx in range(lookup(try(var.instances[zone], {}), "web_count", 0)) : {
+        inx : inx
         id : lookup(try(var.instances[zone], {}), "web_id", 9000) + inx
         name : "${local.web_prefix}-${format("%02d", index(local.zones, zone))}${format("%x", 10 + inx)}"
         zone : zone
         cpu : lookup(try(var.instances[zone], {}), "web_cpu", 1)
-        cpus : join(",", slice(
-          flatten(local.cpus[zone]),
-          inx * lookup(try(var.instances[zone], {}), "web_cpu", 1), (inx + 1) * lookup(try(var.instances[zone], {}), "web_cpu", 1)
-        ))
-        numas : [0] # [inx]
         mem : lookup(try(var.instances[zone], {}), "web_mem", 2048)
 
         hvv4 = cidrhost(local.subnets[zone], 0)
@@ -29,6 +25,18 @@ locals {
   ]) : k.name => k }
 }
 
+module "web_affinity" {
+  for_each = { for zone in local.zones : zone => {
+    zone : zone
+    vms : lookup(try(var.instances[zone], {}), "web_count", 0)
+  } if lookup(try(var.instances[zone], {}), "web_count", 0) > 0 }
+
+  source       = "./cpuaffinity"
+  cpu_affinity = var.nodes[each.value.zone].cpu
+  vms          = each.value.vms
+  cpus         = lookup(try(var.instances[each.value.zone], {}), "web_cpu", 1)
+}
+
 resource "proxmox_virtual_environment_file" "web_machineconfig" {
   for_each     = local.webs
   node_name    = each.value.zone
@@ -109,7 +117,7 @@ resource "proxmox_virtual_environment_vm" "web" {
   cpu {
     architecture = "x86_64"
     cores        = each.value.cpu
-    affinity     = each.value.cpus
+    affinity     = join(",", module.web_affinity[each.value.zone].arch[each.value.inx].cpus)
     sockets      = 1
     numa         = true
     type         = "host"
@@ -120,10 +128,10 @@ resource "proxmox_virtual_environment_vm" "web" {
     # keep_hugepages = true
   }
   dynamic "numa" {
-    for_each = { for idx, numa in each.value.numas : numa => {
-      device = "numa${idx}"
-      cpus   = "0-${each.value.cpu - 1}"
-      mem    = each.value.mem
+    for_each = { for idx, numa in module.web_affinity[each.value.zone].arch[each.value.inx].numa : idx => {
+      device = "numa${index(keys(module.web_affinity[each.value.zone].arch[each.value.inx].numa), idx)}"
+      cpus   = "${idx * (each.value.cpu / length(module.web_affinity[each.value.zone].arch[each.value.inx].numa))}-${(idx + 1) * (each.value.cpu / length(module.web_affinity[each.value.zone].arch[each.value.inx].numa)) - 1}"
+      mem    = each.value.mem / length(module.web_affinity[each.value.zone].arch[each.value.inx].numa)
     } }
     content {
       device    = numa.value.device

proxmox/instances-worker.tf

@@ -6,16 +6,11 @@ locals {
   workers = { for k in flatten([
     for zone in local.zones : [
       for inx in range(lookup(try(var.instances[zone], {}), "worker_count", 0)) : {
+        inx : inx
         id : lookup(try(var.instances[zone], {}), "worker_id", 9000) + inx
         name : "${local.worker_prefix}-${format("%02d", index(local.zones, zone))}${format("%x", 10 + inx)}"
         zone : zone
-        node_name : zone
         cpu : lookup(try(var.instances[zone], {}), "worker_cpu", 1)
-        cpus : join(",", slice(
-          flatten(local.cpus[zone]),
-          (inx + 2) * lookup(try(var.instances[zone], {}), "worker_cpu", 1), (inx + 3) * lookup(try(var.instances[zone], {}), "worker_cpu", 1)
-        ))
-        numas : [0] # [2 + inx]
         mem : lookup(try(var.instances[zone], {}), "worker_mem", 2048)
 
         hvv4 = cidrhost(local.subnets[zone], 0)
@@ -30,9 +25,21 @@ locals {
   ]) : k.name => k }
 }
 
+module "worker_affinity" {
+  for_each = { for zone in local.zones : zone => {
+    zone : zone
+    vms : lookup(try(var.instances[zone], {}), "worker_count", 0)
+  } if lookup(try(var.instances[zone], {}), "worker_count", 0) > 0 }
+
+  source       = "./cpuaffinity"
+  cpu_affinity = var.nodes[each.value.zone].cpu
+  vms          = each.value.vms
+  cpus         = lookup(try(var.instances[each.value.zone], {}), "worker_cpu", 1)
+}
+
 resource "proxmox_virtual_environment_file" "worker_machineconfig" {
   for_each     = local.workers
-  node_name    = each.value.node_name
+  node_name    = each.value.zone
   content_type = "snippets"
   datastore_id = "local"
 
@@ -88,7 +95,7 @@ resource "proxmox_virtual_environment_vm" "worker" {
   cpu {
     architecture = "x86_64"
     cores        = each.value.cpu
-    affinity     = each.value.cpus
+    affinity     = join(",", module.worker_affinity[each.value.zone].arch[each.value.inx].cpus)
     sockets      = 1
     numa         = true
     type         = "host"
@@ -99,10 +106,10 @@ resource "proxmox_virtual_environment_vm" "worker" {
     # keep_hugepages = true
   }
   dynamic "numa" {
-    for_each = { for idx, numa in each.value.numas : numa => {
-      device = "numa${idx}"
-      cpus   = "0-${each.value.cpu - 1}"
-      mem    = each.value.mem
+    for_each = { for idx, numa in module.worker_affinity[each.value.zone].arch[each.value.inx].numa : idx => {
+      device = "numa${index(keys(module.worker_affinity[each.value.zone].arch[each.value.inx].numa), idx)}"
+      cpus   = "${idx * (each.value.cpu / length(module.worker_affinity[each.value.zone].arch[each.value.inx].numa))}-${(idx + 1) * (each.value.cpu / length(module.worker_affinity[each.value.zone].arch[each.value.inx].numa)) - 1}"
+      mem    = each.value.mem / length(module.worker_affinity[each.value.zone].arch[each.value.inx].numa)
     } }
     content {
       device    = numa.value.device