kvm: can now accept parameters external-disk-format, cpu-options and numa

kvm_run script can now create external disk with specified format (qcow2, raw, etc...)
if kvm external disk is created it will never be replaced, even if user
reduce the number of disk with parameter 'external-disk-number' or change the
'external-disk-format' value. All created disk will be added to qemu run command.

numa option can be used to create [fake] numa nodes and expose them to the guest
OS instance. In this case it is important to know the host cpu topology and where
the instance will be allocated.

cpu-options are additional parameters like sockets=1,cores=5,threads=1
in this case if cpu-count parameter is 1 qemu options will be:
-smp 1,sockets=1,cores=5,threads=1

slapos/recipe/kvm/template/kvm_run.in

@@ -9,6 +9,8 @@ import subprocess
 import urllib
 import gzip
 import shutil
+from random import shuffle
+import glob
 
 # XXX: give all of this through parameter, don't use this as template, but as module
 qemu_img_path = '%(qemu-img-path)s'
@@ -30,10 +32,13 @@ listen_ip = '%(ipv4)s'
 mac_address = '%(mac-address)s'
 tap_mac_address = '%(tap-mac-address)s'
 smp_count = '%(smp-count)s'
+smp_options = '%(smp-options)s'.strip()
+numa_list = '%(numa)s'.split()
 ram_size = '%(ram-size)s'
 pid_file_path = '%(pid-file-path)s'
 external_disk_number = %(external-disk-number)s
 external_disk_size = '%(external-disk-size)s'
+external_disk_format = '%(external-disk-format)s'
 disk_storage_dict = {}
 disk_storage_list = """%(disk-storage-list)s""".split('\n')
 map_storage_list = []
@@ -68,21 +73,41 @@ def getSocketStatus(host, port):
     break
   return s
 
-def getMapStorageList(disk_storage_dict):
+def getMapStorageList(disk_storage_dict, external_disk_number):
   map_disk_file = os.path.join(etc_directory, '.data-disk-ids')
+  last_disk_num_f = os.path.join(etc_directory, '.data-disk-amount')
   id_list = []
-  if os.path.exists(map_disk_file):
+  last_amount = 0
+  map_f_exist = os.path.exists(map_disk_file)
+  if os.path.exists(last_disk_num_f):
+    with open(last_disk_num_f, 'r') as lf:
+      last_amount = int(lf.readline())
+  if map_f_exist:
     with open(map_disk_file, 'r') as mf:
       # ID are writen in one line: data1 data3 data2 ...
       content = mf.readline()
-      if content:
-        id_list = [id for id in content.split(' ') if disk_storage_dict.has_key(id)]
+      for id in content.split(' '):
+        if disk_storage_dict.has_key(id):
+          id_list.append(id)
+        else:
+          # Mean that this disk path has been removed (disk unmounted)
+          last_amount -= 1
   for key in disk_storage_dict:
     if not key in id_list:
       id_list.append(key)
+
+  if id_list:
+    if not map_f_exist:
+      # shuffle the list to not write disk in data1, data2, ... everytime 
+      shuffle(id_list)
+    if external_disk_number < last_amount:
+      # Drop created disk is not allowed
+      external_disk_number = last_amount
     with open(map_disk_file, 'w') as mf:
       mf.write(' '.join(id_list))
-  return id_list
+    with open(last_disk_num_f, 'w') as lf:
+      lf.write('%%s' %% external_disk_number)
+  return id_list, external_disk_number
 
 # Download existing hard drive if needed at first boot
 if not os.path.exists(disk_path) and virtual_hard_drive_url != '':
@@ -122,24 +147,34 @@ if not os.path.exists(disk_path):
 
 # Check and create external disk
 additional_disk_list = []
-for storage in sorted(disk_storage_list):
+for storage in disk_storage_list:
   if storage:
     key, val = storage.split(' ')
     disk_storage_dict[key.strip()] = val.strip()
 
-map_storage_list = getMapStorageList(disk_storage_dict)
+if not external_disk_format in ['qcow2', 'raw', 'vdi', 'vmdk', 'cloop']:
+  external_disk_format = 'qcow2'
 
+map_storage_list, external_disk_number = getMapStorageList(disk_storage_dict,
+                                                      int(external_disk_number))
+
+assert len(map_storage_list) == len(disk_storage_dict)
 if disk_storage_dict:
-  if int(external_disk_number) > 0:
+  if external_disk_number > 0:
     index = 0
     while (index < len(disk_storage_dict)) and (index < external_disk_number):
       path = disk_storage_dict[map_storage_list[index]]
       if os.path.exists(path):
-        disk_filepath = os.path.join(path, 'kvm_virtual_disk.qcow2')
-        if not os.path.exists(disk_filepath):
+        disk_filepath = os.path.join(path,
+                                  'kvm_virtual_disk.%%s' %% external_disk_format)
+        disk_list = glob.glob('%%s.*' %% os.path.join(path, 'kvm_virtual_disk'))
+        if disk_list == []:
           print('Creating one additional virtual hard drive...')
-          subprocess.Popen([qemu_img_path, 'create' ,'-f', 'qcow2',
+          subprocess.Popen([qemu_img_path, 'create' ,'-f', '%%s' %% external_disk_format,
               disk_filepath, '%%sG' %% external_disk_size])
+        else:
+          # Cannot change or recreate if disk is exists
+          disk_filepath = disk_list[0]
         additional_disk_list.append(disk_filepath)
       else:
         print('Data folder %%s was not used to create external disk %%r' %% (index +1))
@@ -150,6 +185,7 @@ if disk_storage_dict:
 # XXX: use_tap should be a boolean
 tap_network_parameter = []
 nat_network_parameter = []
+numa_parameter = []
 number = -1
 if use_nat == 'True':
   number += 1
@@ -164,8 +200,11 @@ if use_tap == 'True':
             tap_interface),
           '-device', 'e1000,netdev=lan%%s,mac=%%s' %% (number, tap_mac_address)]
 
+smp = smp_count
+if smp_options:
+  smp += ',%%s' %% smp_options
 kvm_argument_list = [qemu_path,
-  '-enable-kvm', '-smp', smp_count,
+  '-enable-kvm', '-smp', smp,
   '-m', ram_size, '-vga', 'std',
   '-drive', 'file=%%s,if=%%s' %% (disk_path, disk_type),
   '-vnc', '%%s:1,ipv4,password' %% listen_ip,
@@ -173,6 +212,11 @@ kvm_argument_list = [qemu_path,
   '-qmp', 'unix:%%s,server' %% socket_path,
   '-pidfile', pid_file_path,
 ]
+
+for numa in numa_list:
+  numa_parameter.extend(['-numa', numa])
+kvm_argument_list += numa_parameter
+
 if tap_network_parameter == [] and nat_network_parameter == []:
   print 'Warning : No network interface defined.'
 else:
@@ -201,4 +245,5 @@ else:
       '-drive', 'file=%%s,media=cdrom' %% default_disk_image
   ])
 
+print 'Starting KVM: \n %%s' %% ' '.join(kvm_argument_list)
 os.execv(qemu_path, kvm_argument_list)

software/kvm/common.cfg

@@ -93,7 +93,7 @@ mode = 0644
 recipe = hexagonit.recipe.download
 url = ${:_profile_base_location_}/instance-kvm.cfg.jinja2
 mode = 644
-md5sum = 5506e1df6ba32c6ead647636ebece79e
+md5sum = c8f69039aff15e20447c6e522267c152
 download-only = true
 on-update = true
 
@@ -101,7 +101,7 @@ on-update = true
 recipe = hexagonit.recipe.download
 url = ${:_profile_base_location_}/instance-kvm-cluster.cfg.jinja2.in
 mode = 644
-md5sum = 214c46a9aa7605951b8a1f98572dac28
+md5sum = a5c1f0620510a979daf8fd60120f6253
 download-only = true
 on-update = true
 

software/kvm/instance-kvm-cluster-input-schema.json

@@ -166,6 +166,16 @@
               "minimum": 1,
               "maximum": 8
             },
+            "cpu-options": {
+              "title": "Additional options (cores, threads, sockets, maxcpus) to use with cpu-count.",
+              "description": "Additional options to use with cpu-count. Options are separated by coma: [cores=cores][,threads=threads][,sockets=sockets][,maxcpus=maxcpus]. Only set this option if you really know what you're doing.",
+              "type": "string"
+            },
+            "numa": {
+              "title": "Simulate a multi node NUMA system. If mem and cpus are omitted, resources are split equally.",
+              "description": "Simulate a multi node NUMA system. If mem and cpus are omitted, resources are split equally. Each numa option are separated by space: node,nodeid=4,cpus=40-49,mem=64g node,nodeid=1,cpus=10-19,mem=128g. Only set this option if you really know what you're doing.",
+              "type": "string"
+            },
             "nbd-host": {
               "title": "NBD hostname",
               "description": "hostname (or IP) of the NBD server containing the boot image.",
@@ -235,6 +245,13 @@
               "maximum": 100,
               "default": 20
             },
+            "external-disk-format": {
+              "title": "Type of external disk drive to create by QEMU.",
+              "description": "Type of QEMU disk drive, to create.",
+              "type": "string",
+              "default": "qcow2",
+              "enum": ["qcow2", "raw", "vdi", "vmdk", "cloop"]
+            },
             "use-tap": {
               "title": "Use QEMU TAP network interface",
               "description": "Use QEMU TAP network interface, might require a bridge on SlapOS Node.",

software/kvm/instance-kvm-cluster.cfg.jinja2.in

@@ -38,6 +38,8 @@ config-ram-size = {{ dumps(kvm_parameter_dict.get('ram-size', 1024)) }}
 config-disk-size = {{ dumps(kvm_parameter_dict.get('disk-size', 10)) }}
 config-disk-type = {{ dumps(kvm_parameter_dict.get('disk-type', 'virtio')) }}
 config-cpu-count = {{ dumps(kvm_parameter_dict.get('cpu-count', 1)) }}
+config-cpu-options = {{ dumps(kvm_parameter_dict.get('cpu-options', '')) }}
+config-numa = {{ dumps(kvm_parameter_dict.get('numa', '')) }}
 
 {% set nat_rules_list = kvm_parameter_dict.get('nat-rules', [22, 80, 443]) -%}
 config-nat-rules = {{ nat_rules_list | join(' ') }}
@@ -49,6 +51,7 @@ config-virtual-hard-drive-md5sum = {{ dumps(kvm_parameter_dict.get('virtual-hard
 config-virtual-hard-drive-gzipped = {{ dumps(kvm_parameter_dict.get('virtual-hard-drive-gzipped', False)) }}
 config-external-disk-number = {{ dumps(kvm_parameter_dict.get('external-disk-number', 0)) }}
 config-external-disk-size = {{ dumps(kvm_parameter_dict.get('external-disk-size', 20)) }}
+config-external-disk-format = {{ dumps(kvm_parameter_dict.get('external-disk-format', 'qcow2')) }}
 return = 
   backend-url
   url

software/kvm/instance-kvm-input-schema.json

@@ -36,6 +36,16 @@
       "minimum": 1,
       "maximum": 8
     },
+    "cpu-options": {
+      "title": "Additional options (cores, threads, sockets, maxcpus) to use with cpu-count.",
+      "description": "Additional options to use with cpu-count. Options are separated by coma: [cores=cores][,threads=threads][,sockets=sockets][,maxcpus=maxcpus]. Only set this option if you really know what you're doing.",
+      "type": "string"
+    },
+    "numa": {
+      "title": "Simulate a multi node NUMA system. If mem and cpus are omitted, resources are split equally.",
+      "description": "Simulate a multi node NUMA system. If mem and cpus are omitted, resources are split equally. Each numa option are separated by space: node,nodeid=4,cpus=40-49,mem=64g node,nodeid=1,cpus=10-19,mem=128g. Only set this option if you really know what you're doing.",
+      "type": "string"
+    },
 
     "nbd-host": {
       "title": "NBD hostname",
@@ -101,6 +111,13 @@
       "maximum": 100,
       "default": 20
     },
+    "external-disk-format": {
+      "title": "Type of external disk drive to create by QEMU.",
+      "description": "Type of QEMU disk drive, to create.",
+      "type": "string",
+      "default": "qcow2",
+      "enum": ["qcow2", "raw", "vdi", "vmdk", "cloop"]
+    },
 
     "use-tap": {
       "title": "Use QEMU TAP network interface",

software/kvm/instance-kvm.cfg.jinja2

@@ -81,7 +81,9 @@ socket-path = ${directory:var}/qmp_socket
 pid-file-path = ${directory:run}/pid_file
 
 smp-count = ${slap-parameter:cpu-count}
+smp-options = ${slap-parameter:cpu-options}
 ram-size = ${slap-parameter:ram-size}
+numa = ${slap-parameter:numa}
 mac-address = ${create-mac:mac-address}
 tap-mac-address = ${create-tap-mac:mac-address}
 
@@ -110,6 +112,7 @@ disk-storage-list =
 {% endfor -%}
 external-disk-number = ${slap-parameter:external-disk-number}
 external-disk-size = ${slap-parameter:external-disk-size}
+external-disk-format = ${slap-parameter:external-disk-format}
 
 [kvm-vnc-promise]
 recipe = slapos.cookbook:check_port_listening
@@ -275,6 +278,10 @@ disk-size = 10
 disk-type = virtio
 
 cpu-count = 1
+# cpu-option is a string: [cores=cores][,threads=threads][,sockets=sockets][,maxcpus=maxcpus]
+cpu-options = 
+# list of numa options separate by space: node,nodeid=1,cpus=9-15 node,nodeid=2,cpus=1,3,7
+numa = 
 
 nat-rules = 22 80 443
 use-nat = True
@@ -286,3 +293,4 @@ virtual-hard-drive-gzipped = False
 
 external-disk-number = 0
 external-disk-size = 20
+external-disk-format = qcow2