Cinder LVM Oversubscription in thin provisioning

目录

Oversubscription in thin provisioning

• Cinder spec: Over Subscription in Thin Provisioning https://review.openstack.org/#/c/129342/12/specs/kilo/over-subscription-in-thin-provisioning.rst
• cinder bp: Over subscription in thin provisioning https://blueprints.launchpad.net/cinder/+spec/over-subscription-in-thin-provisioning

所谓 Oversubscription in thin provisioning，就相当于是 Thin Provisioning Storage Pool 的超分比限制，防止 Thin Provisioning Storage Pool 被无限放大。对应的配置项是 max_over_subscription_ratio，默认值为 20.0。e.g.

[lvm-1]
volume_group = centos
volume_driver = cinder.volume.drivers.lvm.LVMVolumeDriver
volume_backend_name = lvm-1
iscsi_helper = tgtadm
iscsi_protocol = iscsi
lvm_max_over_subscription_ratio = 25.0


功能生效的地方是 Cinder Scheduler Filter capacity_filter，代码如下：

# cinder/scheduler/filters/capacity_filter.py # Only evaluate using max_over_subscription_ratio if
# thin_provisioning_support is True. Check if the ratio of
# provisioned capacity over total capacity has exceeded over
# subscription ratio.
if (thin and backend_state.thin_provisioning_support and
backend_state.max_over_subscription_ratio >= 1):
provisioned_ratio = ((backend_state.provisioned_capacity_gb +
requested_size) / total)
LOG.debug("Checking provisioning for request of %s GB. "
"Backend: %s", requested_size, backend_state)
if provisioned_ratio > backend_state.max_over_subscription_ratio:
msg_args = {
"provisioned_ratio": provisioned_ratio,
"oversub_ratio": backend_state.max_over_subscription_ratio,
"grouping": grouping,
"grouping_name": backend_state.backend_id,
}
LOG.warning(
"Insufficient free space for thin provisioning. "
"The ratio of provisioned capacity over total capacity "
"%(provisioned_ratio).2f has exceeded the maximum over "
"subscription ratio %(oversub_ratio).2f on %(grouping)s "
"%(grouping_name)s.", msg_args)
return False


如果 provisioned_ratio > backend_state.max_over_subscription_ratio 为 True 则表示当前 provisioned 的比率已经大于 Oversubscription in thin provisioning 设定的比率了，所以当前 Cinder Backend 无法继续创建新的 Volume。

那么，provisioned_ratio 是怎么得到的呢？代码如下：

provisioned_ratio = ((backend_state.provisioned_capacity_gb + requested_size) / total)


首先弄清楚这几个变量的含义及取值：

# cinder/utils.py # provisioned_capacity_gb is the apparent total capacity of
# all the volumes created on a backend, which is greater than
# or equal to allocated_capacity_gb, which is the apparent
# total capacity of all the volumes created on a backend
# in Cinder. Using allocated_capacity_gb as the default of
# provisioned_capacity_gb if it is not set.
allocated_capacity_gb = capability.get('allocated_capacity_gb', 0)
provisioned_capacity_gb = capability.get('provisioned_capacity_gb',
allocated_capacity_gb)
thin_provisioning_support = capability.get('thin_provisioning_support',
False)
total_capacity_gb = capability.get('total_capacity_gb', 0)
free_capacity_gb = capability.get('free_capacity_gb', 0)


官方说明：

• total_capacity: This is an existing parameter already reported by the driver. It is the total physical capacity. Example: Assume backend A has a total physical capacity of 100G.

• available_capacity: This is an existing parameter already reported by the driver. It is the real physical capacity available to be used. Example: Assume backend A has a total physical capacity of 100G. There are 10G thick luns and 20G thin luns (10G out of the 20G thin luns are written). In this case, available_capacity = 100 - 10 -10 = 80G.

• used_capacity: This parameter is calculated by the difference between total_capacity and available_capacity. It is used below for calculating used ratio.

• volume_size: This is an existing parameter. It is the size of the volume to be provisioned.

• provisioned_capacity: This is a new parameter. It is the apparent allocated space indicating how much capacity has been provisioned. Example: User A created 2x10G volumes in Cinder from backend A, and user B created 3x10G volumes from backend A directly, without using Cinder. Assume those are all the volumes provisioned on backend A. The total provisioned_capacity will be 50G and that is what the driver should be reporting.

• allocated_capacity: This is an existing parameter. Cinder uses this to keep track of how much capacity has been allocated through Cinder. Example: Using the same example above for provisioned_capacity, the allocated_capacity will be 20G because that is what has been provisioned through Cinder. allocated_capacity is documented here to differentiate from the new parameter provisioned_capacity.

简要说明：

• allocated_capacity_gb：实际已分配的容量。
• provisioned_capacity_gb：已置备的容量，大于或等于 allocated_capacity_gb。
  
  注：所谓已置备是存储领域的专业术语，表示逻辑上的已经分配出去的虚拟容量，可能是精简置备的也可能是厚置备的。需要与实际已分配的容量作一个区分。
• total_capacity_gb：实际的总容量。
• free_capacity_gb：实际剩余的容量。

在 LVM Driver 中的含义：

• allocated_capacity_gb: 在 cinder 中分配的 cinder volume 总大小
• free_capacity_gb: vg 中空闲的容量
• provisioned_capacity_gb： vg 上面分配的总大小，因为文件稀疏（Thin）的问题，这个值可能很大
• total_capacity_gb: vg 容量的总大小
• max_over_subscription_ratio: 最大超配比

回过头来在看看这条计算公式：

provisioned_ratio = ((backend_state.provisioned_capacity_gb + requested_size) / total)


provisioned_ratio 就是当前已经被置备的容量的比率，包括精简置备或厚置备的情况，也就是我们不希望它太过于大的比率。理应小于 max_over_subscription_ratio。

但有一个问题需要注意，在 LVMDriver 中使用 Thin provisioning 时，provisioned_capacity_gb 马上就等于 VG 的总容量。你会发现虽然你还没有创建任何 Volume，但 provisioned_capacity_gb 就已经等于 VG 的 Size 了。代码如下：

# cinder/volume/drivers/lvm.py
if self.configuration.lvm_mirrors > 0:
total_capacity =\
self.vg.vg_mirror_size(self.configuration.lvm_mirrors)
free_capacity =\
self.vg.vg_mirror_free_space(self.configuration.lvm_mirrors)
provisioned_capacity = round(
float(total_capacity) - float(free_capacity), 2)
elif self.configuration.lvm_type == 'thin':
total_capacity = self.vg.vg_thin_pool_size
free_capacity = self.vg.vg_thin_pool_free_space
provisioned_capacity = self.vg.vg_provisioned_capacity
else:
total_capacity = self.vg.vg_size
free_capacity = self.vg.vg_free_space
provisioned_capacity = round(
float(total_capacity) - float(free_capacity), 2)


这是因为 Cinder 假设，当你使用 LVM Thin provisioning 的时候，那么整个 VG 都应该是 Thin provisioning 的，不存在 Thin 和 Thick 混合的情况，否则无法正确进行容量的计算。因此，当我们使用 LVM Thin provisioning 时，切记要划分一个干净的 VG 给 LVM Backend，否则就会出现资源计算错误的问题。