summaryrefslogtreecommitdiffstats
path: root/roles/kube_nfs_volumes
diff options
context:
space:
mode:
authorJan Safranek <jsafrane@redhat.com>2015-05-20 11:40:35 +0200
committerJan Safranek <jsafrane@redhat.com>2015-05-20 11:40:35 +0200
commit98acbb6d24921ded065fca3b57943ef1ae7fb15f (patch)
tree17571cd1d1ef49cc81b9dcd8f1684ff65c30e37f /roles/kube_nfs_volumes
parent8dedb32348549d8aa279201cfb89c50d5dc83ba5 (diff)
downloadopenshift-98acbb6d24921ded065fca3b57943ef1ae7fb15f.tar.gz
openshift-98acbb6d24921ded065fca3b57943ef1ae7fb15f.tar.bz2
openshift-98acbb6d24921ded065fca3b57943ef1ae7fb15f.tar.xz
openshift-98acbb6d24921ded065fca3b57943ef1ae7fb15f.zip
Add nfs-volumes role.
This role is useful to use (physical) disks as persistent volumes in Kubernetes. It partitions the disks, exports the partitions as NFS shares and registers the shares as physical volumes at remote Kubernetes installation.
Diffstat (limited to 'roles/kube_nfs_volumes')
-rw-r--r--roles/kube_nfs_volumes/README.md111
-rw-r--r--roles/kube_nfs_volumes/defaults/main.yml10
-rw-r--r--roles/kube_nfs_volumes/handlers/main.yml3
-rw-r--r--roles/kube_nfs_volumes/library/partitionpool.py240
-rw-r--r--roles/kube_nfs_volumes/meta/main.yml16
-rw-r--r--roles/kube_nfs_volumes/tasks/main.yml25
-rw-r--r--roles/kube_nfs_volumes/tasks/nfs.yml16
-rw-r--r--roles/kube_nfs_volumes/templates/nfs.json.j223
8 files changed, 444 insertions, 0 deletions
diff --git a/roles/kube_nfs_volumes/README.md b/roles/kube_nfs_volumes/README.md
new file mode 100644
index 000000000..965958bd6
--- /dev/null
+++ b/roles/kube_nfs_volumes/README.md
@@ -0,0 +1,111 @@
+# kube_nfs_volumes
+
+This role is useful to export disks as set of Kubernetes persistent volumes.
+It does so by partitioning the disks, creating ext4 filesystem on each
+partition, mounting the partitions, exporting the mounts via NFS and adding
+these NFS shares as NFS persistent volumes to existing Kubernetes installation.
+
+All partitions on given disks are used as the persistent volumes, including
+already existing partitions! There should be no other data (such as operating
+system) on the disks!
+
+## Requirements
+
+* Running Kubernetes with NFS persistent volume support (on a remote machine).
+
+* Works only on RHEL/Fedora-like distros.
+
+## Role Variables
+
+```
+# Options of NFS exports.
+nfs_export_options: "*(rw,no_root_squash,insecure,no_subtree_check)"
+
+# Directory, where the created partitions should be mounted. They will be
+# mounted as <mount_dir>/sda1 etc.
+mount_dir: /exports
+
+# Comma-separated list of disks to partition.
+# This role always assumes that all partitions on these disks are used as
+# physical volumes.
+disks: /dev/sdb,/dev/sdc
+
+# Whether to re-partition already partitioned disks.
+# Even though the disks won't get repartitioned on 'false', all existing
+# partitions on the disk are exported via NFS as physical volumes!
+foce: false
+
+# Specification of size of partitions to create. See library/partitionpool.py
+# for details.
+sizes: 100M
+
+# URL of Kubernetes API server, incl. port.
+kubernetes_url: https://10.245.1.2:6443
+
+# Token to use for authentication to the API server
+kubernetes_token: tJdce6Fn3cL1112YoIJ5m2exzAbzcPZX
+```
+
+## Dependencies
+
+None
+
+## Example Playbook
+
+With this playbook, `/dev/sdb` is partitioned into 100MiB partitions, all of
+them are mounted into `/exports/sdb<N>` directory and all these directories
+are exported via NFS and added as physical volumes to Kubernetes running at
+`https://10.245.1.2:6443`.
+
+ - hosts: servers
+ roles:
+ - role: kube_nfs_volumes
+ disks: "/dev/sdb"
+ sizes: 100M
+ kubernetes_url: https://10.245.1.2:6443
+ kubernetes_token: tJdce6Fn3cL1112YoIJ5m2exzAbzcPZX
+
+See library/partitionpool.py for details how `sizes` parameter can be used
+to create partitions of various sizes.
+
+## Full example
+Let's say there are two machines, 10.0.0.1 and 10.0.0.2, that we want to use as
+NFS servers for our Kubernetes cluster, running Kubernetes public API at
+https://10.245.1.2:6443.
+
+Both servers have three 1 TB disks, /dev/sda for the system and /dev/sdb and
+/dev/sdc to be partitioned. We want to split the data disks into 5, 10 and
+20 GiB partitions so that 10% of the total capacity is in 5 GiB partitions, 40%
+in 10 GiB and 50% in 20 GiB partitions.
+
+That means, each data disk will have 20x 5 GiB, 40x 10 GiB and 25x 20 GiB
+partitions.
+
+* Create an `inventory` file:
+ ```
+ [nfsservers]
+ 10.0.0.1
+ 10.0.0.2
+ ```
+
+* Create an ansible playbook, say `setupnfs.yaml`:
+ ```
+ - hosts: nfsservers
+ sudo: yes
+ roles:
+ - role: kube_nfs_volumes
+ disks: "/dev/sdb,/dev/sdc"
+ sizes: 5G:10,10G:40,20G:50
+ force: no
+ kubernetes_url: https://10.245.1.2:6443
+ kubernetes_token: tJdce6Fn3cL1112YoIJ5m2exzAbzcPZX
+ ```
+
+* Run the playbook:
+ ```
+ ansible-playbook -i inventory setupnfs.yml
+ ```
+
+## License
+
+Apache 2.0
diff --git a/roles/kube_nfs_volumes/defaults/main.yml b/roles/kube_nfs_volumes/defaults/main.yml
new file mode 100644
index 000000000..e296492f9
--- /dev/null
+++ b/roles/kube_nfs_volumes/defaults/main.yml
@@ -0,0 +1,10 @@
+---
+# Options of NFS exports.
+nfs_export_options: "*(rw,no_root_squash,insecure,no_subtree_check)"
+
+# Directory, where the created partitions should be mounted. They will be
+# mounted as <mount_dir>/sda1 etc.
+mount_dir: /exports
+
+# Force re-partitioning the disks
+force: false
diff --git a/roles/kube_nfs_volumes/handlers/main.yml b/roles/kube_nfs_volumes/handlers/main.yml
new file mode 100644
index 000000000..52f3ceffe
--- /dev/null
+++ b/roles/kube_nfs_volumes/handlers/main.yml
@@ -0,0 +1,3 @@
+---
+- name: restart nfs
+ service: name=nfs-server state=restarted
diff --git a/roles/kube_nfs_volumes/library/partitionpool.py b/roles/kube_nfs_volumes/library/partitionpool.py
new file mode 100644
index 000000000..1ac8eed4d
--- /dev/null
+++ b/roles/kube_nfs_volumes/library/partitionpool.py
@@ -0,0 +1,240 @@
+#!/usr/bin/python
+"""
+Ansible module for partitioning.
+"""
+
+# There is no pyparted on our Jenkins worker
+# pylint: disable=import-error
+import parted
+
+DOCUMENTATION = """
+---
+module: partitionpool
+short_description; Partition a disk into parititions.
+description:
+ - Creates partitions on given disk based on partition sizes and their weights.
+ Unless 'force' option is set to True, it ignores already partitioned disks.
+
+ When the disk is empty or 'force' is set to True, it always creates a new
+ GPT partition table on the disk. Then it creates number of partitions, based
+ on their weights.
+
+ This module should be used when a system admin wants to split existing disk(s)
+ into pools of partitions of specific sizes. It is not intended as generic disk
+ partitioning module!
+
+ Independent on 'force' parameter value and actual disk state, the task
+ always fills 'partition_pool' fact with all partitions on given disks,
+ together with their sizes (in bytes). E.g.:
+ partition_sizes = [
+ { name: sda1, Size: 1048576000 },
+ { name: sda2, Size: 1048576000 },
+ { name: sdb1, Size: 1048576000 },
+ ...
+ ]
+
+options:
+ disk:
+ description:
+ - Disk to partition.
+ size:
+ description:
+ - Sizes of partitions to create and their weights. Has form of:
+ <size1>[:<weigth1>][,<size2>[:<weight2>][,...]]
+ - Any <size> can end with 'm' or 'M' for megabyte, 'g/G' for gigabyte
+ and 't/T' for terabyte. Megabyte is used when no unit is specified.
+ - If <weight> is missing, 1.0 is used.
+ - From each specified partition <sizeX>, number of these partitions are
+ created so they occupy spaces represented by <weightX>, proportionally to
+ other weights.
+
+ - Example 1: size=100G says, that the whole disk is split in number of 100 GiB
+ partitions. On 1 TiB disk, 10 partitions will be created.
+
+ - Example 2: size=100G:1,10G:1 says that ratio of space occupied by 100 GiB
+ partitions and 10 GiB partitions is 1:1. Therefore, on 1 TiB disk, 500 GiB
+ will be split into five 100 GiB partition and 500 GiB will be split into fifty
+ 10GiB partitions.
+ - size=100G:1,10G:1 = 5x 100 GiB and 50x 10 GiB partitions (on 1 TiB disk).
+
+ - Example 3: size=200G:1,100G:2 says that the ratio of space occupied by 200 GiB
+ partitions and 100GiB partition is 1:2. Therefore, on 1 TiB disk, 1/3
+ (300 GiB) should be occupied by 200 GiB partitions. Only one fits there,
+ so only one is created (we always round nr. of partitions *down*). Teh rest
+ (800 GiB) is split into eight 100 GiB partitions, even though it's more
+ than 2/3 of total space - free space is always allocated as much as possible.
+ - size=200G:1,100G:2 = 1x 200 GiB and 8x 100 GiB partitions (on 1 TiB disk).
+
+ - Example: size=200G:1,100G:1,50G:1 says that the ratio of space occupied by
+ 200 GiB, 100 GiB and 50 GiB partitions is 1:1:1. Therefore 1/3 of 1 TiB disk
+ is dedicated to 200 GiB partitions. Only one fits there and only one is
+ created. The rest (800 GiB) is distributed according to remaining weights:
+ 100 GiB vs 50 GiB is 1:1, we create four 100 GiB partitions (400 GiB in total)
+ and eight 50 GiB partitions (again, 400 GiB).
+ - size=200G:1,100G:1,50G:1 = 1x 200 GiB, 4x 100 GiB and 8x 50 GiB partitions
+ (on 1 TiB disk).
+
+ force:
+ description:
+ - If True, it will always overwite partition table on the disk and create new one.
+ - If False (default), it won't change existing partition tables.
+
+"""
+
+# It's not class, it's more a simple struct with almost no functionality.
+# pylint: disable=too-few-public-methods
+class PartitionSpec(object):
+ """ Simple class to represent required partitions."""
+ def __init__(self, size, weight):
+ """ Initialize the partition specifications."""
+ # Size of the partitions
+ self.size = size
+ # Relative weight of this request
+ self.weight = weight
+ # Number of partitions to create, will be calculated later
+ self.count = -1
+
+ def set_count(self, count):
+ """ Set count of parititions of this specification. """
+ self.count = count
+
+def assign_space(total_size, specs):
+ """
+ Satisfy all the PartitionSpecs according to their weight.
+ In other words, calculate spec.count of all the specs.
+ """
+ total_weight = 0.0
+ for spec in specs:
+ total_weight += float(spec.weight)
+
+ for spec in specs:
+ num_blocks = int((float(spec.weight) / total_weight) * (total_size / float(spec.size)))
+ spec.set_count(num_blocks)
+ total_size -= num_blocks * spec.size
+ total_weight -= spec.weight
+
+def partition(diskname, specs, force=False, check_mode=False):
+ """
+ Create requested partitions.
+ Returns nr. of created partitions or 0 when the disk was already partitioned.
+ """
+ count = 0
+
+ dev = parted.getDevice(diskname)
+ try:
+ disk = parted.newDisk(dev)
+ except parted.DiskException:
+ # unrecognizable format, treat as empty disk
+ disk = None
+
+ if disk and len(disk.partitions) > 0 and not force:
+ print "skipping", diskname
+ return 0
+
+ # create new partition table, wiping all existing data
+ disk = parted.freshDisk(dev, 'gpt')
+ # calculate nr. of partitions of each size
+ assign_space(dev.getSize(), specs)
+ last_megabyte = 1
+ for spec in specs:
+ for _ in range(spec.count):
+ # create the partition
+ start = parted.sizeToSectors(last_megabyte, "MiB", dev.sectorSize)
+ length = parted.sizeToSectors(spec.size, "MiB", dev.sectorSize)
+ geo = parted.Geometry(device=dev, start=start, length=length)
+ filesystem = parted.FileSystem(type='ext4', geometry=geo)
+ part = parted.Partition(
+ disk=disk,
+ type=parted.PARTITION_NORMAL,
+ fs=filesystem,
+ geometry=geo)
+ disk.addPartition(partition=part, constraint=dev.optimalAlignedConstraint)
+ last_megabyte += spec.size
+ count += 1
+ try:
+ if not check_mode:
+ disk.commit()
+ except parted.IOException:
+ # partitions have been written, but we have been unable to inform the
+ # kernel of the change, probably because they are in use.
+ # Ignore it and hope for the best...
+ pass
+ return count
+
+def parse_spec(text):
+ """ Parse string with partition specification. """
+ tokens = text.split(",")
+ specs = []
+ for token in tokens:
+ if not ":" in token:
+ token += ":1"
+
+ (sizespec, weight) = token.split(':')
+ weight = float(weight) # throws exception with reasonable error string
+
+ units = {"m": 1, "g": 1 << 10, "t": 1 << 20, "p": 1 << 30}
+ unit = units.get(sizespec[-1].lower(), None)
+ if not unit:
+ # there is no unit specifier, it must be just the number
+ size = float(sizespec)
+ unit = 1
+ else:
+ size = float(sizespec[:-1])
+ spec = PartitionSpec(int(size * unit), weight)
+ specs.append(spec)
+ return specs
+
+def get_partitions(diskpath):
+ """ Return array of partition names for given disk """
+ dev = parted.getDevice(diskpath)
+ disk = parted.newDisk(dev)
+ partitions = []
+ for part in disk.partitions:
+ (_, _, pname) = part.path.rsplit("/")
+ partitions.append({"name": pname, "size": part.getLength() * dev.sectorSize})
+
+ return partitions
+
+
+def main():
+ """ Ansible module main method. """
+ module = AnsibleModule(
+ argument_spec=dict(
+ disks=dict(required=True, type='str'),
+ force=dict(required=False, default="no", type='bool'),
+ sizes=dict(required=True, type='str')
+ ),
+ supports_check_mode=True,
+ )
+
+ disks = module.params['disks']
+ force = module.params['force']
+ if force is None:
+ force = False
+ sizes = module.params['sizes']
+
+ try:
+ specs = parse_spec(sizes)
+ except ValueError, ex:
+ err = "Error parsing sizes=" + sizes + ": " + str(ex)
+ module.fail_json(msg=err)
+
+ partitions = []
+ changed_count = 0
+ for disk in disks.split(","):
+ try:
+ changed_count += partition(disk, specs, force, module.check_mode)
+ except Exception, ex:
+ err = "Error creating partitions on " + disk + ": " + str(ex)
+ raise
+ #module.fail_json(msg=err)
+ partitions += get_partitions(disk)
+
+ module.exit_json(changed=(changed_count > 0), ansible_facts={"partition_pool": partitions})
+
+# ignore pylint errors related to the module_utils import
+# pylint: disable=redefined-builtin, unused-wildcard-import, wildcard-import
+# import module snippets
+from ansible.module_utils.basic import *
+main()
+
diff --git a/roles/kube_nfs_volumes/meta/main.yml b/roles/kube_nfs_volumes/meta/main.yml
new file mode 100644
index 000000000..eb71a7a1f
--- /dev/null
+++ b/roles/kube_nfs_volumes/meta/main.yml
@@ -0,0 +1,16 @@
+---
+galaxy_info:
+ author: Jan Safranek
+ description: Partition disks and use them as Kubernetes NFS physical volumes.
+ company: Red Hat, Inc.
+ license: license (Apache)
+ min_ansible_version: 1.4
+ platforms:
+ - name: EL
+ versions:
+ - 7
+ - name: Fedora
+ versions:
+ - all
+ categories:
+ - cloud
diff --git a/roles/kube_nfs_volumes/tasks/main.yml b/roles/kube_nfs_volumes/tasks/main.yml
new file mode 100644
index 000000000..23b228d32
--- /dev/null
+++ b/roles/kube_nfs_volumes/tasks/main.yml
@@ -0,0 +1,25 @@
+---
+- name: Install pyparted (RedHat/Fedora)
+ yum: name=pyparted,python-httplib2 state=installed
+
+- name: partition the drives
+ partitionpool: disks={{ disks }} force={{ force }} sizes={{ sizes }}
+
+- name: create filesystem
+ filesystem: fstype=ext4 dev=/dev/{{ item.name }}
+ with_items: partition_pool
+
+- name: mount
+ mount: name={{mount_dir}}/{{ item.name }} src=/dev/{{ item.name }} state=mounted fstype=ext4 passno=2
+ with_items: partition_pool
+
+- include: nfs.yml
+
+- name: export physical volumes
+ uri: url={{ kubernetes_url }}/api/v1beta3/persistentvolumes
+ method=POST
+ body='{{ lookup("template", "../templates/nfs.json.j2") }}'
+ body_format=json
+ status_code=201
+ HEADER_Authorization="Bearer {{ kubernetes_token }}"
+ with_items: partition_pool
diff --git a/roles/kube_nfs_volumes/tasks/nfs.yml b/roles/kube_nfs_volumes/tasks/nfs.yml
new file mode 100644
index 000000000..87cf5f9a4
--- /dev/null
+++ b/roles/kube_nfs_volumes/tasks/nfs.yml
@@ -0,0 +1,16 @@
+---
+- name: Install NFS server on Fedora/Red Hat
+ yum: name=nfs-utils state=installed
+
+- name: Start rpcbind on Fedora/Red Hat
+ service: name=rpcbind state=started enabled=yes
+
+- name: Start nfs on Fedora/Red Hat
+ service: name=nfs-server state=started enabled=yes
+
+- name: Export the directories
+ lineinfile: dest=/etc/exports
+ regexp="^{{ mount_dir }}/{{ item.name }} "
+ line="{{ mount_dir }}/{{ item.name }} {{nfs_export_options}}"
+ with_items: partition_pool
+ notify: restart nfs
diff --git a/roles/kube_nfs_volumes/templates/nfs.json.j2 b/roles/kube_nfs_volumes/templates/nfs.json.j2
new file mode 100644
index 000000000..b42886ef1
--- /dev/null
+++ b/roles/kube_nfs_volumes/templates/nfs.json.j2
@@ -0,0 +1,23 @@
+{
+ "kind": "PersistentVolume",
+ "apiVersion": "v1beta3",
+ "metadata": {
+ "name": "pv-{{ inventory_hostname | regex_replace("\.", "-") }}-{{ item.name }}",
+ "labels": {
+ "type": "nfs"
+ }
+ },
+ "spec": {
+ "capacity": {
+ "storage": "{{ item.size }}"
+ },
+ "accessModes": [
+ "ReadWriteOnce"
+ ],
+ "NFS": {
+ "Server": "{{ inventory_hostname }}",
+ "Path": "{{ mount_dir }}/{{ item.name }}",
+ "ReadOnly": false
+ }
+ }
+}