Skip to content

Latest commit

 

History

History
163 lines (120 loc) · 6.71 KB

UPDATING.md

File metadata and controls

163 lines (120 loc) · 6.71 KB
Raw

Updating Information for BCPC clusters

These instructions assume that you basically know what you are doing. =)

20131026

The Ceph storage back-end was revamped to allow for multiple storage tiers, namely keeping SSD and HDD separate in ceph avoid performance hits. So for each node, there is no longer a node[:bcpc][:ceph_disks] parameter but instead node[:bcpc][:ceph][:ssd_disks] and node[:bcpc][:ceph][:hdd_disks]. Disks of each type are automatically added to their respective storage branches (see ceph osd tree after bringing up a cluster). Each disk will also have it's weight setting in CRUSH changed to match the drives size in TB (in floating point).

Also, in the BCPC cookbook's attributes/default.rb, there are new configurables to tune the replica count and the relative size of each pool BCPC creates (target making the portion attributes add up to ~100, but it is not a strict limit). As of Dumpling, both replica counts and pg_num are stable enough to change on-th-fly, so automatically calculate both based on the target number of placement groups per node. Roughly, the ratio of pgs:osds is 100:1, so use an appropriate number for node[:bcpc][:ceph][:pgs_per_node]based on the average number of disks in your hosts. NOTE: this means node[:bcpc][:ceph_node_count] and node[:bcpc][:ceph_s3_replica_count] have gone away.

Also, cinder has been configured to have multiple back-ends to take advantage of the two storage tiers, so now when allocating a volume, you can choose the "Type" of either SSD of HDD.

20131021

The Ceph S3 component relies on several pools which it will happily create using the incorrect defaults. In order to tell chef the correct sizing, you must supply your anticipated number of nodes in an attribute: node[:bcpc][:ceph_node_count]

It is defaulted to 3, if you want it higher, a recommended spot would be an environment file.

node[:bcpc][:ceph_s3_replica_count] can also be set if you want a number of replicas different than 3

20131003

Keystone is now serviced over HTTPS on port 5000. In particular, a new subjectAltName field containing the VIP IP address was added to the SSL certificate.

On all headnodes:

# service nova-api stop
# service keystone stop
# service apache2 stop

On one headnode:

# . /root/keystonerc
# keystone service-delete keystone

Remove the ssl-certificate and ssl-private-key fields from the databag - eg:

# EDITOR=vi knife data bag edit configs Test-Laptop

Then, you can run chef-client to regenerate and redeploy the new certs.

20131002

To upgrade Power DNS to ...: Warning: only '&', '_' and are translated for project names to domain names.

Easiest, unless you put custom information in:

  • pdns.records_static
  • pdns.domains
  • The entire pdns database

Simply run DROP DATABASE pdns and re-run chef-client.

If you have custom DNS data in your pdns database, you can roughly do the following:

  • Migrate MySQL tables and DB to use UTF8 instead of Sweedish collation in the pdns DB:
ALTER DATABASE pdns DEFAULT COLLATE utf8_general_ci;
ALTER TABLE records_static CHARACTER SET utf8 COLLATE utf8_general_ci;
ALTER TABLE records_static CONVERT TO CHARACTER SET utf8 COLLATE utf8_general_ci;
ALTER TABLE domains CHARACTER SET utf8 COLLATE utf8_general_ci;
ALTER TABLE domains CONVERT TO CHARACTER SET utf8 COLLATE utf8_general_ci;
ALTER TABLE domains RENAME TO domains_static;
DROP VIEW records;

Then re-run chef-client to get the new required MySQL views and function.

20130928

On Ubuntu nodes, network interfaces are now templated rather than hard-coded upon first-run. Delete the relevant entries from /etc/network/interfaces and it will be replaced by the templated files in /etc/network/interfaces.d/

Beaver (on all head/worker nodes) and logstash (head nodes only) has been replaced by fluentd. You can safely remove those packages now.

20130824

Automatic boot-from-volumes patches in Grizzly is added back in to the tree. You must upload RAW images again. For example, to convert the public Ubuntu 12.04 QCOW2 image to RAW:

$ curl -O http://cloud-images.ubuntu.com/releases/precise/release/ubuntu-12.04-server-cloudimg-amd64-disk1.img
$ qemu-img convert -O raw ubuntu-12.04-server-cloudimg-amd64-disk1.img ubuntu-12.04-server-cloudimg-amd64-disk1.raw
$ glance image-create --name='Ubuntu 12.04 x86_64' --is-public=True --container-format=bare --disk-format=raw --file ubuntu-12.04-server-cloudimg-amd64-disk1.raw

Apache httpd site configurations were switched to use the site model rather than place server vhost definitions in conf.d. To update an existing cluster, please run the following on each head node before running chef-client:

# rm /etc/apache2/conf.d/openstack-dashboard.conf /etc/apache2/conf.d/zabbix-web.conf /etc/apache2/conf.d/graphite-web.conf

20130817

Due to a change in Vagrant 1.2.5+, the default bootstrap node cannot have its IP end in .1 (ie, 10.0.100.1). In the current Vagrant scripts, the automated bootstrap node will be 10.0.100.3. (The host machine stays at 10.0.100.2.)

The default Ceph version has been upgraded from Cuttlefish (0.61) to Dumpling (0.67).

Follow rolling upgrade notes described in: http://ceph.com/docs/master/release-notes/#v0-67-dumpling

  • Upgrade ceph-common on all nodes that will use the command line ceph utility.
  • Upgrade all monitors (upgrade ceph package, restart ceph-mon daemons). This can happen one daemon or host at a time. Note that because cuttlefish and dumpling monitors can't talk to each other, all monitors should be upgraded in relatively short succession to minimize the risk that an a untimely failure will reduce availability.
  • Upgrade all osds (upgrade ceph package, restart ceph-osd daemons). This can happen one daemon or host at a time.
  • Upgrade radosgw (upgrade radosgw package, restart radosgw daemons).

In Dumpling, the client.bootstrap-osd cephx key has new capabilities (see ceph auth list). Cuttlefish required the following monitor capabilities:

mon 'allow command osd create ...; allow command osd crush set ...; allow command auth add * osd allow\\ * mon allow\\ rwx; allow command mon getmap'

In Dumpling, it is now reduced to:

mon 'allow profile bootstrap-osd'

BCPC now uses the standard Opscode Community ntp cookbook instead of rolling our own ntp recipes. Environments need to be updated from:

"bcpc": {
  ...
  "ntp_servers" : [ "0.pool.ntp.org", "1.pool.ntp.org", "2.pool.ntp.org", "3.pool.ntp.org" ],
  ...
}

to (at the outer level - akin to bcpc, chef_client, and ubuntu):

"ntp": {
   "servers" : [ "0.pool.ntp.org", "1.pool.ntp.org", "2.pool.ntp.org", "3.pool.ntp.org" ]
},