validate: allow running ceph-ansible 3.2 against ansible 2.7
This commit allows ceph-ansible 3.2 to be run against ansible 2.7
However, note that running stable-3.2 against ansible 2.7 doesn't get
any testing upstream this might break the playbook, only ansible 2.6 is
officially supported.
Dimitri Savineau [Tue, 28 Jan 2020 15:27:34 +0000 (10:27 -0500)]
ceph-defaults: remove rgw from ceph_conf_overrides
The [rgw] section in the ceph.conf file or via the ceph_conf_overrides
variable doesn't exist and has no effect.
To apply overrides to all radosgw instances we should use either the
[global] or [client] sections.
Overrides per radosgw instance should still use the
[client.rgw.{instance-name}] section.
To avoid confusion, let's change the default value from `0.0.0.0` to
`x.x.x.x`.
Users might think setting `0.0.0.0` will make the daemon binding on all
interfaces.
This commit adds a playbook to be played before we run purge playbook,
it first creates an rbd image then map an rbd device on client0 so the
purge playbook will try to unmap it.
in containerized context, using the binary provided in atomic os won't
work because it's an old version provided by ceph-common based on
10.2.5.
Using a container could be an idea but for large cluster with hundreds
of client nodes, that would require to pull the image of each of them
just to unmap the rbd devices.
Let's use the sysfs method in order to avoid any issue related to ceph
version that is shipped on the host.
Dimitri Savineau [Wed, 27 Nov 2019 14:29:06 +0000 (09:29 -0500)]
ceph-osd: wait for all osds once
cf8c6a3 moves the 'wait for all osds' task from openstack_config to the
main tasks list.
But the openstack_config code was executed only on the last OSD node.
We don't need to do this check on all OSD node so we need to add set
run_once to true on that task.
Dimitri Savineau [Tue, 26 Nov 2019 16:09:11 +0000 (11:09 -0500)]
ceph-osd: wait for all osd before crush rules
When creating crush rules with device class parameter we need to be sure
that all OSDs are up and running because the device class list is
is populated with this information.
This is now enable for all scenario not openstack_config only.
Dimitri Savineau [Thu, 31 Oct 2019 20:24:12 +0000 (16:24 -0400)]
ceph-osd: add device class to crush rules
This adds device class support to crush rules when using the class key
in the rule dict via the create-replicated sub command.
If the class key isn't specified then we use the create-simple sub
command for backward compatibility.
Dimitri Savineau [Thu, 31 Oct 2019 20:17:33 +0000 (16:17 -0400)]
move crush rule creation from mon to osd role
If we want to create crush rules with the create-replicated sub command
and device class then we need to have the OSD created before the crush
rules otherwise the device classes won't exist.
Dimitri Savineau [Mon, 16 Dec 2019 21:41:20 +0000 (16:41 -0500)]
switch_to_containers: set GUID on lockbox part
The ceph lockbox partition (part number 5) used with non lvm scenarios
and in non containerized deployment don't have a valid PARTUUID.
The value is set to 00000000-0000-0000-0000-000000000000 for each OSD
devices.
When switching to containerized deployment we manually mount the lockbox
partition by using the PARTUUID.
Unfortunately because we have most of the time multiple OSD on the same
node we can't have the right symlink in /dev/disk/by-partuuid because it
will point to only one partition.
After the switch_to_containers playbook then only one OSD will restart
correctly and the other will try to access to the wrong device causing
error like 'xxxx is still in use'.
When deploying with containers and dmcrypt OSDs we force a PARTUUID
value during the ceph-disk prepare task.
When using `osd_auto_discovery`, `devices` is built multiple times due
to multiple runs of `ceph-facts` role. It end up with duplicate
instances of a same device in the list.
Using `unique` filter when building the list fixes this issue.
When using fqdn in inventory, that playbook fails because of some tasks
using the result of ceph osd tree (which returns shortname) to get
some datas in hostvars[].
This commit adds the support of the ceph-iscsi stable repository when
use ceph_repository community instead of always using the devel
repositories.
We're still using the devel repositories for rtslib and tcmu-runner in
both cases (dev and community).
ansible.cfg: do not enforce PreferredAuthentications
There's no need to enforce PreferredAuthentications by default.
Users can still choose to override the ansible.cfg with any additional
parameter like this one to fit their infrastructure.
The systemd unit script wasn't updated with the new container name
format (without the hostname).
We now have the same start/stop docker commands for all scenarios.
During the device to id OSD migration we need to be sure that the
old container with the hostname are stopped.
the previous approach was wrong.
checking if `item.key` is in `osd_auto_discovery_exclude` (`['dm-',
'loop']`) is incorrect because it will obviously not match. Therefore,
the condition will return `True` whatever the device we are checking.
Dimitri Savineau [Wed, 27 Nov 2019 16:27:09 +0000 (11:27 -0500)]
switch_to_containers: fix umount ceph partitions
When a container is already running on a non containerized node then the
umount ceph partition task is skipped.
This is due to the container ps command which always returns 0 even if
the filter matches nothing.
We should run the umount task when:
1/ the container command is failing (not installed) : rc != 0
2/ the container command reports running ceph-osd containers : rc == 0
Also we should not fail on the ceph directory listing.
Dimitri Savineau [Wed, 20 Nov 2019 19:40:52 +0000 (14:40 -0500)]
rolling_update: don't enable ceph-mon unit
On non containerized deployment the ceph-mon hostname/fqdn systemd
service are stopped at the beginning of the mon upgrade.
But the parameter enabled is set to true for both task so even if we're
not using the fqdn then it will enabled the systemd unit based on it.
Dimitri Savineau [Thu, 14 Nov 2019 14:29:29 +0000 (09:29 -0500)]
container: add always tag on gather fact tasks
If we execute the site-container.yml playbook with specific tags (like
ceph_update_config) then we need to be sure to gather the facts otherwise
we will see error like:
The task includes an option with an undefined variable. The error was:
'ansible_hostname' is undefined
This commit also adds missing 'gather_facts: false' to mons plays.
ceph_key: restore file mode after a key is fetched
when `import_key` is enabled, if the key already exists, it will only be
fetched using ceph cli, if the mode specified in the `ceph_key` task is
different from what is applied by the ceph cli, the mode isn't restored because
we don't call `module.set_fs_attributes_if_different()` before
`module.exit_json(**result)`
[1] introduced a regression on the fs.aio-max-nr sysctl value condition.
The enable key isn't a boolean but a string because the expression isn't
evaluated.
This string output "(osd_objectstore == 'bluestore')" is always true
because item.enable condition only matches non empty string. So the
sysctl value was applyied for both filestore and bluestore backend.
[2] added the bool filter to the condition but the filter always returns
false on string and the sysctl wasn't applyed at all.
This commit fixes the enable key value by evaluating the value instead
of using the string.
Harald Jensås [Fri, 6 Sep 2019 14:24:30 +0000 (16:24 +0200)]
Support comma-delimited subnets in firewall
ceph.conf supports a comma separated list of
subnet CIDR's for the public_network and the
cluster network. ceph-ansible should support
setting up the firewall for this configuration.
Dimitri Savineau [Wed, 30 Oct 2019 15:45:44 +0000 (11:45 -0400)]
ceph-osd: Remove ulimit nofile on container start
Even if this improves ceph-disk/ceph-volume performances then it also
impact the ceph-osd process.
The ceph-osd process shouldn't use 1024:4096 value for the max open
files.
Removing the ulimit option from the container engine and doing this kind
of change on the container side [1].
This commit adds a default value in the with_dict because when using
python 2.7, if a task using a with_dict has a condition, it is
evaluated anyway whereas in python 3 it isn't.
Dimitri Savineau [Fri, 25 Oct 2019 20:47:50 +0000 (16:47 -0400)]
rolling_update: fix active mds host value
The active mds host should be based on the inventory hostname and not on
the ansible hostname.
The value returns under the mdsmap structure is based on the OS hostname
so we need to find the right node in the inventory with this value when
doing operation on inventory nodes.
Othewise we could see error like:
The task includes an option with an undefined variable. The error was:
"hostvars[foobar]" is undefined
update: follow new recommandation to upgrade mds cluster
Refact the mds cluster upgrade code in order to follow the documented
recommandation.
See: https://github.com/ceph/ceph/blob/luminous/doc/cephfs/upgrading.rst
Dimitri Savineau [Thu, 17 Oct 2019 18:28:45 +0000 (14:28 -0400)]
tests: fix the size on the second data LV
The commit replaces the pv/vg/lv commands used with the ansible command
module by the lvg and lvol modules.
This also fixes the size of the second data LV because we were only using
50% of the remaining space instead of 100%.
With a 50G device, the result was:
- data-lv1 was 25G
- data-lv2 was 12.5G
Instead of:
- data-lv1 was 25G
- data-lv2 was 25G
common: do not override ceph_release when using custom repo
Otherwise it fails like following:
```
TASK [ceph-mds : allow multimds] **************************************************************************************************************************************************
Monday 22 July 2019 16:37:38 +0800 (0:00:03.269) 0:13:25.651 ***********
fatal: [rhel7u6clone1]: FAILED! => {"msg": "The conditional check 'ceph_release_num[ceph_release] == ceph_release_num.luminous' failed. The error was: error while evaluating conditional (ceph_release_num[ceph_release] == ceph_release_num.luminous): 'dict object' has no attribute u'dummy'\n\nThe error appears to have been in '/usr/share/ceph-ansible/roles/ceph-mds/tasks/create_mds_filesystems.yml': line 43, column 3, but may\nbe elsewhere in the file depending on the exact syntax problem.\n\nThe offending line appears to be:\n\n\n- name: allow multimds\n ^ here\n"}
```
Dimitri Savineau [Tue, 15 Oct 2019 15:32:40 +0000 (11:32 -0400)]
rbd-mirror: fail if the peer is not added
Due the 'failed_when: false' statement present in the peer task then
the playbook continues to ran even if the peer task was failing (like
incorrect remote peer format.
"stderr": "rbd: invalid spec 'admin@cluster1'"
This patch adds a task to list the peer present and add the peer only if
it's not already added. With this we don't need the failed_when statement
anymore.
Dimitri Savineau [Fri, 11 Oct 2019 15:42:36 +0000 (11:42 -0400)]
Remove validate action and notario dependency
The current ceph-validate role is using both validate action and fail
module tasks to validate the ceph configuration.
The validate action is based on the notario python library. When one of
the notario validation fails then a python stack trace is reported to the
ansible task. This output isn't understandable by users.
This patch removes the validate action and the notario depencendy. The
validation is now done with only fail ansible module.
The secondary vagrant variables didn't have the grafana vm variable
set which create an vagrant error.
There was an error loading a Vagrantfile. The file being loaded
and the error message are shown below. This is usually caused by
an invalid or undefined variable.
This patch also changes the ssh-extra-args parameter to ssh-common-args
to get the same values for ssh/sftp/scp. Otherwise we can see warnings
from ansible and some tasks are failing.
[WARNING]: sftp transfer mechanism failed on [mon0]. Use ANSIBLE_DEBUG=1
to see detailed information
It also updates the ssh-common-args value for the rgw-multisite scenario
to reflect the ANSIBLE_SSH_ARGS environment variable value.
Finally changing the IP addresses due to the Vagrant refact done in the
commit 778c51a
When switching from a baremetal deployment to a containerized deployment
we only umount the OSD data partition.
If the OSD is encrypted (dmcrypt: true) then there's an additional
partition (part number 5) used for the lockbox and mount in the
/var/lib/ceph/osd-lockbox/ directory.
Because this partition isn't umount then the containerized OSD aren't
able to start. The partition is still mount by the system and can't be
remount from the container.
9e7972a introduced a regression via the container_binary variable
which is undefined.
The CEPH_CONTAINER_BINARY environment variable isn't used at all.
556052b changed the way the mgr keyring are created but the ceph_key
module need the containerized parameter when the deployment is using
containers.
This module doesn't support CEPH_CONTAINER_[BINARY|IMAGE] environment
variables.
When using the ansible --limit option on one or few OSD nodes and if the
handler is triggered then we will restart the OSD service on all OSDs
nodes instead of the hosts limited by the limit value.
Even if the play is limited by the --limit value we are using all OSD
nodes from the OSD group.
with_items: '{{ groups[osd_group_name] }}'
Instead we should iterate only on the nodes present in both OSD group and
limit list.
Vagrantfile: support more than 9 nodes per daemon type
because of the current ip address assignation, it's not possible to
deploy more than 9 nodes per daemon type.
This commit refact a bit and allows us to get around this limitation.
Kevin Jones [Sat, 10 Aug 2019 19:44:32 +0000 (15:44 -0400)]
Set proper ownership command performance improvement
By changing the set ownership command from using the file module in combination with a with_items loop to a raw chown command, we can achieve a 98% performance increase here.
On a ceph cluster with a significant amount of directories and files in /var/lib/ceph, the file module has to run checks on ownership of all those directories and files to determine whether a change is needed.
In this case, we just want to explicitly set the ownership of all these directories and files to the ceph_uid
Added context note to all set proper ownership tasks
Since we change the way to run the OSD containers with the ID instead
of the device name, we lost the ability to use loop devices.
Loop devices are like nvme or cciss devices because the partitions are
referenced with an extra 'p' before the partition number.
In containerized deployment, the restart OSD handler couldn't be
triggered in most ansible execution.
This is due to the usage of run_once + a condition on the inventory
hostname and the last filter.
The run_once is triggered first so ansible will pick a node in the
osd group to execute the restart task. But if this node isn't the
last one in the osd group then the task is ignored. There's more
probability that the task will be ignored than executed.
The admin keyring isn't present by default on the rbd mirror nodes so
the rbd commands related to the mirroring confguration will fail.
Instead we can use the rbd mirror client keyring.
The rbd mirror configuration was only available for non containerized
deployment and was also imcomplete.
We now enable the mirroring on the pool and add the remote peer in both
scenarios.
The default mirroring mode is set to 'pool' but can be configured via
the ceph_rbd_mirror_mode variable.
This commit also fixes an issue on the rbd mirror command if the ceph
cluster name isn't using the default value (ceph) due to a missing
--cluster parameter to the command.
Dimitri Savineau [Fri, 30 Aug 2019 18:36:12 +0000 (14:36 -0400)]
tests: update dedidated mgr node all_daemons
5b29144 change the mgr node to a dedicated node instead of the first
monitor node.
But the change didn't update the switch-to-containers inventory which
cause this playbook to fail.
Also update the ubuntu inventory to have the same configuration.
We don't have a reason to not apply firewall rules on the host when
using a containerized deployment.
The TripleO environments already manage the ceph firewall rules outside
ceph-ansible and set the configure_firewall variable to false.
The "osd blacklist" isn't an osd caps but should be used with mon caps.
Also the correct caps for this is: 'allow command "osd blacklist"'.
The current change is breaking the openstack and clients keyrings.
By using the profile rbd (which is already used) we already rely on the
ability to blacklist dead client.
On containerized deployment, the OSD entrypoint runs some ceph-volume
commands (lvm/simple scan and/or activate) which perform badly without
the ulimit option.
This option was added for all previous ceph-volume commands but not on
the ceph-osd container startup.
Also updating hard limit value to 4096 to reflect default baremetal
value.
validate: fail if gpt header found on unprepared devices
ceph-volume will complain if gpt headers are found on devices.
This commit checks whether a gpt header is present on devices passed in
`devices` variable and fail early.
update: use ids to restart osds instead of device name
we must use the ids instead of device names in the tasks executed in
`post_tasks` for the osd rolling update otherwise it ends up with old
systemd units enabled.
osd: copy systemd-device-to-id.sh on all osd nodes before running it
Otherwise it will fail when running rolling_update.yml playbook because
of `serial: 1` usage.
The task which copies the script is run against the current node being
played only whereas the task which runs the script is run against all
nodes in a loop, it ends up with the typical error:
Since bedc0ab we now manage ceph-osd systemd unit scripts based on ID
instead of device name but it was not present in the shrink-osd
playbook (ceph-disk version).
To keep backward compatibility on deployment that didn't do yet the
transition on OSD id then we should stop unit scripts for both device
and ID.
This commit adds the ulimit nofile container option to get better
performance on ceph-disk commands.
It also fixes an issue when the OSD id matches multiple OSD ids with
the same first digit.
Dimitri Savineau [Tue, 26 Feb 2019 14:16:37 +0000 (09:16 -0500)]
rgw: add beast frontend
Allow to configure the rgw beast frontend in addition to civetweb
(default value).
Add rgw_thread_pool_size variable with 512 as default value and keep
backward compatibility with num_threads option when using civetweb.
Update radosgw_civetweb_num_threads to reflect rgw_thread_pool_size
change.
When creating OpenStack pools, we only check if the return code from
the pool list command isn't 0 (ie: if it doesn't exist). In that case,
the return code will be 2. That's why the next condition is rc != 0 for
the pool creation.
But in containerized deployment, the return code could be different if
there's a failure on the container engine command (like container not
running). In that case, the return code could but either 1 (docker) or
125 (podman) so we should fail at this point and not in the next tasks.
projects / ceph-ansible.git / log