if we don't assign the rbd application tag on this pool,
the cluster will get `HEALTH_WARN` state like following:
```
HEALTH_WARN application not enabled on 1 pool(s)
POOL_APP_NOT_ENABLED application not enabled on 1 pool(s)
application not enabled on pool 'rbd'
```
We're using fuser command to see if a process is using a ceph unix
socket file. But the fuser command runs through every PID present in
/proc/<PID> to see if one of them is using the file.
On a system running thousands processes, the fuser command can take
a long time to finish.
The current lvm_osds only tests filestore on one OSD node.
We also have bs_lvm_osds to test bluestore and encryption.
Let's use only one scenario to test filestore/bluestore and with or
without dmcrypt on four OSD nodes.
Also use validate_dmcrypt_bool_value instead of types.boolean on
dmcrypt validation via notario.
Mike Christie [Thu, 9 May 2019 19:52:08 +0000 (14:52 -0500)]
igw: Fix rolling update service ordering
We must stop tcmu-runner after the other rbd-target-* services
because they may need to interact with tcmu-runner during shutdown.
There is also a bug in some kernels where IO can get stuck in the
kernel and by stopping rbd-target-* first we can make sure all IO is
flushed.
We only need to set the wal dedicated device when there's three tiers
of storage used.
Currently the block.wal partition will also be created on the same
device than block.db.
Currently only rbd-target-gw service is restarted during an update.
We also need to restart tcmu-runner and rbd-target-api services
during the ceph iscsi upgrade.
In containerized deployment the default mds cpu quota is too low
for production environment.
This is causing performance degradation compared to bare-metal.
In containerized deployment the default osd cpu quota is too low
for production environment using NVMe devices.
This is causing performance degradation compared to bare-metal.
Matthew Vernon [Fri, 12 Apr 2019 15:36:12 +0000 (16:36 +0100)]
ceph-mon: increase timeout waiting for admin and bootstrap keys
With a large and/or busy cluster, it can take significantly more than
30s for a restarted monitor to get to the point where
`ceph-create-keys` returns successfully. A recent upgrade of our
production cluster failed here because it took a couple of minutes for
the newly-upgraded `mon` to be ready. So increase the timeout
significantly.
This patch is applied to stable-3.2, because the affected code is
refactored in stable-4.0 and ceph-create-keys is no longer called.
We don't need to have multiple ceph-override.json copies. We
currently already have symlink to all_daemons/ceph-override.json so
we can do it for all scenarios.
We don't need to use the cephfs variable for the application pool
name because it's always cephfs.
If the cephfs variable is set to something else than the default
value it will break the appplication pool task.
Matthew Vernon [Wed, 27 Mar 2019 13:34:47 +0000 (13:34 +0000)]
UCA: Uncomment UCA variables in defaults, fix consequent breakage
The Ubuntu Cloud Archive-related (UCA) defaults in
roles/ceph-defaults/defaults/main.yml were commented out, which means
if you set `ceph_repository` to "uca", you get undefined variable
errors, e.g.
```
The task includes an option with an undefined variable. The error was: 'ceph_stable_repo_uca' is undefined
The error appears to have been in '/nfs/users/nfs_m/mv3/software/ceph-ansible/roles/ceph-common/tasks/installs/debian_uca_repository.yml': line 6, column 3, but may
be elsewhere in the file depending on the exact syntax problem.
The offending line appears to be:
- name: add ubuntu cloud archive repository
^ here
```
Unfortunately, uncommenting these results in some other breakage,
because further roles were written that use the fact of
`ceph_stable_release_uca` being defined as a proxy for "we're using
UCA", so try and install packages from the bionic-updates/queens
release, for example, which doesn't work. So there are a few `apt` tasks
that need modifying to not use `ceph_stable_release_uca` unless
`ceph_origin` is `repository` and `ceph_repository` is `uca`.
When using monitor_address_block or radosgw_address_block variables
to configure the mon/rgw address we're getting the first ip address
from the ansible facts present in that cidr.
When there's VIP on that network the first filter could return the
wrong value.
This seems to affect only IPv6 setup because the VIP addresses are
added to the ansible facts at the beginning of the list. This is the
opposite (at the end) when using IPv4.
This causes the mon/rgw processes to bind on the VIP address.
In containerized deployment the default radosgw quota is too low
for production environment.
This is causing performance degradation compared to bare-metal.
Dimitri Savineau [Wed, 20 Mar 2019 17:40:33 +0000 (13:40 -0400)]
purge-docker-cluster: Remove ceph-osd service
The systemd ceph-osd@.service file used for starting the ceph osd
containers is used in all osd_scenarios.
Currently purging a containerized deployment using the lvm scenario
didn't remove the ceph-osd systemd service.
If the next deployment is a non-containerized deployment, the OSDs
won't be online because the file is still present and override the
one from the package.
Dimitri Savineau [Fri, 22 Mar 2019 19:03:15 +0000 (15:03 -0400)]
Add uca to ceph_repository choices validation
Ubuntu cloud archive is configurable via ceph_repository variable but
the uca choice isn't accepted.
This commit fixes this issue and also validates the associated uca
repository variables.
defaults: change default value for ceph_docker_image_tag
Since nautilus has been released, it's now the latest stable release, it
means the tag `latest` now refers to nautilus.
`stable-3.2` isn't intended to deploy nautilus, therefore, we should
change the default value for this variable to the latest release
stable-3.2 is able to deploy (mimic).
Dimitri Savineau [Wed, 20 Mar 2019 19:30:46 +0000 (15:30 -0400)]
ceph-osd: Ensure lvm2 is installed
When using osd_scenario lvm, we never check if the lvm2 package is
present on the host.
When using containerized deployment and docker on CentOS/RedHat this
package will be automatically installed as a dependency but not for
Ubuntu distribution.
OSD deployed via ceph-volume require the lvmetad.socket to be active
and running.
osd: backward compatibility with old disk_list.sh location
Since all files in container image have moved to `/opt/ceph-container`
this check must look for new AND the old path so it's backward
compatible. Otherwise it could end up by templating an inconsistent
`ceph-osd-run.sh`.
Dimitri Savineau [Thu, 14 Mar 2019 20:22:01 +0000 (16:22 -0400)]
ceph-validate: fail if there's no ipaddr available in monitor_address_block subnet
When using monitor_address_block to determine the ip address of the
monitor node, we need an ip address available in that cidr to be
present in the ansible facts (ansible_all_ipv[46]_addresses).
Currently we don't check if there's an ip address available during
the ceph-validate role.
As a result, the ceph-config role fails due to an empty list during
ceph.conf template creation but the error isn't explicit.
TASK [ceph-config : generate ceph.conf configuration file] *****
fatal: [0]: FAILED! => {"msg": "No first item, sequence was empty."}
With this patch we will fail before the ceph deployment with an
explicit failure message.
Gregory Orange [Mon, 18 Mar 2019 01:50:41 +0000 (09:50 +0800)]
Change docker_container parameter network to network_mode
Addressing "populate kv_store with custom ceph.conf":
Unsupported parameters for (docker_container) module. Looking at
https://docs.ansible.com/ansible/latest/modules/docker_container_module.html
shows that the correct parameter is network_mode, not network.
Currently the default crush rule value is added to the ceph config
on the mon nodes as an extra configuration applied after the template
generation via the ansible ini module.
This implies two behaviors:
1/ On each ceph-ansible run, the ceph.conf will be regenerated via
ceph-config+template and then ceph-mon+ini_file. This leads to a
non necessary daemons restart.
2/ When other ceph daemons are collocated on the monitor nodes
(like mgr or rgw), the default crush rule value will be erased by
the ceph.conf template (mon -> mgr -> rgw).
This patch adds the osd_pool_default_crush_rule config to the ceph
template and only for the monitor nodes (like crush_rules.yml).
The default crush rule id is read (if exist) from the current ceph
configuration.
The default configuration is -1 (ceph default).
Dimitri Savineau [Mon, 11 Mar 2019 14:44:47 +0000 (10:44 -0400)]
ceph-osd: Install numactl package when needed
With 3e32dce we can run OSD containers with numactl support.
When using numactl command in a containerized deployment we need to
be sure that the corresponding package is installed on the host.
The package installation is only executed when the
ceph_osd_numactl_opts variable isn't empty.
As of testinfra 2.0.0, the binary name is `py.test`.
But let's pin the version to 1.19.0.
Indeed, migrating to 2.0.0 requires our current testing to be reworked a bit.
Since we don't have the bandwidth ATM for this, it's better to simply
keep testing with testinfra 1.19.0.
Note that I've replaced all `testinfra` occurences by `py.test` anyway.
removing the content of this directory seems a bit agressive and cause a
redeployment to fail after a purge on debian based distrubition.
Typical error:
```
fatal: [mon0]: FAILED! => changed=false
attempts: 3
msg: No package matching 'ceph' is available
```
The following task will consider the cache is still valid, so apt
doesn't refresh it:
```
- name: update apt cache if cache_valid_time has expired
apt:
update_cache: yes
cache_valid_time: 3600
register: result
until: result is succeeded
```
since the task installing ceph packages has a `update_cache: no` it
fails:
```
- name: install ceph for debian
apt:
name: "{{ debian_ceph_pkgs | unique }}"
update_cache: no
state: "{{ (upgrade_ceph_packages|bool) | ternary('latest','present') }}"
default_release: "{{ ceph_stable_release_uca | default('') }}{{ ansible_distribution_release ~ '-backports' if ceph_origin == 'distro' and ceph_use_distro_backports else '' }}"
register: result
until: result is succeeded
```
/tmp/* isn't specific to ceph as well, so we shouldn't remove everything
in this directory.
using `shell` module seems to be the only way to make this task working
on rhel based distribution AND debian based distributions.
on ubuntu, using `command` ansible module fails like following
(not due to `sudo` usage or not):
```
ok: [osd1] => changed=false
cmd: command -v ceph-volume
failed_when_result: false
msg: '[Errno 2] No such file or directory: ''command'': ''command'''
rc: 2
```
ToprHarley [Mon, 18 Feb 2019 18:02:03 +0000 (19:02 +0100)]
Convert interface names to underscores
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1540881 Signed-off-by: Tomas Petr <tpetr@redhat.com>
(cherry picked from commit 573adce7dd4f306c384b3308c8049ae49ef59716)
Kevin Coakley [Tue, 26 Feb 2019 17:30:31 +0000 (09:30 -0800)]
Set permissions on monitor directory to u=rwX,g=rX,o=rX recursive
Set directories to 755 and files to 644 to
/var/lib/ceph/mon/{{ cluster }}-{{ monitor_name }} recursively instead of
setting files and directories to 755 recursively. The ceph mon
process writes files to this path with permissions 644. This update stops
ansible from updating the permissions in
/var/lib/ceph/mon/{{ cluster }}-{{ monitor_name }} every time ceph mon writes
a file and increases idempotency.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1683997 Signed-off-by: Kevin Coakley <kcoakley@sdsc.edu>
(cherry picked from commit d327681b99915578fc8b389fda69556966db905f)
Dimitri Savineau [Wed, 27 Feb 2019 16:07:38 +0000 (11:07 -0500)]
mon: Add mds permissions to client.admin
The administrator keyring needs full capabilities on mds like mon,
osd and mgr.
Whithout this, the client.admin key won't be able to run commands
against mds (like ceph tell mds.0 session ls)
osd: make the 'wait for all osd to be up' task configurable
introduce two new variables to make the check that 'wait for all osd to
be up' configurable.
It's possible that for some deployments, OSDs can take longer to be seen
as UP and IN.
David Waiting [Mon, 10 Dec 2018 14:54:18 +0000 (09:54 -0500)]
ensure at least one osd is up
The existing task checks that the number of OSDs is equal to the number of up OSDs before continuing.
The problem is that if none of the OSDs have been discovered yet, the task will exit immediately and subsequent pool creation will fail (num_osds = 0, num_up_osds = 0).
In this change, we also check that at least one OSD is present. In our testing, this results in the task correctly waiting for all OSDs to come up before continuing.
setup_ntp: call handler to disable ntpd if chronyd used
The task setup chronyd called the handler disable chronyd, which of
course defeats the purpose.
Changing the task to disable ntpd instead fixes the issue of chronyd
being disabled after it got enabled.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1673664 Fixes: #3582 Signed-off-by: Patrick C. F. Ernzer pcfe@redhat.com
(cherry picked from commit c605ff6a68720ab43b63086c3ac1d529a651f585)
Typical error:
```
fatal: [iscsi-gw0]: FAILED! =>
msg: 'an error occurred while trying to read the file ''/home/guits/ceph-ansible/tests/functional/all_daemons/fetch/e5f4ab94-c099-4781-b592-dbd440a9d6f3/iscsi-gateway.key'': [Errno 13] Permission denied: b''/home/guits/ceph-ansible/tests/functional/all_daemons/fetch/e5f4ab94-c099-4781-b592-dbd440a9d6f3/iscsi-gateway.key'''
```
`become: True` is not needed on the following task:
`copy crt file(s) to gateway nodes`.
Since it's already set in the main playbook (site.yml/site-container.yml)
The thing is that the files get generated in the 'fetch_directory' with
root user because there is a 'delegate_to' + we run the playbook with
`become: True` (from main playbook).
The idea here is to create files under ansible user so we can open them
later to copy them on the remote machine.
Leah Neukirchen [Thu, 7 Feb 2019 17:09:21 +0000 (18:09 +0100)]
Fix uses of default(omit) with string concatenation
When {{omit}} is concatenated with another string, it expands to something
like __omit_place_holder__63eea0d96dd6ed867b95405e11d87dddf61f448d.
However, in these use-cases we need an empty string.
Sébastien Han [Mon, 26 Nov 2018 16:58:49 +0000 (17:58 +0100)]
osd: expose udev into the container
In order to be able to retrieve udev information, we must expose its
socket. As per, https://github.com/ceph/ceph/pull/25201 ceph-volume will
start consuming udev output.
Noah Watkins [Tue, 6 Nov 2018 16:49:39 +0000 (08:49 -0800)]
Fixup shrink_osd[_container] scenario config
** configuration seems to be for filestore:
[ERROR]: [ceph-osd0] Validation failed for variable: lvm_volumes
** Removing `radosgw_interface: eth1` to resolve:
The task includes an option with an undefined variable. The error was:
'ansible.vars.hostvars.HostVarsVars object' has no attribute
u'ansible_eth1'
The error appears to have been in
'/home/nwatkins/src/ceph-ansible/roles/ceph-defaults/tasks/set_radosgw_address.yml':
line 21, column 5, but may be elsewhere in the file depending on the
exact syntax problem.
The offending line appears to be:
- name: set_fact _radosgw_address to radosgw_interface - ipv4
^ here
projects / ceph-ansible.git / log