Erwan Velu [Fri, 1 Jun 2018 16:53:10 +0000 (18:53 +0200)]
ceph-defaults: Enable local epel repository
During the tests, the remote epel repository is generating a lots of
errors leading to broken jobs (issue #2666)
This patch is about using a local repository instead of a random one.
To achieve that, we make a preliminary install of epel-release, remove
the metalink and enforce a baseurl to our local http mirror.
That should speed up the build process but also avoid the random errors
we face.
This patch is part of a patch series that tries to remove all possible yum failures.
Since the openstack_config.yml has been moved to `ceph-osd` we must move
this `set_fact` in ceph-osd otherwise the tasks in
`openstack_config.yml` using `openstack_keys` will actually use the
defaults value from `ceph-defaults`.
osds: wait for osds to be up before creating pools
This is a follow up on #2628.
Even with the openstack pools creation moved later in the playbook,
there is still an issue because OSDs are not all UP when trying to
create pools.
Adding a task which checks for all OSDs to be UP with a `retries/until`
condition should definitively fix this issue.
Fix a typo in `tag` target, double quote are missing here.
Without them, the `make tag` command fails like this:
```
if [[ "v3.0.35" == ]]; then \
echo "e5f2df8 on stable-3.0 is already tagged as v3.0.35"; \
exit 1; \
fi
/bin/sh: -c: line 0: unexpected argument `]]' to conditional binary operator
/bin/sh: -c: line 0: syntax error near `;'
/bin/sh: -c: line 0: `if [[ "v3.0.35" == ]]; then echo "e5f2df8 on stable-3.0 is already tagged as v3.0.35"; exit 1; fi'
make: *** [tag] Error 2
```
When playing ceph-mds role, mon nodes have set a fact with the default
pg num for osd pools, we can simply default to this value for cephfs
pools (`cephfs_pools` variable).
At the moment the variable definition for `cephfs_pools` looks like:
and we have a task in `ceph-validate` to ensure `pgs` has been set to a
valid value.
We could simply avoid this check by setting the default value of `pgs`
to `hostvars[groups[mon_group_name][0]]['osd_pool_default_pg_num']` and
let to users the possibility to override this value.
in `ceph-osd` there is no need to set `docker_exec_cmd` since the only
place where this fact is used is in `openstack_config.yml` which
delegate all docker command to a monitor node. It means we need the
`docker_exec_cmd` fact that has been set referring to `ceph-mon-*`
containers, this fact is already set earlier in `ceph-defaults`.
By the way, when collocating an OSD with a MON it fails because the container
`ceph-osd-{{ ansible_hostname }}` doesn't exist.
Removing this task will allow to collocate an OSD with a MON.
For a few moment we can see failures in the CI for containerized
scenarios because VMs are running out of space at some point.
The default in the images used is to have only 3Gb for root partition
which doesn't sound like a lot.
Typical error seen:
```
STDERR:
failed to register layer: Error processing tar file(exit status 1): open /usr/share/zoneinfo/Atlantic/Canary: no space left on device
```
Indeed, on the machine we can see:
```
Every 2.0s: df -h Tue May 29 17:21:13 2018
Filesystem Size Used Avail Use% Mounted on
/dev/mapper/atomicos-root 3.0G 3.0G 14M 100% /
```
The idea here is to expand this partition with all the available space
remaining by issuing an `lvresize` followed by an `xfs_growfs`.
```
-bash-4.2# lvresize -l +100%FREE /dev/atomicos/root
Size of logical volume atomicos/root changed from <2.93 GiB (750 extents) to 9.70 GiB (2484 extents).
Logical volume atomicos/root successfully resized.
```
In the CI we can see at many times failures like following:
`Failure talking to yum: Cannot find a valid baseurl for repo:
base/7/x86_64`
It seems the fastest mirror detection is sometimes counterproductive and
leads yum to fail.
This fix has been added in the `setup.yml`.
This playbook was used until now only just before playing `testinfra`
and could be used before running ceph-ansible so we can add some
provisionning tasks.
When collocating mds on monitor node, the cephpfs will fail
because `docker_exec_cmd` is reset to `ceph-mds-monXX` which is
incorrect because we need to delegate the task on `ceph-mon-monXX`.
In addition, it wouldn't have worked since `ceph-mds-monXX` container
isn't started yet.
Moving the task earlier in the `ceph-mds` role will fix this issue.
Sébastien Han [Mon, 16 Apr 2018 13:57:23 +0000 (15:57 +0200)]
rgw: container add option to configure multi-site zone
You can now use RGW_ZONE and RGW_ZONEGROUP on each rgw host from your
inventory and assign them a value. Once the rgw container starts it'll
pick the info and add itself to the right zone.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1551637 Signed-off-by: Sébastien Han <seb@redhat.com>
Since we fixed the `gather and delegate facts` task, this exception is
not needed anymore. It's a leftover that should be removed to save some
time when deploying a cluster with a large client number.
When deploying a large number of OSD nodes it can be an issue because the
protection check [1] won't pass since it tries to create pools before all
OSDs are active.
The idea here is to move cephfs pools creation in `ceph-mds` role.
let's move this variable in group_vars/all.yml in all testing scenarios
accordingly to this commit 1f15a81c480f60bc82bfc3a1aec3fe136e6d3bc4 so
we keep consistency between the playbook and the tests.
When deploying a large number of OSD nodes it can be an issue because the
protection check [1] won't pass since it tries to create pools before all
OSDs are active.
The idea here is to move openstack pools creation at the end of `ceph-osd` role.
Luigi Toscano [Tue, 22 May 2018 09:46:33 +0000 (11:46 +0200)]
ceph-radosgw: disable NSS PKI db when SSL is disabled
The NSS PKI database is needed only if radosgw_keystone_ssl
is explicitly set to true, otherwise the SSL integration is
not enabled.
It is worth noting that the PKI support was removed from Keystone
starting from the Ocata release, so some code paths should be
changed anyway.
Also, remove radosgw_keystone, which is not useful anymore.
This variable was used until fcba2c801a122b7ce8ec6a5c27a70bc19589d177.
Now profiles drives the setting of rgw keystone *.
Signed-off-by: Luigi Toscano <ltoscano@redhat.com>
Vishal Kanaujia [Wed, 16 May 2018 09:58:31 +0000 (15:28 +0530)]
Skip GPT header creation for lvm osd scenario
The LVM lvcreate fails if the disk already has a GPT header.
We create GPT header regardless of OSD scenario. The fix is to
skip header creation for lvm scenario.
Sébastien Han [Tue, 22 May 2018 23:52:40 +0000 (16:52 -0700)]
rolling_update: fix get fsid for containers
When running ansible2.4-update_docker_cluster there is an issue on the
"get current fsid" task. The current task only works for
non-containerized deployment but will run all the time (even for
containerized). This currently results in the following error:
This is not really representative on the real error since the 'ceph' cli is available on that machine.
On other environments we will have something like "command not found: ceph".
Fix restarting OSDs twice during a rolling update.
During a rolling update, OSDs are restarted twice currently. Once, by the
handler in roles/ceph-defaults/handlers/main.yml and a second time by tasks
in the rolling_update playbook. This change turns off restarts by the handler.
Further, the restart initiated by the rolling_update playbook is more
efficient as it restarts all the OSDs on a host as one operation and waits
for them to rejoin the cluster. The restart task in the handler restarts one
OSD at a time and waits for it to join the cluster.
Alfredo Deza [Mon, 21 May 2018 12:09:00 +0000 (08:09 -0400)]
validate: split schema for lvm osd scenario per objecstore
The bluestore lvm osd scenario does not require a journal entry. For
this reason we need to have a separate schema for that and filestore or
notario will fail validation for the bluestore lvm scenario because the
journal key does not exist in lvm_volumes.
Andrew Schoen [Mon, 21 May 2018 15:11:22 +0000 (10:11 -0500)]
ceph-validate: do not check ceph version on dev or rhcs installs
A dev or rhcs install does not require ceph_stable_release to be set and
instead generates that by looking at the installed ceph-version.
However, at this point in the playbook ceph may not have been installed
yet and ceph-common has not be run.
Fixes: https://github.com/ceph/ceph-ansible/issues/2618 Signed-off-by: Andrew Schoen <aschoen@redhat.com>
Andrew Schoen [Tue, 8 May 2018 14:26:07 +0000 (09:26 -0500)]
ceph-defaults: remove backwards compat for containerized_deployment
The validation module does not get config options with the template
syntax rendered, so we're gonna remove that and just default it to
False. The backwards compat was schedule to be removed in 3.1 anyway.
Andrew Schoen [Wed, 2 May 2018 17:55:33 +0000 (12:55 -0500)]
adds a requiremnts.txt file for the project
With the addition of the validate module we need to ensure
that notario is installed. This will be done with the use
of this requirments.txt file and pip.
Andrew Schoen [Tue, 1 May 2018 16:22:31 +0000 (11:22 -0500)]
ceph-defaults: fix failing tasks when osd_scenario was not set correctly
When devices is not defined because you want to use the 'lvm'
osd_scenario but you've made a mistake selecting that scenario these
tasks should not fail.
Sébastien Han [Fri, 18 May 2018 12:43:57 +0000 (14:43 +0200)]
defaults: restart_osd_daemon unit spaces
Extra space in systemctl list-units can cause restart_osd_daemon.sh to
fail
It looks like if you have more services enabled in the node space
between "loaded" and "active" get more space as compared to one space
given in command the command[1].
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1573317 Signed-off-by: Sébastien Han <seb@redhat.com>
take-over: fix bug when trying to override variable
A customer has been facing an issue when trying to override
`monitor_interface` in inventory host file.
In his use case, all nodes had the same interface for
`monitor_interface` name except one. Therefore, they tried to override
this variable for that node in the inventory host file but the
take-over-existing-cluster playbook was failing when trying to generate
the new ceph.conf file because of undefined variable.
Typical error:
```
fatal: [srvcto103cnodep01]: FAILED! => {"failed": true, "msg": "'dict object' has no attribute u'ansible_bond0.15'"}
```
Including variables like this `include_vars: group_vars/all.yml` prevent
us from overriding anything in inventory host file because it
overwrites everything you would have defined in inventory.
Andy McCrae [Mon, 19 Feb 2018 16:57:18 +0000 (16:57 +0000)]
Fix template reference for ganesha.conf
We can simply reference the template name since it exists within the
role that we are calling. We don't need to check the ANSIBLE_ROLE_PATH
or playbooks directory for the file.
Sébastien Han [Wed, 16 May 2018 15:37:10 +0000 (17:37 +0200)]
switch: disable ceph-disk units
During the transition from jewel non-container to container old ceph
units are disabled. ceph-disk can still remain in some cases and will
appear as 'loaded failed', this is not a problem although operators
might not like to see these units failing. That's why we remove them if
we find them.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1577846 Signed-off-by: Sébastien Han <seb@redhat.com>
there is some leftover on devices when purging osds because of a invalid
device list construction.
typical error:
```
changed: [osd3] => (item=/dev/sda sda1) => {
"changed": true,
"cmd": "# if the disk passed is a raw device AND the boot system disk\n if parted -s \"/dev/sda sda1\" print | grep -sq boot; then\n echo \"Looks like /dev/sda sda1 has a boot partition,\"\n echo \"if you want to delete specific partitions point to the partition instead of the raw device\"\n echo \"Do not use your system disk!\"\n exit 1\n fi\n echo sgdisk -Z \"/dev/sda sda1\"\n echo dd if=/dev/zero of=\"/dev/sda sda1\" bs=1M count=200\n echo udevadm settle --timeout=600",
"delta": "0:00:00.015188",
"end": "2018-05-16 12:41:40.408597",
"item": "/dev/sda sda1",
"rc": 0,
"start": "2018-05-16 12:41:40.393409"
}
Error: Could not stat device /dev/sda sda1 - No such file or directory.
```
the devices list in the task `resolve parent device` isn't built
properly because the command used to resolve the parent device doesn't
return the expected output
For instance, it will result with a devices list like:
`['/dev/sda sda1', '/dev/sdb', '/dev/sdc sdc1']`
where we expect to have:
`['/dev/sda', '/dev/sdb', '/dev/sdc']`
Sébastien Han [Wed, 16 May 2018 14:02:41 +0000 (16:02 +0200)]
rolling_update: move osd flag section
During a minor update from a jewel to a higher jewel version (10.2.9 to
10.2.10 for example) osd flags don't get applied because they were done
in the mgr section which is skipped in jewel since this daemons does not
exist.
Moving the set flag section after all the mons have been updated solves
that problem.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1548071 Co-authored-by: Tomas Petr <tpetr@redhat.com> Signed-off-by: Sébastien Han <seb@redhat.com>
Ken Dreyer [Thu, 10 May 2018 23:08:05 +0000 (17:08 -0600)]
Makefile: add "make tag" command
Add a new "make tag" command. This automates some common operations:
1) Automatically determine the next Git tag version number to create.
For example:
"3.2.0beta1 -> "3.2.0beta2"
"3.2.0rc1 -> "3.2.0rc2"
"3.2.0" -> "3.2.1"
2) Create the Git tag, and print instructions for the user to push it to
GitHub.
3) Sanity check that HEAD is a stable-* branch or master (bail on
everything else).
4) Sanity check that HEAD is not already tagged.
Note, we will still need to tag manually once each time we change the
format, for example when moving from tagging "betas" to tagging "rcs",
or "rcs" to "stable point releases".
Signed-off-by: Ken Dreyer <kdreyer@redhat.com> Co-authored-by: Guillaume Abrioux <gabrioux@redhat.com>
Andy McCrae [Tue, 1 May 2018 11:20:19 +0000 (12:20 +0100)]
Install packages as a list
To make the package installation more efficient we should install
packages as a list rather than as individual tasks or using a
"with_items" loop. The package managers can handle a list passed to them
to install in one go.
We can use a specified list and substitute any packages that are not to
be installed with the ceph-common package, which is installed on every
package install, then apply the unique filter to the package install
list.
There is no need to stat for created mgr keyrings since they are created
anyway when deploying a ceph cluster > jewel. In case of a jewel
deployment we won't enter that block.
rolling_update: fix dest path for mgr keys fetching
the role `ceph-mgr` that is played later in the playbook fails because
the destination path for the fetched keys is wrong.
This patch fix the destination path used in the task `fetch ceph mgr
key(s)` so there is no mismatch.
{{ fsid }} points to {{ cluster_uuid.stdout }} which is not defined in
this part of the rolling_update playbook.
Since we need to call {{ fsid }} we must get the fsid and register it to
`cluster_uuid`.
Sébastien Han [Thu, 10 May 2018 17:38:55 +0000 (10:38 -0700)]
rolling_update: move mgr key creation
Until all the mons haven't been updated to Luminous, there is no way to
create a key. So we should do the key creation in the mon role only if
we are not part of an update.
If we are then the key creation is done after the mons upgrade to
Luminous.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1574995 Signed-off-by: Sébastien Han <seb@redhat.com>
trying to mask target when `/etc/systemd/system/target.service` doesn't
exist seems to be a bug.
There is no need to mask a unit file which doesn't exist.
Andy McCrae [Thu, 10 May 2018 10:15:30 +0000 (11:15 +0100)]
Allow os_tuning_params to overwrite fs.aio-max-nr
The order of fs.aio-max-nr (which is hard-coded to 1048576) means that
if you set fs.aio-max-nr in os_tuning_params it will effectively be
ignored for bluestore scenarios.
To resolve this we should move the setting of fs.aio-max-nr above the
setting of os_tuning_params, in this way the operator can define the
value of fs.aio-max-nr to be something other than 1048576 if they want
to.
Additionally, we can make the sysctl settings happen in 1 task rather
than multiple.
Ken Dreyer [Thu, 10 May 2018 20:39:07 +0000 (14:39 -0600)]
Makefile: bail out on unknown Git tag formats
Prior to this change, if we created entirely new Git tags patterns like
"3.2.0alpha" or "3.2.0foobar", the Makefile would incorrectly translate
the Git tag name into a Name-Version-Release that would prevent upgrades
to "newer" versions.
This happened for example in
https://bugs.centos.org/view.php?id=14593, "Incorrect naming scheme for
a build of ceph-ansible prevents subsequent updates to be installed"
If we encounter a new Git tag format that we cannot parse,
pessimistically bail out early instead of trying to build an RPM.
The purpose of this safeguard is to prevent Jenkins from building RPMs
that cannot be easily upgraded.
client: remove default value for pg_num in pools creation
trying to set the default value for pg_num to
`hostvars[groups[mon_group_name][0]]['osd_pool_default_pg_num'])` will
break in case of external client nodes deployment.
the `pg_num` attribute should be mandatory and be tested in future
`ceph-validate` role.
Sébastien Han [Wed, 9 May 2018 21:30:12 +0000 (14:30 -0700)]
contrib: update backport script to reflect stable branch
Since we now do backports on stable-3.0 and stable-3.1 we have to use
the name of the stable branch in the backport branch name. If we don't
do this we will end up with conflicting branch names.
mon: fix mgr keyring creation when upgrading from jewel
On containerized deployment,
when upgrading from jewel to luminous, mgr keyring creation fails because the
command to create mgr keyring is executed on a container that is still
running jewel since the container is restarted later to run the new
image, therefore, it fails with bad entity error.
To get around this situation, we can delegate the command to create
these keyrings on the first monitor when we are running the playbook on the last monitor.
That way we ensure we will issue the command on a container that has
been well restarted with the new image.
projects / ceph-ansible.git / log