qa/workunits/cephadm/test_cephadm: mark services unmanaged for test

We are deploying containers manually.  Mark them unmanaged so that we
do not fight against mgr/cephadm cleaning up orphan daemons.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: do not reconfig unmanaged services

Fixes: https://tracker.ceph.com/issues/44615
Signed-off-by: Sage Weil <sage@redhat.com>

qa/workunits/cephadm/test_cephadm: output file for pub key

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33866 into octopus

* refs/pull/33866/head:
mgr/cephadm: limit number of times check host is performed in the serve loop

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #34005 into octopus

* refs/pull/34005/head:
python-common: reduce output of ServiceSpec.to_json()
python-common,mgr/cephadm: move assert_valid_host to service_spec
mgr/cephadm: add HostAssignment.validate()

Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #34013 into octopus

* refs/pull/34013/head:
qa/workunits/cephadm/test_cephadm: --skip-monitoring-stack

Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge pull request #33994 from dillaman/wip-librbd-poll-event-race

librbd: defer event socket completion until after callback issued

Reviewed-by: Dehao Shang <dehao.shang@intel.com>
Reviewed-by: Mykola Golub <mgolub@suse.com>

qa/workunits/cephadm/test_cephadm: --skip-monitoring-stack

We're deploying these things manually later, and they use fixed ports.

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33974 into octopus

* refs/pull/33974/head:
use quay octopus tip until 15.2 tag is available
update default container images

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #33442 into octopus

* refs/pull/33442/head:
cephadm: add allow_ptrace option to enable SYS_PTRACE

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #33997 into octopus

* refs/pull/33997/head:
mon/MgrMonitor: make 'mgr fail' work with no arguments

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge PR #34000 into octopus

* refs/pull/34000/head:
mgr/dashboard: adapt create_osds interface change

Reviewed-by: Laura Paduano <lpaduano@suse.com>

use quay octopus tip until 15.2 tag is available

Signed-off-by: Sage Weil <sage@redhat.com>

python-common: reduce output of ServiceSpec.to_json()

Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

python-common,mgr/cephadm: move assert_valid_host to service_spec

Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

mgr/cephadm: add HostAssignment.validate()

Fixes: https://tracker.ceph.com/issues/44607
Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

mgr/dashboard: adapt create_osds interface change

Support the parameter change of create_osds since
https://github.com/ceph/ceph/pull/33922.

Fixes: https://tracker.ceph.com/issues/44632
Signed-off-by: Kiefer Chang <kiefer.chang@suse.com>

mon/MgrMonitor: make 'mgr fail' work with no arguments

Fail the active mgr so you don't have to check ceph -s and cut&paste it or
do some jq shenanigans with 'mgr dump'.

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: add allow_ptrace option to enable SYS_PTRACE

In any environments it is helpful to have SYS_PTRACE so that you can
gdb attach or strace a daemon.

Leave this off by default so that the container is more secure.

Enable this in teuthology and vstart.

Signed-off-by: Sage Weil <sage@redhat.com>

update default container images

- For tests, use bleeding-edge octopus branch
- For production defaults, use ceph/ceph:v15.2 tag
- For bootstrap, grab cephadm script from latest octopus branch

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: limit number of times check host is performed in the serve loop

we only need this to happen every 10 min on hosts

Signed-off-by: Daniel-Pivonka <dpivonka@redhat.com>

Merge PR #33961 into octopus

* refs/pull/33961/head:
cephadm: add-repo: add --version
cephadm: rename distro args in repo methods

Reviewed-by: Sebastian Wagner <swagner@suse.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge PR #33952 into octopus

* refs/pull/33952/head:
qa/workunits/cephadm: --skip-mon-network when using 127.0.0.1
cephadm: add tests
qa/tasks/cephadm: pass -v to bootstrap
mgr/cephadm: only try to place mons on hosts matching public_network
mgr/cephadm: keep track of host networks, ips
cephadm: automatically infer mon public_network, if we can
cephadm: add list-networks command

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #33990 into octopus

* refs/pull/33990/head:
mgr/orch: add --all-available-devices to 'orch apply osd'

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #33955 into octopus

* refs/pull/33955/head:
mgr/cephadm: respect 'unmanaged' flag in spec
mgr/orch: orch ls: show <no spec> or <unmanaged> as appropriate
mgr/orch: orch ls: rename SPEC -> PLACEMENT
mgr/orch: add 'unmanaged' property to ServiceSpec
mgr/orch: combine 'orch daemon add <type> ...' into one command
mgr/orch: combine 'orch apply <type> [<placement>]' into one command

Reviewed-by: Sebastian Wagner <swagner@suse.com>

Merge PR #33936 into octopus

* refs/pull/33936/head:
cephadm: bootstrap: deploy monitoring stack by default

Reviewed-by: Sebastian Wagner <swagner@suse.com>
Reviewed-by: Patrick Seidensal <pseidensal@suse.com>

mgr/orch: add --all-available-devices to 'orch apply osd'

Provide a super-simple "use any avaialable device" command for
'orch apply osd'.  This will work for many (maybe even most?) users.

Signed-off-by: Sage Weil <sage@redhat.com>

qa/workunits/cephadm: --skip-mon-network when using 127.0.0.1

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: add tests

Signed-off-by: Sage Weil <sage@redhat.com>

qa/tasks/cephadm: pass -v to bootstrap

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: only try to place mons on hosts matching public_network

Only try to schedule new mons on hosts that match the configured
public_network, if any.  If we do not have one configured, then don't
try to place new mons at all.

Note that there are other restrictions that ceph-mon supports that we
aren't considering, here: public_network_interface in particular, which
might further limit which IPs we consider binding to.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: keep track of host networks, ips

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: automatically infer mon public_network, if we can

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: add list-networks command

Summarize CIDR networks and our IPs in them.  Result looks like this:

{
    "10.3.64.0/24": [
        "10.3.64.23",
        "10.3.64.27"
    ],
    "10.88.0.0/16": [
        "10.88.0.1"
    ],
    "192.168.122.0/24": [
        "192.168.122.1"
    ]
}

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: bootstrap: deploy monitoring stack by default

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33979 into octopus

* refs/pull/33979/head:
cephadm: implement 'install' command

Reviewed-by: Michael Fritch <mfritch@suse.com>

librbd: defer event socket completion until after callback issued

A change post-Nautilus fixed an issue where multiple threads could
concurrently invoke callbacks to librbd clients. However, it also
introduced the potential that a callback hasn't yet fired by the
time the event socket is completed. This resulted in a crash of
fio under high-throughput testing since it expected both a callback
and the event socket completion, in that order.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #33988 from dillaman/wip-44391

rbd-mirror: hold lock while updating local image name

Reviewed-by: Mykola Golub <mgolub@suse.com>

cephadm: add-repo: add --version

Instead of --release octopus, which would get the latest octopus
version (whatever it might be), or possibly a repo with many build versions
inside, you can instead do --version 15.2.1 to get a repo with a
specific version and that version only.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: respect 'unmanaged' flag in spec

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/orch: orch ls: show <no spec> or <unmanaged> as appropriate

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/orch: orch ls: rename SPEC -> PLACEMENT

More accurate, I think.  We only ever show placement here.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/orch: add 'unmanaged' property to ServiceSpec

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: implement 'install' command

This was present, but a no-op.

By default, install cephadm.

Optionally take a list of packages to install instead (e.g., ceph-common).

Signed-off-by: Sage Weil <sage@redhat.com>

rbd-mirror: hold lock while updating local image name

There is a potential for an independently scheduled status update to
request the local image name from the bootstrap state machine during its
initialization.

Fixes: https://tracker.ceph.com/issues/44391
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #33987 from tchaikov/wip-octopus-formatting

doc/releases/octopus.rst: formatting tweaks

Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #33980 into octopus

* refs/pull/33980/head:
cephadm: bootstrap: allow --output-dir
cephadm: do not infer image for bootstrap
cephadm: write output files to /etc/ceph by default
cephadm: verify the output files' containing directory exists

Reviewed-by: Michael Fritch <mfritch@suse.com>

doc/releases/octopus.rst: use note directive for notes

so the note stands out in the text

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/releases/octopus.rst: add '$' as prompt

to help the BashSession lexer to render the text starting with "#" as comments

see https://pygments.org/docs/lexers/#lexers-for-various-shells

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/releases/octopus: decrease the indent level

for less and better horizontal spacing

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/_static/css: fine tune the spacing

* `ul.simple > li` is also used in front page, where the items are too
  sparse now. so we need to restore the default spacing.
* `ul.simple > li > ul > li:last-child` is used to control the spacing
  between nested unordered list. we need have more spacing there.
* `div.section > ul > li > p` is added to decrease the spacing of nested
  list items.

Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #33971 from tchaikov/octopus-doc-release

doc/releases/octopus.rst: format tweaks

Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #33968 into octopus

* refs/pull/33968/head:
mgr/cephadm: do not include 'osd' service in 'orch ls' output

Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge PR #33970 into octopus

* refs/pull/33970/head:
python-common: 'all:true' -> '*'

Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge PR #33973 into octopus

* refs/pull/33973/head:
mgr/orch,cephadm: elevate --force flag for daemons to interface
cephadm: backup data when we remove stateful daemons (mon, osd, prometheus)
cephadm: do not zap disk after removing an OSD daemon

Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge PR #33935 into octopus

* refs/pull/33935/head:
mgr/cephadm: fix host connection exception
doc/cephadm: use ssh-copy-id

Reviewed-by: Joshua Schmid <jschmid@suse.de>

doc/releases/octopus.rst: reindent top-level items

reduce the indent level, so these items are better formated

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/releases/octopus.rst: formatting tweaks

* use console lexer for highlighting command lines
  otherwise they will be rendered as python code
* use bold font instead italic in subsection headers,
  they look more formal this way
* fix a typo: delineated

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/releases: fix hyper link

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc/release: fix "CephFS" subsection heading

it's a regression introduced by 26db06cf212be029e5628a40de3410632c0b86a0

Signed-off-by: Kefu Chai <kchai@redhat.com>

cephadm: bootstrap: allow --output-dir

This is easier than specifying each of output-config, output-keyring, and
output-pub-ssh-key.

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: do not infer image for bootstrap

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: write output files to /etc/ceph by default

This streamlines the install for most users by putting the config and
keyring files in /etc/ceph, where other ceph CLI commands will find them.

The cost is making isolated installs a bit more painful, by requiring
that users specify a local location for --output-config etc.

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: verify the output files' containing directory exists

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33977 into octopus

* refs/pull/33977/head:
pybind/mgr: apply_drivegroups should return Sequence[Completion]

Reviewed-by: Sage Weil <sage@redhat.com>

pybind/mgr: apply_drivegroups should return Sequence[Completion]

* change annotation of `orchestrator.Orchestrator.apply_drivegroups()` so it returns
  `Sequence[Completion]`
  see
  https://mypy.readthedocs.io/en/latest/generics.html#variance-of-generic-types
* to follow the interface defined by `orchestrator.Orchestrator`. change
  all its implementations. see `orchestrator/_interface.py`.

Signed-off-by: Kefu Chai <kchai@redhat.com>

mgr/orch,cephadm: elevate --force flag for daemons to interface

Apply this check in the generic orchestrator code so that it applies
consistently to all implementations.

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: backup data when we remove stateful daemons (mon, osd, prometheus)

This avoids doing permanent damage when removing stateful daemons.

We include OSDs here even though in most cases no precious data lives in
the osd data directory because there might be some important metadata
there--particularly for adopted OSDs from older clusters.

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: do not zap disk after removing an OSD daemon

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33922 into octopus

* refs/pull/33922/head:
qa/tasks/mgr/test_orchestrator_cli: adapt
qa/suites/rados/cephadm/upgrade: start from compatible build
mgr/cephadm: refine log message
python-common: correctly dictify DriveGroupSpec objects when paths are used
mgr/cephadm: adapt test_cephadm.sh to new syntax
mgr/cephadm: adapt qa->test_cephadm.sh
qa/tasks/cephadm: use 'orch daemon osd add'
mgr/cephadm: adapt osd deployment to service_apply

Reviewed-by: Sage Weil <sage@redhat.com>

qa/tasks/mgr/test_orchestrator_cli: adapt

Signed-off-by: Sage Weil <sage@redhat.com>

qa/suites/rados/cephadm/upgrade: start from compatible build

The osd create command is changing, so upgrade from something that the
latest cephadm.py task can use.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: fix host connection exception

1- catch exception thrown by _get_connection itself
2- make the error message more informative.  the old one was

Error ENOENT: New host dael (10.3.64.25) failed to connect: `ssh -F /tmp/cephadm-conf-9tvrgrls -i /tmp/cephadm-identity-sj66ado1 root@10.3.64.25`

which was both misleading and had irrelevant information.  Now,

Error ENOENT: Failed to connect to dael (10.3.64.25).  Check that the host is reachable and accepts connections using the cephadm SSH key

Signed-off-by: Sage Weil <sage@redhat.com>

doc/cephadm: use ssh-copy-id

Signed-off-by: Sage Weil <sage@redhat.com>

python-common: 'all:true' -> '*'

'all:true' is no longer a valid PlacementSpec.

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/cephadm: do not include 'osd' service in 'orch ls' output

Signed-off-by: Sage Weil <sage@redhat.com>

cephadm: rename distro args in repo methods

Consistently prefix distro vars with distro_

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/orch: combine 'orch daemon add <type> ...' into one command

Signed-off-by: Sage Weil <sage@redhat.com>

mgr/orch: combine 'orch apply <type> [<placement>]' into one command

Declutter the CLI!

Signed-off-by: Sage Weil <sage@redhat.com>

Merge remote-tracking branch 'gh/octopus' into octopus

Merge PR #33520 into octopus

* refs/pull/33520/head:
mgr/dashboard: Isolate each RBD component

Reviewed-by: Stephan Müller <smueller@suse.com>

Merge PR #33758 into octopus

* refs/pull/33758/head:
mgr/dashboard: Replace ng2-tree with angular-tree-component

Reviewed-by: Stephan Müller <smueller@suse.com>
Reviewed-by: Alfonso Martínez <almartin@redhat.com>
Reviewed-by: Laura Paduano <lpaduano@suse.com>

Merge PR #33941 into octopus

* refs/pull/33941/head:
mgr/dashboard: hide daemon table when orchestrator is disabled

Reviewed-by: Volker Theile <vtheile@suse.com>

Merge PR #33948 into octopus

* refs/pull/33948/head:
python-common: PlacementSpec.from_string: drop `host_pattern:` prefix
python-common: remove `all_hosts` from `PlacementSpec`

Reviewed-by: Sage Weil <sage@redhat.com>

mgr/cephadm: refine log message

Avoid a lot of these

2020-03-13T14:58:38.785360-0500 mgr.x [INF] No data_devices, skipping DriveGroup: osd.all

every time we consider the OSD spec and find nothing to do.  Instead, log
something like this each time we actually have work to do:

2020-03-13T15:04:22.803356-0500 mgr.x [INF] Applying osd.all on host gnit...
2020-03-13T15:04:33.723731-0500 mgr.x [INF] Deploying daemon osd.5 on gnit

Signed-off-by: Sage Weil <sage@redhat.com>

Merge PR #33777 into octopus

* refs/pull/33777/head:
rgw: svc.config_key_rados: get() warns if potentially insecure connection
rgw: svc.rados: add clog_warn()
mon: log monitor: add command level param
auth registry: helpers for checking secure method/mode

Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #33950 into octopus

* refs/pull/33950/head:
ceph.in: use os._exit when we don't shut down

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge PR #33951 into octopus

* refs/pull/33951/head:
mgr/orch: dump service spec by name

Reviewed-by: Joshua Schmid <jschmid@suse.de>
Reviewed-by: Sebastian Wagner <swagner@suse.com>

ceph.in: use os._exit when we don't shut down

If we experience a timeout, we don't bother shutting down rados, because
it can't currently handle one thread blocked (or running) library init
at the same time that another thread calls rados_shutdown().  This
behavior was recently introduced by c8f353c50bcd2a30be9e3600dba91912b8cd0429

However, sys.exit() runs all kinds of shutdown work that will also
interfere with the running librados threads.

Fix by using os._exit instead of sys.exit.

Fixes: https://tracker.ceph.com/issues/44566
Signed-off-by: Sage Weil <sage@redhat.com>

15.1.1

mgr/orch: dump service spec by name

add optional arg to allow a dump by name:
`ceph orch spec dump [svc_name]`

Signed-off-by: Michael Fritch <mfritch@suse.com>

Merge PR #33953 into octopus

* refs/pull/33953/head:
doc: Move ceph-deploy docs to doc/install/ceph-deploy

Reviewed-by: Sage Weil <sage@redhat.com>

doc: Move ceph-deploy docs to doc/install/ceph-deploy

Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

Merge PR #33924 into octopus

* refs/pull/33924/head:
doc/mgr: Enhance placement specs
doc/cephadm: Fix toc of Drive Groups

Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #33891 into octopus

* refs/pull/33891/head:
mgr/cephadm: fix dump output by formatting to json first

Reviewed-by: Sebastian Wagner <swagner@suse.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge PR #33905 into octopus

* refs/pull/33905/head:
install-deps.sh: remove the workaround to bootstrap a new venv
install-deps.sh: install additional repo for RHEL8
install-deps.sh: install rhel-7-server-optional-rpms on el7 only

Reviewed-by: David Galloway <dgallowa@redhat.com>

python-common: correctly dictify DriveGroupSpec objects when paths are used

Signed-off-by: Joshua Schmid <jschmid@suse.de>

Merge pull request #33940 from tchaikov/octopus-release-notes-header

doc: use subsection for representing components in release notes

Reviewed-by: Sage Weil <sage@redhat.com>

doc/mgr: Enhance placement specs

Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

doc/releases: use h3 for rendering subsections in release notes

to highlight the components with subsection header, and allow readers to
navigate through them with TOC.

Signed-off-by: Kefu Chai <kchai@redhat.com>