On POWER systems, /proc/cpuinfo provides a 'processor' entry as a
counter, and a 'cpu' entry with a description (similar to 'model name'
on x86). Support for POWER in get_cpu_facts was added via the 'cpu'
entry in commit 8746e692c1. Subsequent
support for ARM64 in commit ce4ada93f9
used the 'processor' entry, resulting in double-counting of cores on
POWER systems.
When unit tests were later written for this code in
commit 55306906cf, the erroneous values
were just accepted in the test instead of being diagnosed.
Signed-off-by: Yaakov Selkowitz <yselkowi@redhat.com>
* Clean up comments in integration tests.
Tests reference soon to be outdated paths and implementation details.
* Remove unused test/runner/ reference in test.
* remove external grep call and parse with python
* use function for repeated code
* use module.get_bin_path() for iscsiutil on HPUX
* some code opt for HPUX
* clean up non-module code, module being defined is a requirement for this code
* import get_bin_path() directly and use without module prefix
* Add integration tests for AIX and HP-UX
* add changelog fragment
* Apply suggestions from code review
Co-Authored-By: mator <matorola@gmail.com>
* Apply suggestions from code review #2
Co-Authored-By: Sam Doran <sdoran@redhat.com>
* Remove strict requirement on executable to exist for get_bin_path() as
it will allow facts gathering to continue without an error. Almost all
other files under facts do not have "required=True" (except 2 files,
which should be probably fixed). And check return value for
get_bin_path() , before run attempt.
* add check for AIX lsattr run_command return code
Fixes a bug where parse_distribution_file_ClearLinux() was called on CoreOS (and probably many other distros) and it returned True since it successfully parses the distribution file. Since this file exists on many Linux distributions and they are a very similar format, add an additional check to make sure it is Clear Linux.
Change the order in which distribution files are processed so NA is last. This prevents a match on CoreOS hosts since they also have /etc/os-release and the called matching function for NA is very general and will match CoreOS.
* Add changelog
* Add unit tests
Only add tests for Clear Linux parsing since that was the cause of this issue.
* Facts parsing for cmdline can now handle multiple values for a single key.
* Unit tests for cmdline fact parsing
* Review comments
Fixes: #22766
Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
* parallelize getting mount info
* fixed timeout and made 8 max thread count
- minor cleanup
- avoid empty mount entries
- set timeout on get
- enforce timeout per mount/thread
- make note on failure per mount
- make note on timeout per mount
- ensure proper pool control
- minor fixes
- less vars, simpler code
- move filter 'pre threading'
- remove timeout for all mounts, now per mount
- also use cpu count from multiprocessing lib
- moved 'bind' options out of thread as per comments
- warn on error, more info on failure to get info
* Revert "allow caller to deal with timeout (#49449)"
This reverts commit 63279823a7.
Flawed on many levels
* Adds poor API to a public function
* Papers over the fact that the public function is doing something bad
by catching exceptions it cannot handle in the first place
* Papers over the real cause of the issue which is a bug in the timeout
decorator
* Doesn't reraise properly
* Catches the wrong exception
Fixes#49824Fixes#49817
* Make the timeout decorator properly raise an exception outside of the function's scope
signal handlers which raise exceptions will never work well because the
exception can be raised anywhere in the called code. This leads to
exception race conditions where the exceptions could end up being
hanlded by unintended pieces of the called code.
The timeout decorator was using just that idiom. It was especially bad
because the decorator syntactically occurs outside of the called code
but because of the signal handler, the exception was being raised inside
of the called code.
This change uses a thread instead of a signal to manage the timeout in
parallel to the execution of the decorated function. Since raising of
the exception happens inside of the decorator, now, instead of inside of
a signal handler, the timeout exception is raised from outside of the
called code as expected which makes reasoning about where exceptions are
to be expected intuitive again.
Fixes#43884
* Add a common case test.
Adding an integration test driven from our unittests. Most of the time
we'll timeout in run_command which is running things in a subprocess.
Create a test for that specific case in case anything funky comes up
between threading and execve.
* Don't use OSError-based TimeoutError as a base class
Unlike most standard exceptions, OSError has a specific parameter list
with specific meanings. Instead follow the example of other stdlib
functions, concurrent.futures and multiprocessing and define a separate
TimeoutException.
* Add comment and docstring to point out that this is not hte Python3 TimeoutError
* Move ansible.compat.tests to test/units/compat/.
* Fix unit test references to ansible.compat.tests.
* Move builtins compat to separate file.
* Fix classification of test/units/compat/ dir.
* Properly handle default package manager vs apt
For distros where apt might be installed but is not the default
package manager for the distro, properly identify the default distro
package manager during fact finding and re-use fact finding from
DistributionFactCollector and instead of reimplementing small
portions of it in PkgMgrFactCollector
Add unit test to always check the apt + Fedora combination to test
the new code.
Fixes#34014
Signed-off-by: Adam Miller <admiller@redhat.com>
* remove q debugging output I accidentally left behind
Signed-off-by: Adam Miller <admiller@redhat.com>
* add os_family to the conditional so we're only hitting that code path when needed
Signed-off-by: Adam Miller <admiller@redhat.com>
* setup for a _check* pattern for general os_family group pkg_mgr checking
Signed-off-by: Adam Miller <admiller@redhat.com>
* use Mock.patch decorator for os.path.exists in TestPkgMgrFactsAptFedora
Signed-off-by: Adam Miller <admiller@redhat.com>
Add deps/requires for fact collectors
Fact collectors can now set a required_facts
class attribute that will be a set of the names
of fact collectors they require to be run first.
ie, if a collector needs to know the ansible_distribution,
it should set it's required_facts to include 'distribution'
required_facts = set(['distribution'])
If a collector requires another collector, it gets added
to the selected collector names.
We then topological sort the ordering of the collectors
so that deps work out (ie, 'distribution' will run before
'service_mgr')
required_facts were added to the collectors for:
- network (requires 'distribution', 'platform')
- hardware (requires 'platform')
- service_mgr (requires 'distribution', 'platform')
Fix name references for facts (need 'ansible_' prefix)
is service_mgr
Fixes#30753
The accumulated collected_facts was being update
with new facts _after_ filtering them. So only
facts that pass the filter would ever be passed
to other fact collectors.
For 'filter=ansible_service_mgr', even though it requires
the platform and distribution facts and even collects them,
they would get filtered out and never passed to the other
collectors that need them (service_mgr for ex).
Fix is just to add the unfiltered facts to collected_facts.
Adds unit tests for fact filter and collected_facts.
Fixes#32286
* Make ansible_selinux facts a consistent type
Rather than returning a bool if the Python library is missing, return a dict with one key containing a message explaining there is no way to tell the status of SELinux on the system becasue the Python library is not present.
* Fix unit test
* Fix fact failures cause by ordering of collectors
Some fact collectors need info collected by other facts.
(for ex, service_mgr needs to know 'ansible_system').
This info is passed to the Collector.collect method via
the 'collected_facts' info.
But, the order the fact collectors were running in is
not a set order, so collectors like service_mgr could
run before the PlatformFactCollect ('ansible_system', etc),
so the 'ansible_system' fact would not exist yet.
Depending on the collector and the deps, this can result
in incorrect behavior and wrong or missing facts.
To make the ordering of the collectors more consistent
and predictable, the code that builds that list is now
driven by the order of collectors in default_collectors.py,
and the rest of the code tries to preserve it.
* Flip the loops when building collector names
iterate over the ordered default_collectors list
selecting them for the final list in order instead
of driving it from the unordered collector_names set.
This lets the list returned by select_collector_classes
to stay in the same order as default_collectors.collectors
For collectors that have implicit deps on other fact collectors,
the default collectors can be ordered to include those early.
* default_collectors.py now uses a handful of sub lists of
collectors that can be ordered in default_collectors.collectors.
fixes#30753fixes#30623
* Fix pkg_mgr fact on OpenBSD
Add a OpenBSDPkgMgrFactCollector that hardcodes pkg_mgr
to 'openbsd_pkg'. The ansible collector will choose the
OpenBSD collector if the system is OpenBSD and the 'Generic'
one otherwise.
This removes PkgMgrFactCollectors depenency on the
'system' fact being in collected_facts, which also
avoids ordering issues (if the pkg mgr fact is collected
before the system fact...)
Fixes#30623
previously gather_subset=['!all'] would still gather the
min set of facts, and there was no way to collect no facts.
The 'min' specifier in gather_subset is equilivent to
exclude the minimal_gather_subset facts as well.
gather_subset=['!all', '!min'] will collect no facts
This also lets explicitly added gather_subsets override excludes.
gather_subset=['pkg_mgr', '!all', '!min'] will collect only the pkg_mgr
fact.
* Add 2.0-2.3 facts api compat (ansible_facts(), get_all_facts())
These are intended to provide compatibilty for modules that
use 'ansible.module_utils.facts.ansible_facts' and
'ansible.module_utils.facts.get_all_facts' from 2.0-2.3 facts
API.
Fixes#25686
Some related changes/fixes needed to provide the compat api:
* rm ansible.constants import from module_utils.facts.compat
Just use a hard coded default for gather_subset/gather_timeout
instead of trying to load it from non existent config if the
module params dont include it.
* include 'external' collectors in compat ansible_facts()
* Add facter/ohai back to the valid collector classes
facter/ohai had gotten removed from the default_collectors
class used as the default list for all_collector_classes by
setup.py and compat.py
That made gather_subset['facter'] fail.
It was in lib/ansible/modules/system/setup.py since it
was the only thing using it, but move it back to module_utils
and add a ansible_collector.get_ansible_collector() to build
a facts collector just like the one used by setup.py
mv test_setup.py -> test_ansible_collector.py
All the code it was testing is now in ansible_collector
rm code to create 'ansible_facts' subkey from namespace
Just leave it up to the caller to do, and just return a
flat dictionary from AnsibleFactCollector.collect()
* Add more mount point statvfs info including sizes
Based on https://github.com/ansible/ansible/pull/12073
facts.utils.get_mount_size() now returns a dict of most
of the posix statvfs data, including block_size and inode
counts.
Update the facts.hardware classes that use get_mount_size() to
use the new info by mount_info.update(mount_statvfs_inof) to merge.
* add back unit tests for LinuxHardware mount/fs facts
* add test cases for facts.utils.get_mount_size
* Support NetBSD 7.1+ style ifconfig -a output
network facts on NetBSD after 7.1 cvs would fail
because of format changes in 'ifconfig -a' output.
update code to support new and old format.
add unit tests for both based on
examples from Bruce V Chiarelli.
* wrap use of interfaces.keys() in list() for py3 compat
* sort interface ids for stability
* Fix ansible_cmdline initrd fact for UEFI
UEFI cmdline paths use \ path sep which would
get munged by cmdline fact collection.
* Make CmdLineFactCollector easier to test
extract the parsing of the /proc/cmdline content to
_parse_proc_cmdline()
add a wrapper method for get_file_content _get_proc_cmdline()
Add unit tests of _parse_proc_cmdline based on examples
from issue #23647Fixes#23647
Facts Refresh (2.4 roadmap)
This commit implements most of the 2.4 roadmap 'Facts Refresh'
- move facts.py to facts/__init__.py
- move facts Distribution() to its own class
- add a facts/utils.py
- move get_file_content and get_uname_version to facts/utils.py
- move Facts() class from facts/__init__ to facts/facts.py
- mv get_file_lines to facts/utils.py
- mv Ohai()/Facter() class to facts/ohai.py and facter.py
- Start moving fact Hardware() classes to facts/hardware/*.py
- mv HPUX() hardware class to facts/hardware/hpux.py
- move SunOSHardware() fact class to facts/hardware/sunos.py
- move OpenBSDHardware() class to facts/hardware/openbsd.py
- mv FreeBsdHardware() and DragonFlyHardware() to facts/hardware/
- mv NetBSDHardware() to facts/hardware/netbsd.py
- mv Darwin() hardware class to facts/hardware/darwin.py
- pep8/etc cleanups on facts/hardware/*.py
- Mv network facts classes to facts/network/*.py
- mv Virtual fact classes to facts/virtual
- mv Hardware.get_sysctl to facts/sysctl.py:get_sysctl
- Also mv get_uname_version from facts/utils.py -> distribution.py
since distribution.py is the only thing using it.
- add collector.py with new BaseFactCollector
- add a subclass for AnsibleFactCollector
- hook up dict key munging FactNamespaces
- add some test cases for testing the names of facts
- mv timeout stuff to facts.timeout
- rm ansible_facts()/get_all_facts() etc
- Instead of calling facts.ansible_facts(), fact collection
api used by setup.py is now to create an AnsibleFactCollector()
and call it's collect method.
- replace Facts.get_user_facts with UserFactCollector
- add a 'systems' facts package, mv UserFactCollector there
- mv get_dns_facts to DnsFactCollector
- mv get_env_facts to EnvFactCollector
- include the timeout length in exception message
- modules and module_utils that use AnsibleFactCollector
can now theoretically set the 'valid_subsets'
May be useful for network facts module that currently have
to reimplement a good chunk of facts.py to get gather_subsets
to work.
- get_local_facts -> system/LocalFactCollector
- get_date_time -> system/date_time.py
- get_fips_facts -> system/fips.py
- get_caps_facts() -> system/caps.py
- get_apparmor_facts -> system/apparmor.py
- get_selinux_facts -> system/selinux.py
- get_lsb_facts -> system/lsb.py
- get_service_mgr_facts -> system/service_mgr.py
- Facts.is_systemd_managed -> system/service_mgr.py
- get_pkg_mgr_facts -> system/pkg_mgr.py
- Facts()._get_mount_size_facts() -> facts.utils.get_mount_size()
- add unit test for EnvFactCollector
- add a test case for minimal gather_subsets
- add test case for collect_ids
- Make gather_subset match existing behavior or '!all'
If 'gather_subset' is provided as '!all', the existing behavior
(in 2.2/2.3) is that means 'dont collect any facts except those
from the Facts() class'. So 'skip everything except
'apparmor', 'caps', 'date_time', 'env', 'fips', 'local', 'lsb',
'pkg_mgr', 'python', 'selinux', 'service_mgr', 'user', 'platform', etc.
The new facts setup was making '!all' mean no facts at all, since
it can add/exclude at a finer granularity. Since that makes more
sense for the ansible collector, and the set of minimal facts to
collect is really more up to setup.py to decide we do just that.
So if setup.py needs to always collect some gather_subset, even
on !all, setup.py needs to have the that subset added to the
list it passes as minimal_gather_subset.
This should fix some intg tests that assume '!all' means that
some facts are still collected (user info and env for example).
If we want to make setup.py collect a more minimal set, we can do that.
- force facts_dicts.keys() to a list so py3 works
- split fact collector tests to test_collectors.py
- convert Facter(Facts) -> other/facter.py:FacterFactCollector
- add FactCollector.collect_with_namespace()
regular .collect() will return a dict with the key names
using the base names ('ip_address', 'service_mgr' etc)
.collect_with_namespace() will return a dict where the key names
have been transformed with the collectors namespace, if there is
one. For most, this means a namespace that adds 'ansible_' to the
start of the key name.
For 'FacterFactCollector', the namespace transforms the key to
'facter_*'.
- add test cases for collect_with_namespace
- move all the concrete 'which facts does setup.py' stuff to setup.py
The caller of AnsibleFactCollector.from_gather_subset() needs to
pass in the list of collector classes now.
- update system/setup.py to import all of the fact classes and pass
in that list.
- split the Distribution fact class up a bit
extracted the 'distro release' file handling (ie, linux
boxes with /etc/release, /etc/os-release etc) into its
own class.
- extract get_cmdline_facts -> cmdline.py
- extract get_public_ssh_host_keys -> system/ssh_pub_keys.py
- extract get_platform_facts -> system/platform.py
platform.py may be a good candidate for further splitting.
- rm test for plain Facts() base class
- let the base class for Collector unit tests provide collected_facts
some Collectors and/or their migrated Facts() subsclasses need
to look at facts collected by other modules ('ansible_architecture'
the main one...).
Collector.collect() has the collected_facts arg for this, so add
a class variable to BaseFactsTest so we can specify it.
- mv Ohai to other/ohai.py and convert to Collector
- update hardware/*.py to return facts (no side effects)
- mv AnsibleFactCollector to setup.py
- extra collector class gathering to module method in
facts/__init__.py (collector_classes_from_gather_subset)
- add a CollectorMetaDataCollector collector used to provide
the 'gather_setup' fact
- add unit test module for 'setup' module
(test/units/modules/system/setup.py)
- Collector init now doesnt need a module, but collect does
An instance of a FactCollector() isnt tied to a AnsibleModule
instance, but the collect() method can be, so optionally pass
in module to FactCollector.collect() (everywhere)
- add a default_collectors for list of default collectors
import and use it from setup.py module
eventually, would like to replace this with a plugin loader
style class finder/loader
- unit tests for module_utils/facts/__init__.py
- add unit tests for ohai facts collector
- remove self.facts side effect on populate() in hardware/sunos.py
- convert OpenBSDHardware() to rm side effects on self.facts
- try to rm some self.facts side effects in Network()
plumb in collected_facts from populate() where it is needed.
stop passing collected_facts into Network() [via cached_facts=,
where it eventually becomes self.facts]
- nothing provides Fact() cached_facts arg now, rm it
Facts() should be internal only implementation so nothing
should be using it.
Of course, now someone will.
- add a Collector.name attr to build a map of name->_fact_ids
To properly exclude a gather_subset spec like '!hardware', we
need to know that 'hardware' also means 'devices', 'dmi', etc.
Before, '!hardware' would remove the 'hardware' collector name
but not 'devices'. Since both would end up in id_collector_map,
we would still end up with the HardwareCollector in the collector
list. End result being that '!hardware' wouldn't stop hardware
from being collected.
So we need to be able to build that map, so add the Collector.name
attribute that is the primary name (like 'hardware') and let
Collector._fact_ids be the other fact ids that a collector is
responsible for.
Construct the aliases_map of Collector.name -> set of _fact_ids
in fact/__init__.py get_collector_names, and use it when we are
populating the exclude set.
- refactor of distribution.py
make the big OS_FAMILY literal a little easier to read
Also keys can now be any string instead of python literals
99% sure the test for 'KDE Neon' was wrong
I don't see how/where it should or could get 'Neon' instead
of 'KDE Neon' as provided in os-release NAME=
Use 'distribution' string for key to OS_MAP
ie, we dont need to make it a valid python label anymore so dont.
move _has_dist_file to module as _file_exists
easier to mock without mucking with os.path
mv platform.system() calls to within get_distribution_facts() instead
of Distribution() init.
- remove _json compat module
The code in here was to support:
-a 'json' python module that was not the standard one included
with python since 2.6.
- potentially fallback to simplejson if 'json' was not available.
'json' is available for all supported python versions now so
no longer needed.
- mv get_collector_names -> facts.collector
- mv collector_classes_from_gather_subset -> facts.collector
- mv collector tests from test_facts -> test_collector
- Use six's reduce() in sunos/netbsd hardware facts
- rm extraneous get_uname_version in utils
only system/distribution.py uses it
- Remove Facts() subclass metaclass usage
- using fact_id and a platform id for matching collectors
gut most of Facts() subclasses
rm Facts() subclasses with weird metaclass
only add collectors that match the fact_ids and the platform_info
to the list of collectors used.
atm, a collectors platform_id will default to 'Generic', and
any platform matches 'Generic'
goal is to select collector classes including matching the
systems platform in collector.py, instead of relying on
metaclasses in hardware/*. To finish this, the various
Facts() subclasses will need to be replaced entirely with
Collector() subclasses.
use collector classmethod platform_match() to match the platform
This lets the particular class decide if it is compatible with
a given platform_info. platform_info is a dict like obj, so it could be
expanded in the future.
Add a default platform_match to BaseFactCollector that matches
platform_info['system'] == cls._platform
They were needed previously to trigger a module
load on all the collector classes when we import
facts/hardare so that the Hardware() and related
classes that used __new__ and find_all_subclasses()
would work.
Now that is done in collectors based on platform matching
at runtime we dont need to do it py module import/parse
time. So the non empty __init__.pys are no longer needed
and their is a more flexible mechanism for selection
platform specific stuff.
facts/facts.py is no longer used, rm'ed
- if we dont find an implement class for gather spec.. just ignore it.
Would be useful to add a warn to warn about this case.
- Fix SD-UX typo (should be HP-UX)
- Port fix for #21893 (0 sockets) to this branch
This readds the change from 8ad182059d
that got lost in merge/rebase
Fixes#21893
- port sunos fact locale fix for #24542 to this branch
based on e558ec19cdFixes#24542
Solaris fact fix (#24793)
ensure locale for solaris fact gathering
fixes issue with locale interfering with proper reading of decimals
- raise exceptions in the air like we just dont care.
Pretty much ignore any not exit exception in facts
collection. And add some test cases.
- added new selinux fact to clarify python lib
the selinux fact is boolean false when the library is not installed,
a dictionary/hash otherwise, but this is ambigous
added new fact so we can eventually remove the type dichtomy and normalize it as a dict
Re-add of devel commit 85c7a7b844 to
the new code layout, since it got removed in merge/rebase
The timeout for gathering facts needs to be settable from three places
(highest precedence to lowest):
* programmatically
* ansible.cfg (equivalent to the user specifying it explicitly when
calling setup)
* from the default value
The code was changed in b4bd6c80de to
allow programmatically and the default value to work correctly but
setting via ansible.cfg/parameter was broken.
This change should fix setting via ansible.cfg and adds unittests for
all three cases
Fixes#23753
