* Fix AWS EC2 inventory plugin caching of groups
* Added changelog fragment for aws_ec2 caching fix
* Store the AWS query results
The underlying inventory object contains inventory from other sources,
so caching it as ours would be wrong.
It seems easiest and safest to just cache the boto query results
instead.
* Remove unused functions
(cherry picked from commit 7ba09adee1)
* Documentation for docker_container: fix documentation for memory_swappiness. Default value will not equal 0, it will be inherited from the host machine (#47296)
(cherry picked from commit 4c2efa4b67)
* Add changelog.
* [2.7] Update some cliconf plugins (#47141)
* Add `check_all` to many community cliconf plugins
(cherry picked from commit 7844a40)
Co-authored-by: Nathaniel Case <this.is@nathanielca.se>
* Add changelog entry
* Adoc fixes (#47137)
* removed hardcoded loader/plugins list
* updated a few errors to keep orig object
* fix httpapi/cliconf listing
* ansible-doc fixes
* show undocumented as UNDOCUMENTEd
* added missing undoc
(cherry picked from commit fce9673ac1)
* An earlier optimization of ansible-doc -l caused failures. (#47012)
The optimization quickly searches the plugin code for short_description
fields and then uses that in the -l output. The searching was a bit too
naive and ended up pulling out malformed yaml. This caused those
plugins to be omitted from the list of plugins of that type with
a warning that their documentation strings were wrong.
This change makes the documentation parser aware that the documentation
string could have a relative indent for all of its fields which makes it
robust in the face of this particular problem.
* Don't search for space after short_description:
Any whitespace would be valid. In particular newline
(cherry picked from commit 61ae6424a3)
Python sets the SIGPIPE handler to SIG_IGN. On execv() signal handlers are
reset to their defaults, EXCEPT those that are SIG_IGN which are left ignored.
In Python 3 subprocess.popen explicitly resets the SIGPIPE handler to SIG_DFL,
but unfortunately in Python 2.7 it does not. This leads to subprocesses being
executed with SIGPIPE ignored. This is often a problem with bash scripts which
rely on SIGPIPE to terminate commands in a pipe, but can easily be a problem
with other applications.
This implements the Python 3 behaviour for Python 2.7 by using a preexec_fn.
(cherry picked from commit f2dccb9)
Co-authored-by: Jonathan Oddy <jonathan.oddy@transferwise.com>
* Fix in confirmed_commit capability in netconf_config modules
Fixes#46804
* If confirm value is greater than zero or confirm_commit option is set and confirmed-commit
capability is not supported but Netconf server only in that case fail the module
* Update confirm-commit flag
* Update changelog
(cherry picked from commit 5394638047)
* Locate prtdiag even when absent from /usr/bin
On Solaris 8 hosts, this prevents fact collection from aborting with:
Argument 'args' to run_command must be list or string
* Lint fix.
* Style: pass /usr/platform/.../sbin as optional path to get_bin_path().
(cherry picked from commit 40fb992a6f)
* Refactoring related to network device
* Assign unique random temporary key while creating SCSI or/and IDE controller devices
* Add testcase for this change
Fixes: #38679
Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
(cherry picked from commit fd985db72d)
Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
If enable_snat is False, this should be used to build the
request, because the default value in the OpenStack Networking
API is True.
Fixes the issue #45915.
(cherry picked from commit 452a4ab781)
* yum module properly check for None config_file (#46641)
* yum module properly check for None config_file
* add conf_file test cases to yum integration tests
Signed-off-by: Adam Miller <admiller@redhat.com>
(cherry picked from commit fb6e91bf98)
* add changelog for 2.7 backport
Signed-off-by: Adam Miller <admiller@redhat.com>
* yum module handle list optional empty strings properly (#46634)
Fixes#46517
Signed-off-by: Adam Miller <admiller@redhat.com>
(cherry picked from commit ad405fc21e)
* add changelog
Signed-off-by: Adam Miller <admiller@redhat.com>
* [aws_ssm_parameter_store] use describe_parameters paginator (#45632)
Fix the service reaching an internal limit while processing the results and returning unexpected data
(cherry picked from commit 7aaa5da41d)
* changelog
* nmcli: fix vlan connection modification Fixes#42322 (#42415)
* ensure optional items are set to empty strings rather than not presented
fix syntax of vlan modification command
* extended tests for nmcli
(cherry picked from commit fb72a5424c)
* Changelog fragment for nmcli fix
* win_reboot: Fix for ignore post_reboot_delay (#46360)
* win_reboot: Fix for ignore post_reboot_delay
This fixes an issue where win_reboot would be ignoring the provided
post_reboot_delay (and on Windows timing/waiting is everything!)
This must be backported to the v2.7 branch.
* Merge post-reboot handling into run()
(cherry picked from commit 8c64b4af7c)
* Add a changelog for the win_reboot fix
* Return correct version on installed VyOS (#39115)
* Return correct version on installed VyOS
Previously existing regexp will shows only "VyOS" without numeric output of router version.
For example: from "Version: VyOS 1.1.6" only VyOS will be written in ansible_net_version variable
For more informative output numeric value should be returned as well
* Fixed unittests
(cherry picked from commit 235b11f681)
* Added changelog
* module/systemd: fix logic determining if a service needs to be enabled (#46245)
* modules/systemd: fix logic: allow scope to default to 'system'
Fix logic introduced in 7ea909418e5: if 'scope' param is not specified,
it defaults to system, but the value of module.params['scope'] is None,
not 'system' - so allow for that.
* modules/systemd: fix logic: disabled means disabled
Fix logic determining whether a service with both systemd and initd files is enabled or disabled.
In situations where systemd thinks service is disabled, but rc.d symlinks mark it as enabled,
this module wrongly assumes the service is enabled.
Fix this logic: disabled means disabled
Only when the output from does NOT include disabled, consider the status of rc.d symlinks.
This essentially replicates the fixes done to the systemd handling in the "service" module in 3c89a21e0cFixes#22303Fixes#44409
(cherry picked from commit ef131c7556)
* backport/2.7/46245: add changelog fragment
* [aws] route53 module: fix idempotency for CAA records (#46049)
* Fixing record order for CAA records to properly handle idempotency.
* Add integration tests that reproduce CAA failure
(cherry picked from commit a727a1ee67)
* Added changelog.
* Sorting args.
* Doing comparisons of options with container parameters in a more context-sensitive way.
This prevents unnecessary restarts, or missing restarts (f.ex. if parameters are removed from ``cmd``).
* Make blkio_weight work.
* Fix cap_drop idempotency problem.
* Making groups idempotent if it contains integers.
* Make cpuset_mems work.
* Make dns_opts work.
* Fixing log_opts: docker expects string values, returns error for integer.
* Adding tests from felixfontein/ansible-docker_container-test#2.
* Make uts work.
* Adding changelog entry.
* Forgot option security_opts.
* Fixing typo.
* Explain strict set(dict) comparison a bit more.
* Improving idempotency tests.
* Making dns_servers a list, since the ordering is relevant.
* Making dns_search_domains a list, since the ordering is relevant.
* Improving dns_search_domains/dns_servers.
* Fixing entrypoint test.
* Making sure options are only supported for correct docker-py versions.
* [2.7] Clean up after ansible-connection if failure occurred in start() (#45929)
(cherry picked from commit 0d143ed)
Co-authored-by: Nathaniel Case <this.is@nathanielca.se>
* Add changelog fragment
* rewrite get_resource_pool method for correct resource_pool selection
* only keep name if path is given for cluster, esxi_hostname or resource_pool
* Revert "only keep name if path is given for cluster, esxi_hostname or resource_pool"
* This reverts commit 50293ec763c024b0eaceac5d775ccc0ad3ff8bd7.
* if the name argument contains a path, only use the last part for matching
* remove path from cluster argument in tests
* remove find_objs in favour of reusing find_obj with an extra folder argument
* fix find_obj ignoring first if name is not given
(cherry picked from commit 1a810f8f11)
* Don't simply ignore container in present() if image is not specified.
* Use image from existing container for recreation if not specified.
* Added changelog.
* Improve comment.
(cherry picked from commit 895019c59b)
* [rabbitmq_binding] Fix the quoting of vhost and other names, which was broken in PR #35651
* Merge missing urllib_parse.quote from PR #42422
* Missed one line, where also needs to be escaped, i.e., the destination
(cherry picked from commit d5f8738bf2)
Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
Fixes#35120 : the redis cache plugin keeps key/value
entries in an in-memory cache to avoid hitting the
redis database each time.
The problem is that a cache entry is only set when
a value is get or set but it is always deleted when
trying to delete a value.
When the --flush-cache ansible-playbook option is used,
the redis cache plugin is first asked to remove every
entry corresponding to every hostname present in the inventory.
As no value as been set/get so far, it then tries to delete
an unexisting value from the cache and hence crashes with
a KeyError exception.
(cherry picked from commit ee3dfef016)
Jonathan Oddy