* Limit askpass prompts to single attempt
OpenSSH client option NumberOfPasswordPrompts defaults to 3 so in case
an incorrect connection password is provided it is excessively tried 3
times. Not only that but running the `_ssh_askpass` entry point multiple
times (via ssh) results in `json.decoder.JSONDecodeError` as after the
first run the shared memory is zero'd and the subsequent runs end up
calling `json.loads` on empty data.
`json.decoder.JSONDecodeError` does not happen prior to Python 3.13 as
the share memory is unlinked automatically on `.close()` and the
`_ssh_askpass` entry point exits with return code 1 before attempting to
load zero'd memory.
Fixes#85359
* changelog and tests
* Update changelogs/fragments/85359-askpass-incorrect-password-retries.yml
Co-authored-by: Matt Davis <6775756+nitzmahone@users.noreply.github.com>
* Update lib/ansible/cli/_ssh_askpass.py
Co-authored-by: Martin Krizek <martin.krizek@gmail.com>
* Avoid race condition in second unlink
---------
Co-authored-by: Matt Davis <6775756+nitzmahone@users.noreply.github.com>
Co-authored-by: Matt Clay <matt@mystile.com>
(cherry picked from commit 54ccad9e46)
* replace usage of `IOError` as it is an alias to `OSError`
* replace usage of `socket.error` as it is an alias to `OSError`
* use subclasses of `OSError` rather than inspecting `errno`s
* utilize `exist_ok` parameter of `os.makedirs` rather than ignoring
`FileExistsError`
Make the following changes to the exception handling this patch already
updates to be consistent with the new code:
* use `ex` as a name for exception being handled
* use `from ex` when re-raising exception for additional context
* use f-strings and `!r` for quoting
* pass exceptions to the `exception` parameter of `fail_json`
* use `display.error_as_warning` rather than passing stringified
exception into `display.warning`
Co-authored-by: Matt Clay <matt@mystile.com>
(cherry picked from commit 600c1e67b4)
* Add ssh-agent launching, and ssh-agent python client
* Move things around, is this better??
* docs
* postpone creating dir after bin lookup
* fix method name
* changelog ssh agent
* address reviews
* fix typing
* do not redefine public_key
* typing
* more typing
* Catch OSError when starting ssh agent
* likely copy pasted old code
* var type fix
* why is this needed?
ci_complete
* ignoring the change for now
* write out pub key file atomically
* defensive timeout for the socket
* _populate_agent docstring
* do not allow setting these in config
* check expected length before slicing blobs
* test all key types
* remove lock/unlock functionality
* docstring
* private _ssh_agent
* .
* launch agent in cli and ansible_ssh_*
* additional info for ssh-agent comment
* Add tests for remove and remove_all
* comment on os.rename
* hopefully mitigate agent startup/delays problems
* exceptions
* unused import
* fix sanity
* perf
---------
Co-authored-by: Matt Martz <matt@sivel.net>
* Improve performance of including tasks into the play
PlayIterator.add_tasks is used to insert tasks from an include into the
play for particular host. It makes a copy of the current block including
the tasks within the block and inserts the new tasks from the include
into the copied block. But there is no need to make copies of tasks
within the block, what we want is a "shallow" copy of the block. This
PR changes that to copy the block excluding the tasks within.
On a contrived playbook with 50 include_role tasks, each role has 1 task,
running on 10 hosts the running time is reduced from ~55s to ~44s in my
environment.
ci_complete
* Add changelog
* do not mention subprocess.Popen, it is an implementation detail
* use "shell injection vulnerability" to describe the problem that is
well-known, or easily searched for on $search_engine and do not link
to a 3rd-party site that is also specific to a tool for Python
Prevents `Group state for \"customenvgroup\" not found` error which may
or may not be a regression in dnf5. Just name groups/envs uniquely to
workaround the issue.
* Ensure implicit flush_handlers have a parent block
To avoid getting tracebacks when calling methods like ``get_play()`` on
them.
* task needs to be copied
* copy only when necessary
It is rare that we need to provide information like "before and after
the change". In majority of cases we just ignore "Additional
information" when submitting PRs. We can just put the needed information
into the summary if needed and remove the section from the template,
simplifying it.
In ``Handler.notify_host`` we ensure that ``Handler.notified_hosts`` can contain
particular host at most once. Therefore for removing a host it should be
faster to use ``list.remove`` which removes the first item in the list,
as opposed to using list comprehension removing all such items.
This greatly reduces run time on large inventories since meta tasks are
executed in the main process sequentially and just executing them is expensive.
This change avoids running the following implicit meta tasks:
* ``flush_handlers`` on hosts where no handlers are notified
* ``noop`` for the linear strategy's lockstep, instead hosts that are
not executing the current task are just not part of the current host loop
A playbook consiting of two simple plays both running on ~6000 hosts
runs in:
devel: 37s
this PR: 1.3s
Co-authored-by: Sloane Hertel <19572925+s-hertel@users.noreply.github.com>
Co-authored-by: Sviatoslav Sydorenko (Святослав Сидоренко) <wk.cvs.github@sydorenko.org.ua>
Since we don't really care about the type we don't have to
query for it and just retrieve the filename value.
Unfortunately we cannot use module_utils.urls.get_response_filename
as we don't have the response object, so just utilize
email.message.Message to parse the filename
instead of manually doing the work ourselves.
Fixes: #83690
* Rely on dnf.base.remove, no special handling isn't needed,
let the dnf internals figure out what is needed to be done.
This is more in line with what dnf cli does.
* "already installed" in Exception (if it is even a thing) should be
caught by special exceptions like MarkingError or CompsError. This
appears to be a historic check that is no longer needed.
Supersedes: #83295
This change simplifies construction and footprint of testing rpm
repository created by rpmfluff:
* all packages default to noarch
* only when necessary build package for a specific architecture(s)
* do not build several repositories each for specific arch
* remove duplicate "incompatible arch" test
* skip_broken_and_nobest: move testing packages from an external repo
into our dummy repo for transparency
* remove compatibility code from create_repo.py for versions no longer
needed
* remove support for old OS versions from setup_rpm_repo
* simplify representation of an rpm in create_repo.py to allow easier
changes
Notes
* there is one more external testing repo used at
https://ci-files.testing.ansible.com/test/integration/targets/setup_rpm_repo/repo-with-updateinfo
* dnf: follow-up on is_newer_installed arches fix
* fix for the non package object case
* prevent early bailout in dnf _is_newer_version_installed
* non-installed available arches would fail out of the check early
---------
Co-authored-by: Matt Davis <nitzmahone@redhat.com>
* remove no longer needed tasks
* remove workarounds
* consolidate conditionals
* remove distros/versions from conditionals that are no longer present
in CI
This fixes "Arch dependent binaries in noarch package" error cause by
including files created by make_elf function in noarch packages. While the
error only manifests itself on EL 7 and 8 it is better to use files
suitable for noarch packages to prevent the error potentially
re-occuring in the future.
In #80094 support for var substitution for cachedir was added but there
are more options that should be supported. Using an API for
prepend_installroot which should be done anyway provide that feature
so use that. In addition, perform the operation once all substitutes
are in place (releasever as well).
best/nobest options are one of the options whose default values are set
by an OS distribution. For example in our CI, both Fedora and RHEL set
the best option to different default values. As such we should defer to
the distributions for the default value and not change it by default but
if users wish to change it they can do so explicitly.
Currently the dnf module sets the nobest option inconsistenly and not for
all cases. This patch fixes that to reflect the behavior described
above. In addition adding the best option for both dnf and dnf5 modules
since the best option is prefer to nobest in dnf while in dnf5 nobest is
completely removed in favor of best.
Fixes#82616
Jinja may generate an invalid Python source code from a template. Trying
to compile such source code into a Python code object results in
SyntaxError being thrown. An example of such a template is providing the
same keyword argument into a lookup twice, resulting in:
`SyntaxError: keyword argument repeated`.
Since `jinja2.exceptions.TemplateSyntaxError` does not cover such a
case, as it is not a Jinja parsing error, we need to catch SyntaxError
explicitly ourselves.
Fixes#82606
This moves handling of callbacks and play recap stats from
_load_included_file to individual strategies so include_role tasks are
accounted for, not just include_tasks.
Fixes#77336
When templating tags (which happens outside of standard `post_validate`) we
need to template each object in the inheritance chain and set the templated
values on those objects individually. That way when `task.tags` is called the
`extend` functionality properly picks up the templated values of all
parents into one flatten list.
Fixes#81053
