* Properly handle redirected stderr
When running subprocess.Popen.communicate() on a command run with stderr
redirected to subprocess.STDOUT, the 2nd element of the return tuple
(representing the stderr) will be None, since all of the standard error
was sent to stdout.
Functions interpreting the return data from cmd.run_all will be
expecting the ``stderr`` key in the return dict to contain a string, not
a NoneType, so this commit forces the stderr to be set to an empty
string when redirect_stderr is True.
* Look at stdout instead of stderr for error text when redirect_stderr is True
* Hash fileclients by opts
There was an issue whereby the cache of the fileclient was being overwritten
by dueling minion instances in multimaster mode. This protects them by hashing
by the id of opts.
Closes#25040
* Silly typo!
* Remove tests which do not test any actual functionality or are too tightly coupled to the implementation
* Handle empty acl_name in linux_acl state
Calls to setfacl interpret an empty group or user name to mean to be the
owner of the file they're operating on. For example, for a directory
owned by group 'admin', the ACL 'default:group::rwx' is equivalent to
'default:group:admin:rwx'.
The output of the getfacl execution module returns ACLs in the format of
'group:admin:rwx' instead of 'group::rwx'. This commit changes the
acl.present state to look for the owner of the file if the acl_name
paremeter is empty.
* Fix acl.present/acl.absent changing default ACLs
The behaviour of the acl.present and acl.absent is to check the data
structure returned by getfacl contains a key by the name of acl_type.
However, this data structure does not contain any default ACLs if none
exist, so this check will fail. We omit the check if a default ACL was
passed into the state functions.
Unfortunately, the call to modfacl may fail if the user passes in an
acl_type such as 'default:random'. In this case the state will appear to
succeed, but do nothing.
This fixes the state module to allow setting default ACLs on files which
have none.
* If cache_jobs: True is set, populate the local job cache
Fixes#32834
Allows a masterless minion to query the job cache.
* Refactor cache_jobs functionality to be DRY
When creating a new user, if a group of the same name already exists,
the usermgmt.conf file is consulted to determine the primary group.
It's in these cases that the parsing bug is triggered.
This code change addresses several of the existing issues:
- The previous split statement explicitly specified a single space.
Since a config line may have any number of spaces and/or tabs
surrounding the entries, the resulting array's elements may be
incorrect.
- According to the man pages for usermgmt.conf, the "group" config
entry accpets a single parameter -- so we shouldn't iterate.
- The "val[1]" was returning the 2nd letter of each word and not the
second word on the config line as intended.
* salt.utils.gitfs: fix formatting for warning messages
When git_pillar support was added to salt.utils.gitfs, the
recommendation globals had string formatting placeholders added to them,
but the locations where these values are referenced do not call
``.format()`` to properly replace them. This commit fixes that
oversight.
* Remove more gitfs and master-specific wording from log messages
* Fix servermanager module
- Added check for 2008 version of windows
- Added Import-Module ServerManager to _pshell_json.
Apparently this needs to run each time we issue a
servermanager command.
* Fix list_available
Method _finger_fail method from SAuth to AsyncAuth class to make method available
in both class and fix an issue where _finger_Fail is called inside AsyncAuth.
* salt-cloud: fix ipv6-only virtual machines
* fix hostname for rsync fallback in scp_file function
* use 4 spaces instead of 2
* remove global variable, use direct socket call instead
