summaryrefslogtreecommitdiff
path: root/0d9e938e92a25673c3712866e2d2108a255b67d5:
diff options
context:
space:
mode:
authorMicah Anderson <micah@riseup.net>2013-02-21 15:30:52 -0500
committerkwadronaut <kwadronaut@leap.se>2013-09-07 01:59:47 +0200
commitf92d09226cfddb0c7e5e342dd199d8ea05b497cb (patch)
tree20b89161162cebdc3301274c8e6129d3f2682bf9 /0d9e938e92a25673c3712866e2d2108a255b67d5:
parenta4ee585cb17751dca8cf49d9361124e438a39d87 (diff)
Due to a network anomaly during a vcsrepo git clone you can get a
'non-repository' at the path. That can be even just an empty directory. This makes it so further puppet runs cannot ever complete properly, because of this error: change from absent to present failed: Could not create repository (non-repository at path) This requires manual intervention to resolve (ie. removing the directory). It is possible to pass the 'force => true' parameter to vcsrepo to handle this, however, all this will do is remove whatever is there, even if it is a valid repository. That means on every puppet run you get the tree removed and then a new clone is done. This is less than ideal for a number of reasons. So this commit adds a test to determine if the location contains a valid repository and then it is used when 'force => true' to determine if it should replace the existing content with a fresh clone. Note: the valid_repo? check is also added to the 'def bare_git_config_exists?' because the test that is done there is intended to look for a bare git repository by checking to see if there 'config' is in the directory. This isn't a particularly robust test because 'config' could be in a directory, when it isn't a bare git repository (such as is common in a rails app) causing vcsrepo to mistakenly think it was a proper bare repository, when it isn't.
Diffstat (limited to '0d9e938e92a25673c3712866e2d2108a255b67d5:')
0 files changed, 0 insertions, 0 deletions