In another confusing addendum to
I41e81d6bac98875eecde2376e0865784626e11a8 (which was already a
confusing addendum to Ieb6a6e9f55bd93f63c3d0a71828c276c2d02e1b9), we
have decided that the refspec used to fetch here is not sufficient to
clear out remote branches everything when updating.
"+refs/heads/*:refs/heads/*" says to replicate everything from the
remote refs/heads into our local refs/heads, but leaves out
refs/remotes/*
The upshot of this is that I41e81d6bac98875eecde2376e0865784626e11a8
will remove the local branches (refs/heads/stable/icehouse, say) but
not remove the remote branches (refs/remotes/origin/stable/icehouse).
The devstack caching script keeps picking up these remote branches,
checking them out, and consequently trying to download old images.
*Nothing* ever removes these branches. In the main dib cache git
update, we also have --prune, but our refspec there is even more
limited (+master:master). This explains why these branches never seem
to die.
Note, an even better mirror would be "+refs/*:refs/*" (in fact, if you
do git clone --mirror, this is what the repo would be setup with to
fetch by default). However, this drags in "refs/changes/*" and all
sorts of other gerrit things. We don't really need them so we just
keep the limiting on.
Change-Id: Ia9c3ffdb2b5f72a45d629961338b415308d6dd21
In a rather confusing addendum to
Ieb6a6e9f55bd93f63c3d0a71828c276c2d02e1b9, we have actually mirrored a
version of the source-repo script from dib and munged it to cache
devstack early so we can use it to find the vm images to download and
cache.
However, we are not ensuring that we remove old branches in this
clone, which is leading to the problems of us picking up images from
old branches that don't exist any more.
Change-Id: I41e81d6bac98875eecde2376e0865784626e11a8
On OVH and Bluebox, the memory layout is such that there is still
significant (~900M) memory above the 8192M address. Increase the
limit to encompass that, which will bring these providers up to
approximately 8G, while increasing hpcloud (which is the actual
target of this restriction) to about 8.5G.
Change-Id: I5c121be55cadad13ad5807968f33b492f9b1e215
When running disk image builder use the build specific checkout of
system-config to install puppet and puppet modules. This allows for easy
local testing of changes to those scripts. Without it master is always
used.
Change-Id: Ifc86d77f702633d630adc1ee619f58607cb56c82
The release has happened, the migration worked, all's right with the
world.
This reverts commit 04eb36588d80dd77a9c692f53470b7d727567604.
Change-Id: I28a3e9cf3b420a6ccca20c52c4e67adc3f5710c5
This commit adds a temporary version cap on subunit2sql to be < 1.0.0.
The 1.0.0 release includes a very large database migration which will
be slow to execute. The python DB api from >=v1.0.0 will not work with
a database that doesn't have the updated schema. So while the migration
is running let's cap the version we install to prevent everything from
breaking while the migration is running. (which might take days)
Change-Id: Iab89beb5c7aba8b744a62f5063e513b72cab0ec2
Turns out we can't do this without dirtying our images by
preinstalling all the libraries/headers our requirements want to
link against. We should revisit the wheel mirror solution which has
already been written as a saner solution.
This reverts commit 1503ef327817bd42cbf888e773f5639e660169a7.
Change-Id: I69ea73b6eb4b620f1ac77467ee3784ec3c813ee2
This installs everything in upper-constraints in a throw away venv,
which has the knock on effect of fully populating the wheel cache for
root with wheels for everything. This hopefully speeds up all devstack
runs quite a bit.
Change-Id: I429f353c33d892f76552c83d34ac1329ed18f97f
The gate-neutron-dsvm-functional-py34 job needs Python.h, and it is
currently not available when running Python 3.
Change-Id: I00ce26864db682a4577dc5360519918007aa5499
Partial-Bug: #1500400
Since I6c5a962260741dcf6f89da9a33b96372a719b7b0 dib has had a
standardised method for ensuring consistency of tracing and error
detection. Bring the tracing for these elements up to that standard,
but maintain the status-quo of flags such as "-e" and "pipefail" by
adding ignore flags where appropriate (we can update these separately
to avoid breakage)
Other minor changes are alphabetical-ordering in the element-deps
files and permissions on prepare-node script
With this, "tox -edib" passes
Change-Id: Ibba1dadb9e819f94294c9d583b83ff698252f93f
We are still working on the caching story for dnf-based builds
(i.e. Fedora 22). There are a couple of options which we will work
through on the linked page. We will sort this out before we move the
devstack job (yet to be created, because the nodes aren't there yet)
out of experimental.
In the mean time, disable the caching in these elements so we can get
Fedora 22 image builds (currently they're in a big looping failure
[1]).
[1] http://nodepool.openstack.org/image.log
Change-Id: I3a435889fc5109d7365240068047aac98abc605e
Rather then explicitly forcing a security context for unbound, we
should restore the default selinux contexts defined by the OS.
Change-Id: I4bb21dcbbcbcff6a5458ebf9478f58d95ad7240a
Signed-off-by: Paul Belanger <pabelanger@redhat.com>
This commit creates a venv for installing os-testr which will enable
all test jobs to have access to the subunit2html utility which has
been moved to live inside the os-testr package instead of as a slave
script.
Change-Id: I2050b54eb2def10438764f3eeb55ecf9caa874dc
This help when reading log files from nodepool. Otherwise we see the
following in the log files:
[1;31mWarning: Config file /etc/puppet/hiera.yaml not found, using
Hiera defaults[0m
Change-Id: I3a865e5107e2749ed44c144539af49e311e0125f
Signed-off-by: Paul Belanger <pabelanger@redhat.com>
The nodepool base element really needs the puppet element, but we didn't
express that in the element deps. That makes a later reorg of elements
more difficult than it needs to be, because our consumption element list
is expressing internal dependency chains.
Change-Id: Ib2296bed1d7f289e7aa2720a38555275de1b6402
The nodepool-base element relies on a couple environement variables.
Describe NODEPOOL_SCRIPTDIR and NODEPOOL_STATIC_NAMESERVER in the
README.rst file.
Change-Id: I56f2aab095a0504e19598d7296d072e7a51b07c2
Invoke bindep from the bindir for its virtualenv in /usr/bindep-env,
and guard its exit code with true since its normal behavior is to
return zero only when all needed packages are already present.
While here, replace calls to tempfile with mktemp since the former
is not readily available on CentOS 6.x.
Change-Id: I926e73f305b1b54f6855cc4acf3407e91b88213a
While git show is being used to build up a list of candidate
packages to cache, it can cause the cache hook to abort if it
returns nonzero because of a nonexistent bindep list file. Guard
that with a fallback to true so that the loop can continue
normally.
Change-Id: Ic3b773ad2e8efa2c450d37a0115355c0e95da853
The version of git on CentOS 6.x is too ancient to support git
branch --list which is necessary to disambiguate -r with a name
pattern. Instead just list all remote branches and filter with grep
like a good Neanderthal.
Change-Id: I4e673e86acb70c205d159e22b5864cf60ae831df
A quirk of diskimage-builder is that no two elements you compose are
allowed to have parts with the same filename. Rename cache-bindep's
install.d/50-download-pkgs so that it doesn't conflict with the one
of the same name in the cache-devstack element.
Change-Id: I5709147fc5bd4c06fb1bba8ea0e068c0eda7d010
Passing a source-repositories ref of "*" signals fetching all heads
and tags rather than just a specific refname. This is desirable for
the warm cache on our job workers, since jobs may run against
arbitrary branches (or possibly even tags). The point is to have a
reasonably complete mirror of each repo, rather than a fairly
minimal clone.
Change-Id: I624ed1e259e007d4246afb45c3a0560598bfbe3f
Depends-On: I4562c9689a8d235ebe09b2f7178aa5890dbc85f1
We intend to use bindep to tell us what distro packages should be
installed at job run-time, so need to similarly identify those in
such a way that they can be cached on our worker images.
Install the latest bindep release from PyPI into /usr/bindep-env
(consistent with the way we provide virtualenvs for access to
zuul-cloner and libraries used by our log uploads to swift).
Concatenate bindep lists from all branches of every hosted Git repo
along with our fallback list in project-config. Run bindep to render
them into a platform-specific manifest of package names and feed
that to the relevant package management tool for the platform to
download this set into the package cache.
Note that similar tooling is not applied to snapshot nodes, since we
only intend to run jobs which use bindep on diskimage-builder
created nodes.
Change-Id: I0c408fee35dfe9b4f700c51565bfaae0a3d03beb
The build for node-devstack will fail because of missing scripts unless
nodepool-base is executed before, so add this as dependency.
Change-Id: Iabc06a542529500ad248ae067beed9639d24e506
Cleanup every use of `` for subshells in the nodepool and tools
directory , replacing them with $(), and finally making the scripts
consistent.
Change-Id: I2b05cd20f9c9a30ab88f8db235aa81da93b1fad3
Some packages may depend on python-setuptools, which is not
installed and cannot be reinstalled on CentOS 6.x once yum has
erased them, so use --skip-broken to avoid aborting. Also on this
platform --downloadonly causes yum to return nonzero even when it
succeeds, so ignore its exit code.
Change-Id: Iaada39ae81e1e47fe9d0bedba80fd19e4e0e6f38
When we're building nodepool images on top of minimal elements, there
will be no cloud-init, and therefore no need to disable cloud-init
datasources. In fact, trying to do so will be an error.
Change-Id: I98887c43566e07f2be9d2dc5fae6538078c7348e
If someone is building an image locally and don't want to do anything
special with nodepool scripts, they should not need to set an env var.
The env var obviously still needs to be honored for nodepool operation
and for override.
Change-Id: I08076a8eafe7019e715b1b46633e8f603031a1ea
The --downloadonly option to yum is provided by the
yum-plugin-downloadonly package. This is merely a virtual package
satisfied by yum itself in newer releases, but an optional package
in older ones such as CentOS/RHEL 6.x. Install it just to be sure it
will work, since we use this to pre-cache RPMs on nodepool images.
Change-Id: I9e2e1605f3721c410180aa46a81b7b731d08503a
On CentOS 6.x the rpm executable is in /bin instead of /usr/bin, but
the cache-devstack element ends up running yum if it wants to
download RPM packages anyway. Look for yum instead of rpm as an
indication of which packages to install.
Commit fa18656 fixed this in cache_devstack.py for snapshot images,
but missed that it was affecting diskimage-builder elements as well.
Change-Id: I4b76a48564cd5d703d79cfda88bbbfe97216a70a
Since DIB creates a chroot but doesn't actually chroot in an
extra-data script, look for /etc/redhat-release relative to the
temporary chroot base path. Also given that the centos element
leaves DIB_RELEASE exported as am empty string in the calling
environment rather than unset, place it first in the conditional
list.
Change-Id: I0a51f88ff5ea71b6aa6e6406b656cefb0e135e88
Change fd9c80208c00021a49df6907973bbbb3fa242069 inadvertently
removed the subprocess import from 55-cache-devstack-repos even
though it was used in a function defined within that element, unlike
in its snapshot prep script counterpart. Readd it so that this
element works again.
Change-Id: I954a1e7e99af96b08e0a59e99e513902731fd773
In the DevStack caching script/element for Nodepool images, use
run_local instead of subprocess.check_output in the _find_images
function. The latter wasn't introduced until Python 2.7 and so won't
work on CentOS 6. The script called from this function returns
quickly and doesn't benefit from non-blocking I/O anyway.
Change-Id: I3129f1f5b3fece321ae132ea1a52b0e156e58365
The centos element (which is CentOS 6.x specific) doesn't provide
DIB_RELEASE so we have to identify it another way.
Change-Id: Ibf3b25b99f03c6077538dec6c3770dc0f690e3f5
Rather than deleting cloud-init, which is going to take longer, just
disable ec2 metadata service. This will be a no-op on rackspace, which
already does this.
Change-Id: I5e8baee50800f7aae474288a914333c21466855a
Once the migration has been run manually we need to use
subunit2sql>=0.4.0 because that is the version of the DB schema running
on the infra db.
This reverts commit aa87c57ca7fe95f9785719acb6b626f12171e240.
Change-Id: I747083bc15ecda7e1926a87c2adcdaf29e565e23
Subunit2sql 0.4.0 was recently released which included a schema change
however the infra db migration takes too long and can't be automated.
(yet?) Until we run the migration we can't use 0.4.0 on any of the
tooling. This caps the version to be less than 0.4.0
Change-Id: I08f113fa904fa962e8f2dc04187ff44e764de47e