Age | Commit message (Collapse) | Author |
|
Instead of creating tenant/user/role and service/endpoint for all
openstack services in keystone postinstall, now each of the services
creates its own keystone identities by queueing them up in its postinstall
to a file /etc/keystone/service-user-setup. service-user-setup
script, when run as the last postinstall, calls identity.sh with keystone
identity parameters to create necessary identities for the services.
Signed-off-by: Andy Ning <andy.ning@windriver.com>
|
|
Ceilometer ceilometer/hardware/inspector/snmp.py and some
some ceilometer builtin tests require python pysnmp package.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
|
|
ceilometer tests against various databases: mongodb, db2,
sqlalchemy, hbase. For mongodb and db2, tests use environment
variables: CEILOMETER_TEST_MONGODB_URL and CEILOMETER_TEST_DB2_URL
for supplying URL database connection into tests. As we do not
support mangodb or db2, by not setting or setting bogus URLs to
these two environment variables will cause tests to fail.
These tests should be skipped as prerequisite is not satisfied.
Therefore we modify ceilometer testcases for mongodb and db2
to be skipped instead of failed if CEILOMETER_TEST_MONGODB_URL and
CEILOMETER_TEST_DB2_URL are not set.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
|
|
Commit e5b7606eaa94033256eb1aca5f6a6eed4f9e54a0 from
https://github.com/openstack/ceilometer.git master
With the recent introduction of the identity_uri in keystoneclient,
there is a small backward incompatible change which means that weird
protocols like the ones we use during tests aren't support properly
anymore. The patch replaces "foottp" and "barttp" by "file" hopefully
keeping testing coverage the same.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
|
|
Many ceilometer built-in testcases failed because they
can not find ceilometer configuration file, as these tests
use relative paths to find these configuration files.
So forcing these failure testcases to look for these
configuration file using absolute path.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
|
|
There are some missing ceilometer meters (e.g. vcpu) which have
origin from notification. This is due to ceilometer-agent-notification
service is not started on controller node.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Editing the files in ${WORKDIR} using sed or similar tools as part of
do_install means they can only be edited once. Supplying a modified
CONTROLLER_IP in local.conf and building the image again will not
result in the CONTROLLER_IP being properly updated since the
substitution placeholders will no longer exist. We therefore simply
swap the other of things, installing the configuration files first,
then editing them to swap the placeholders. This means we can run the
do_install again and again and get the results we expect.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Updating to the icehouse ceilometer, which has more functionality and
features than the havana version.
Also, we have to fix the resource query for postgres versus mysql.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Package and enable the bash completion suppor that comes with the
various client packages.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Memory leak happens when the dynamic list times grows without anything to
reset it.
In ceilometer and cinder configuration files, the new option is created:
[nova_client]
max_timing_buffer=<value>
In all clients found that uses extends the HTTPClient and uses the times
list, we limit the size of the list by popping off the oldest item in the
list to maintain a maximum size. A default size of 200 is chosen,
configurable by the above configuration option.
Signed-off-by: Amy Fong <amy.fong@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Currently all the openstack components have default start level
of 20. There are other services such as glusterfs, rabbbitmq,
database... are also starting at the same start level. On some
platform, this can cause racing condition between services which
in turn causes some of openstack components not started.
By adjusting the openstack components start level to higher will
ensure that system services start in the determistic way.
Signed-off-by: Vu Tran <vu.tran@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Several python packages require 'python-pbr' both at build and
runtime, as listed in their respective setup.py files, yet this
dependency is not included in their recipe. Adding python-pbr
to the RDEPENDS to correct this.
In addition this situation is complicated by the fact that the
setuptools will actually fetch python-pip and python-pbr eggs,
regardless of the value of BB_NO_NETWORK, if any of these packages are
built before python-pip and python-pbr are in the sysroot. Most
dramitically if you were to attempt to build any of these packages
with no network connectivity the do_compile() task will fail with the
following:
| DEBUG: Executing shell function do_compile
| Download error: [Errno 110] Connection timed out -- Some packages may not be found!
| Couldn't find index page for 'pip' (maybe misspelled?)
| Download error: [Errno 110] Connection timed out -- Some packages may not be found!
| No local packages or download links found for pip>=1.0
| Traceback (most recent call last):
| File "setup.py", line 21, in <module>
| pbr=True)
Adding the missing DEPENDS will ensure these packages are available
without the need for setuptools to fetch them, and avoid possible
build issues due to network connectivity.
In order to test these modifications all of these packages have been
built with a populated sstate cache and the network crippled using:
iptables -A OUTPUT -p tcp --destination-port 80 -j DROP
to ensure no extra fetches are taking place.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
The havana ceilometer postgres (sqlalchemy) support has significant issues.
Rather that perform signficicant backports, we'll uprev and pick up the
latest development stream, that addresses many issues out of the box.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Installation from package feeds shows some missing REDPENDS for the
-setup packages.
Signed-off-by: Rob Wolley <Rob.Woolley@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
From the patch:
sqlalchemy: fix ceilometer resource query
Implement the abandoned ceilometer fix from:
https://review.openstack.org/#/c/59204/ [Fix for get_statistics with postgresql]
Without this fix, the ceilometer resource tab in horizon returns an error due
to badly formed SQL and a database backend error.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Cherry picking two ceilometer master changes to address postgresql database
issues:
https://bugs.launchpad.net/ceilometer/+bug/1241526
https://review.openstack.org/#/c/49456/
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Create and package initscripts from alarm-notifier and alarm-evaluator. These
are added as dependencies of the controller package, so the functional mix
at runtime doesn't change, with the exception that the agents are started on
boot of the controller.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
After moving all database creation initialization packages, we also
remove it from the RDEPENDS of the various control node recipes.
This allows images to select database initialization or skip it.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Many OpenStack modules require a first boot action to setup up users,
databases, bridges, etc. These same packages install initscripts to start
daemons and servers.
The 1st boot package post install actions immediately exit to indicate
that the action cannot be performed in the cross environment and instead
should be done on first boot. The update-rc.d post install actions are
intended to be run in the cross environment to symlink scripts into the
proper runlevels.
The early exit from the db setup routines, means that the rc files are
not linked in host cross. If the rootfs doesn't contain update-rc.d,
they also will not be set up on first boot. The end result is a system
that does not start all of its required services on boot.
To fix this, we split out db and other first boot setup tasks into
dedicated (but empty) -setup packages. These run on first boot, while
update-rc.d is left to create the proper symlinks.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Import and package the basic ceilometer package, havana version. This
minimal configuration creates:
- init scripts
- packages the compute, control and APIs
- creates the configuration and database
It is expected to need extension in the future.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|