... to resolve the following warning.
[WARNING] Deprecated '--debug' flag is set to true (use
'--log-level=debug' instead
Change-Id: Idb412cea64dfc42e3d1223b77f134804eeb7bd60
Cinder and etcd are enabled by default and by default
cinder uses etcd as a distributed lock manager with
tooz as an intermediary. We see a lot of ToozConnectionErrors [1]
in the cinder logs when etcd is backed up [2] which results in
cinder operations timing out causing test failures, like
when a volume is not deleted within a given time.
This changes ETCD_USE_RAMDISK=True by default to try and
alleviate some of the pressure. An alternative is if we know
we're in a single-node job we could just not use a DLM for
Cinder.
[1] http://status.openstack.org/elastic-recheck/#1810526
[2] etcd[26824]: sync duration of 12.076762123s, expected less than 1s
Change-Id: I5f82aa40e9d84114e7b7b5cf19ec4942d6552490
Partial-Bug: #1810526
Seems like for etcd-heavy services like Kubernetes, the fsync
performance of gate VM's are too low [1]. This commit implements an
option to put etcd data directory on RAM disk (tmpfs) to work this
around.
[1] http://lists.openstack.org/pipermail/openstack-discuss/2019-January/001849.html
Change-Id: I5a17099cb9d6941b1a009dc82daefd2c7946d892
This will enable us to run one etcd for devstack and another
for kubernetes in the same box if necessary
Change-Id: Ib71ded24727b80afd4d98eb68bade0f8c0f72311
(1) when checksum fails, better delete the broken files and try the second time;
(2) amazon s3 is not good in mainland China, better try one more time with wget
Change-Id: I24ee73f216b78bd80564863cd335e5d5a9b56360
The client are told to connect to SERVICE_HOST instead of HOST_IP, so
we need to start etcd3 with matching listening parameters.
Change-Id: I96389090180d21d25d72df8f9e8905b850bcaee9
Partial-Bug: 1656329
We need this for every devstack run now, so downloading it from github
every time isn't the most awesome thing in the world.
Add an extra variable EXTRA_CACHE_URLS which will be appended to the
output of tools/image_list.sh. This way, these files will be
downloaded during the daily nodepool build, but they will not be in
the IMAGE_LIST and hence be considered as images to upload.
Add a function get_extra_file which echos the path to a file given the
URL. It will first check the cache at $FILES, and if not present
download it.
Update the documentation in image_list.sh to reflect what's happening.
Move the defaults for etcd variables into stackrc, since it is a base
service now.
Change-Id: I86104824a29d973a6288df1f24b7891feb86267c
This separates out the install phase early from the start phase to
make this mirror other services in devstack.
Depends-On: I4124dc7e3fd3b4d973979da85209ec991c0f8c4b
Change-Id: I76f8740448b25a48869ee80006e826baa6cafc2b
In Multi host deployments, it is possible to run ETCD in a different
host than the SERVICE_HOST (where all the controllers run). This patch
brings that distinction.
Change-Id: I15fe6f25eedf1efebaab81cce26b080577b856cc
Signed-off-by: Antoni Segura Puimedon <antonisp@celebdor.com>
Function "_install_etcd" is trying to use "files" directory
to download a file. Instead of this, this directory should be
$FILES, which is defined previously in parent script.
TrivialFix
Change-Id: I643ce3b9aba1f65f03524430c748bf120d071509
gpg verification requires network connectivity which is non
mirrorable. We try to avoid that in devstack whenever possible. A
sha256sum is a totally reasonable way of knowing if the downloaded
package is valid.
Closes-Bug: #1693092
Change-Id: Id496ab53f76444f08dc6961f1ecd25f450cc96d7
We need to handle this better, for now, just don't install
etcd in the sub nodes. We need to setup the proper clustering
mechanism if we want to have etcd3 running in multiple nodes
Change-Id: I8dd385e3c993942473e67d04367cdf74495dbeef
ETCD_DOWNLOAD_URL is set to github url, in our CI, we can point
ETCD_DOWNLOAD_URL to a url in tarballs.openstack.org possibly
in devstack-gate
Download the etcd binaries and drop them into /opt/stack/bin and
use it from there. Cache the tgz for subsequent use (local workflow)
daemon-reload is called twice once from inside the write_user_unit_file
and then when we adjust the entries with additional things recommended
by the etcd team. We need a better way to do this in the future.
Added a TODO to verify the downloaded artifact later. The etcd team
posts gpg signature, we could verify that or run sha256sum and hard
code that in lib/etcd3 file. We would have to update it whenever we
bump the etcd3 version.
We use the public key "CoreOS Application Signing Key <security@coreos.com>"
with ID FC8A365E to verify the integrity of the downloaded file
Any jobs that need to be run on architectures where v3.1.7 is not available
should rey the v3.2.0-rcX release candidates. We can switch to v3.2.0
when it gets released.
Initial version of this code was borrowed from the dragonflow
repo:
http://git.openstack.org/cgit/openstack/dragonflow/tree/devstack
Change-Id: Ibbb430fb1dbf66942168e0cb52d990ab6a2eb8d7
