This reverts commit f1bdf6751a.
this causes distribution to be either 50/50 or 100/0. in the latter
case, the single worker usually fails to process pipeline queues in a
timely manner
Change-Id: I7c34acb29290de641ccbef4844e1e6982fe8b173
Enabling everything to be polled and transformed by default is quite heavy and
can lead to a massive amount of unused metrics. Limit the default to a few
metrics known to be actually useful to common deployers.
We keep all pollsters enabled on unit tests and devstack.
Change-Id: Ifb5aa684e575b7904726276b7205824f2dcaa7ce
The [notification] section was missed in the configuration of devstack script.
Change-Id: If955fad4d5c1ceede8a24cd135c6c7b3195be3cf
Closes-Bug: #1678380
We recommend to set pipeline_processing_queues with the
same number of workers agent the infra have.
So do it for devstack too.
Change-Id: I64cc4770a9598f1fdeca4f7476e8965651af4097
We have moved the pipeline config files from etc/ceilometer/ to
ceilometer/pipeline/data/, but ceilometer don't support to load pipeline
config file from /etc/ceilometer/ and only support to load these config
files from the source code path. This is not user-friendly.
Change-Id: Iedebf6b05e123c94eb7bd200f64beb09382c3969
Closes-Bug: #1670238
When the database backend is unconfigured, we currently
try to connection to nothing, and due to the retry logic around
the database connection, the user experience is "ceilometer-upgrade is
stuck".
This change keeps the database upgrade is the database is not
configured. And don't use the connection retry logic for upgrade.
Change-Id: I3fa2a332ecac6cb52c145f437f79b2aba43a4757
These two config options only be used in collector service, see[1], Since
we have deprecated collector service and won't install collector service
in devstack defaultly, these two options will be silently ignored.
Instead, we need to config dispachers by publishers configuration in
pipeline.yaml and event_pipeline.yaml.
[1]
https://github.com/openstack/ceilometer/blob/master/ceilometer/collector.py#L67
Change-Id: I5bbf8f3d11defb2afefddbc756447a8e691c157a
devstack has issues setting multiple value options. it does not
know how to replace properly and ends up just replacing everything.
just delete all previous options if they exist to avoid unexpected
residual values.
Change-Id: I1cb1e4fdddfb9ec7dac321b9e70f688c575a7775
To optimise performance, We have to update ceilometer architecture.
The most important step is to deprecate collector. From now on, we
can configure multiple publishers in pipeline for pushing data to
internal or external system.
Highlight using multiple dispatchers.
Change pipeline publisher and disable ceilometer-collector by default.
Co-Authored-By: gordon chung <gord@live.ca>
Change-Id: I25a6e0b9221844adb4412f1829d9e290b6e198a3
add support for polling specific definition file.
this splits the existing polling specific options out of
pipeline.yaml as transformations only exists on notifcation agent
and polling interval/discovery only exists on polling agents.
backward compatibility is maintained so pipeline.yaml file from
previous releases can still be passed in as polling definition file.
Change-Id: I206566349f98d6b17336cd5ea36ceb1e304dd90c
Since Ie4c6329f7b3d1b4686555f7eea0525649d4b31f6 it's not possible to
deploy Ceilometer without backend, but with panko-api enabled, which
is a valid use case.
Split the if statement into two to make sure we do not execute the
else clause and die early.
Change-Id: Icfafc7bccd7b923707483d903bc5338921d14097
clear any existing on start since devstack doesn't know how to properly
iniset a single option of multiopt
Change-Id: Ie4c6329f7b3d1b4686555f7eea0525649d4b31f6
Since Gnocchi is the recommended backend, and other are deprecated.
We can switch instance_discovery to the libvirt_metadata.
Change-Id: Ib846b87bfb76c359fdec28d6cd258781fd201348
This change introduces a new method to get instances
metadata on the ceilometer-compute-agent.
This switches devstack/gate to libvirt_metadata for gnocchi
Change-Id: Ice1918659be49589a45d7a406044adc0a187aa27