search

openflighthpc / concertim-ansible-playbook

Ansible playbook for building a Concertim appliance
Eclipse Public License 2.0
0 stars 0 forks source link

First pass at fake ganglia data generator #8

Closed benarmston closed 1 year ago

benarmston commented 1 year ago

Add fake ganglia data generator.

The fake ganglia data generator is now installed. The installed source code is patched to remove things that (1) are better done by the ansible playbook; or (2) are specific to the old way of doing things, e.g., shutting down daemons via monit.

In order to have the metrics recognised by the appliance, the meryl daemon needs to be installed too. meryl runs periodic metric parsing and caching. That is it periodically parses the metrics from gmetad and caches them into memcache using Ruby marshal to serialize them.

Much of meryl is written as a C++ library using the Ruby API. This is known as MerylNumberCruncher. Previously, the number cruncher was packaged into a .deb file and installed when the meryl daemon was installed too. Unfortunately, using the previous .deb did not work as it was compiled for i486 not amd64.

MerylNumberCruncher has been compiled for amd64, packaged into a .deb and uploaded to S3. The playbook has been updated to download and install that .deb.

Compiling MerylNumberCruncher exposed some bugs/issues, which are probably due to using a newer compiler. Those issues have been addressed. A future PR will contain a new playbook to automate building our assets and will include the patch.

What works:

  1. Devices, e.g., racks, servers, switches, sensors, pdus, etc. have metrics.
  2. Graphs of those metrics can be viewed.

Limitations:

  1. In-band metrics (in addition to out-of-band metrics) are generated.
  2. Group metrics do not work as the group metric aggregation daemon is not yet installed.
  3. User metrics do not work as the user metrics aggregation daemon is not yet installed.
  4. Threshold states are not calculated as the threshold state calculation daemon is not yet installed.
  5. Metrics are not generated for MIA itself. Previously, the local gmond was used to get real in-band metrics. FSR, gmond isn't working at the moment.
  6. The exact demo data is dependant on the name of the devices. When the demo/default data is added, we probably want to look at the data generated for those devices and see if anything is to be adjusted.