| *** akhil_jain has joined #openstack-watcher | 05:11 | |
| *** akhil_jain has left #openstack-watcher | 05:11 | |
| josecastroleon | hi | 07:24 |
|---|---|---|
| *** licanwei has joined #openstack-watcher | 07:44 | |
| *** chenke has joined #openstack-watcher | 07:46 | |
| chenke | hi~ | 07:46 |
| licanwei | josecastroleon: hi~ | 07:46 |
| josecastroleon | sorry i went for a coffee | 07:53 |
| josecastroleon | we did an investigation during summer about Watcher, and we tried several strategies | 07:55 |
| licanwei | josecastroleon: so what's your plan? | 07:56 |
| josecastroleon | we found some rough edges but in general we were pretty happy with the tool | 07:56 |
| josecastroleon | we have some use cases that we want to contribute that are variations of the existing ones | 07:57 |
| josecastroleon | we are using extensively cells for scale the infrastructure out, and in some cases they are not well balanced | 07:57 |
| josecastroleon | we can use the VM workload balancing strategy but we need to limit it somehow to a subset of hosts | 07:58 |
| licanwei | about your use cases, can you give me some details ? | 07:58 |
| josecastroleon | 1) VM workload balancing for cells | 07:59 |
| licanwei | we can set audit scope through parameter | 07:59 |
| josecastroleon | 2) Scheduler for preemptible instances | 07:59 |
| licanwei | we have a scope parameter in audit | 08:00 |
| josecastroleon | 3) continuous audit on hyperconverged servers | 08:00 |
| josecastroleon | on the scheduler of preemptibles, normally we receive hardware and it takes some time to be filled by our clients. As we have other workloads that can get benefit, we can set a continuous audit looking for holes and spawning new instances there | 08:01 |
| josecastroleon | on the continuous audit for hyperconverged, we have a setup in which the compute and storage lives together. In this setup we didn't have any fixed limit to protect CPU or IO and we want to do again workload balance keeping some free resources for IO | 08:03 |
| josecastroleon | those are the use cases that we want to contribute at the moment | 08:04 |
| licanwei | great! | 08:05 |
| josecastroleon | As rough edges, and linked to what I said previously, when fetching data on a large cloud, some operations on nova are pretty expensive. | 08:05 |
| josecastroleon | retrieving data from 10k servers and all the machines hosted takes too long | 08:05 |
| josecastroleon | this is why we would like to add a scope | 08:05 |
| licanwei | I got it | 08:07 |
| licanwei | you want to add a scope to limit Watcher datamodel | 08:08 |
| josecastroleon | yes | 08:09 |
| josecastroleon | finally, we are not using neither monasca, nor ceilometer nor gnocchi for the metrics. We have all the metrics in a set of influxDBs with grafana in front. | 08:09 |
| licanwei | good idea :) | 08:09 |
| josecastroleon | so we may add a datasource for grafana-proxy, if someone is interested in it | 08:10 |
| licanwei | How you get metrics from influx? Rest API or sql? | 08:11 |
| josecastroleon | our monitoring team has almost a influxdb instance per metric, so we are using grafana-proxy for that http://docs.grafana.org/http_api/data_source/ (rest api) | 08:13 |
| licanwei | if you don't use ceilometer, how you get metrics from data sources? | 08:17 |
| licanwei | and about rough edges, Watcher need add a new scope for datamodel and a new datasource for metrics, right? | 08:22 |
| josecastroleon | yes | 08:23 |
| josecastroleon | I need to run now for a meeting, i hope to come back soon | 08:23 |
| josecastroleon | sorry | 08:23 |
| licanwei | ok | 08:23 |
| licanwei | we can discuss tomorrow if no time today :) | 08:24 |
| *** licanwei has quit IRC | 09:24 | |
| *** chenke has quit IRC | 09:25 | |
| *** josecastroleon has quit IRC | 15:01 | |
| *** ianychoi has joined #openstack-watcher | 17:06 | |
| *** Dantalion has joined #openstack-watcher | 17:58 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!