docker compose setup #15

Open

pradykaushik wants to merge 5 commits from pradykaushik/issues/issue-13-docker-compose-setup into master

pradykaushik commented

2019-11-08 00:17:48 +00:00

(Migrated from github.com)

Added Dockerfile to help containerize Elektron.
Added docker-compose yaml file to setup mesos cluster for elektron.

The docker image tagged rdelvalle/mesos-master:1.5.1 was used as the mesos master.
As the mesos agent for Elektron requires additional software (docker, PCP) pkaushi1/mesos-agent-elektron:1.5.1 was built using rdelvalle/mesos:1.5.1 as the base image.

Use the below commands to run elektron.

docker-compose run elektron

1. Added Dockerfile to help containerize Elektron. 2. Added docker-compose yaml file to setup mesos cluster for elektron. The docker image tagged [`rdelvalle/mesos-master:1.5.1`](https://github.com/ridv/aurora-docker/blob/master/mesos/master/Dockerfile) was used as the mesos master. As the mesos agent for Elektron requires additional software (docker, PCP) [`pkaushi1/mesos-agent-elektron:1.5.1`](https://github.com/spdfg/elektron-mesos-agent-dockerfile) was built using [`rdelvalle/mesos:1.5.1`](https://github.com/ridv/aurora-docker/blob/master/mesos/Dockerfile) as the base image. Use the below commands to run elektron. ```commandline docker-compose run elektron ```

ridv (Migrated from github.com) reviewed 2019-11-08 00:52:48 +00:00

docker-compose.yaml

ridv (Migrated from github.com) commented

2019-11-08 00:52:48 +00:00

We own the organization spdf at dockerhub https://hub.docker.com/orgs/spdf/repositories

We own the organization `spdf` at dockerhub https://hub.docker.com/orgs/spdf/repositories

ridv commented

2019-11-08 00:54:01 +00:00

(Migrated from github.com)

Were you able to find a workaround for getting RAPL metrics?

pradykaushik (Migrated from github.com) reviewed 2019-11-08 20:53:25 +00:00

docker-compose.yaml

pradykaushik (Migrated from github.com) commented

2019-11-08 20:53:25 +00:00

Good to know. I'll push the image there then.

pradykaushik commented

2019-11-08 21:10:06 +00:00

(Migrated from github.com)

Were you able to find a workaround for getting RAPL metrics?

Not yet. Currently, the CPU and Memory utilization is measured of the physical host (not the containers).
There are multiple ways forward from here.

If we stick to the current approach (the one in the PR), then it would require PCP to be installed on the host machine. Definitely not the ideal scenario.
If we want to monitor the CPU and Memory utilization of the docker containers then we could do so by using pmdadocker and monitoring the below metrics. However, note that PCP would still need to be installed on the host machine. If we want to measure RAPL metrics, we would need to pull data from the host machine.
- docker.cpu_stats.cpu_usage.usage_in_kernelmode
- docker.cpu_stats.cpu_usage.usage_in_usermode
- docker.memory_stats.usage
If we want to refrain from having to install anything on the host machine, then I believe the only way forward is to read off a PCP log file baked into the image (this is the approach you suggested the other day). I can see two advantages with this.
- Testing becomes easier as now we have a fixed PCP log.
- Power measurements can also be part of the PCP log file.
If we go with option 3, it would not require installation of PCP on the host machine.

Let me know your thoughts on this.

> Were you able to find a workaround for getting RAPL metrics? Not yet. Currently, the CPU and Memory utilization is measured of the physical host (not the containers). There are multiple ways forward from here. 1. If we stick to the current approach (the one in the PR), then it would require PCP to be installed on the host machine. Definitely not the ideal scenario. 2. If we want to monitor the CPU and Memory utilization of the docker containers then we could do so by using `pmdadocker` and monitoring the below metrics. However, note that PCP would still need to be installed on the host machine. If we want to measure RAPL metrics, we would need to pull data from the host machine. - `docker.cpu_stats.cpu_usage.usage_in_kernelmode` - `docker.cpu_stats.cpu_usage.usage_in_usermode` - `docker.memory_stats.usage` 3. If we want to refrain from having to install anything on the host machine, then I believe the only way forward is to read off a PCP log file baked into the image (this is the approach you suggested the other day). I can see two advantages with this. - Testing becomes easier as now we have a fixed PCP log. - Power measurements can also be part of the PCP log file. If we go with option 3, it would not require installation of PCP on the host machine. Let me know your thoughts on this.

This repository is archived. You cannot comment on pull requests.

No reviewers