Multi-disk usage
In case you have multiple disks on your server, you can configure Supervisely to store data on multiple disks, including the Agent data. This will allow you to better utilize your server resources and increase the overall performance of your instance, sometimes it's necessary to store data on multiple disks due to the limited capacity of the OS drive.
Docker data
To change the docker data folder location you first need to find its current location. To do that, run the following command:
After locating the directory you need to stop the docker service:
(Optional) Then you need to move the data to the new location. This step is optional, because you can always easily download the images again, moving the docker folder if there are many docker images will take a lot of time:
Now you need to create a symlink to the new location:
Finally, you need to start the docker service:
Now all of your docker data will be stored in the new location /new/path/docker
Supervisely server data
Please check this article first for more details. You can link any folder in DATA_PATH
to a different location.
Here's an example of how you could link storage to a different location:
I'm assuming that you already have a folder /new/path/for/storage
with the correct permissions and the value of DATA_PATH
from .env
is /supervisely/data
Agent data
Agent uses 2 folders to store its data:
user persistent data
agent internal data, like tasks logs and caches
You can configure each folder using the advanced settings in the Agent Instructions UI. Start -> Team Cluster -> 3 dots -> Instructions.
You should configure both "Agent host directory" and "Folder to mount" with the path to the new location. For example, you might want to have a dedicated folder for agents on another drive: /mnt/storage/supervisely/agents
.
Then you can use it like this: Agent host directory: /mnt/storage/supervisely/agents
Folder to mount: /mnt/storage/supervisely/agents/agent-1
After making the changes, click the blue button to copy the command and run it on the agent machine.
Last updated