Make Garage more operator friendly #69

New Issue

Closed

opened 2021-05-01 12:51:02 +00:00 by quentin · 1 comment

quentin commented

2021-05-01 12:51:02 +00:00

Owner

The aim of this issue is to make Garage's development more open and more transparent by clearly recording our early thinking.

Currently, Garage is a bit rough around the edges for operators.
First, I would start with some concrete propositions to make it more operator friendly:

(DONE) Put default values to --ca-cert, --client-cert and --client-key to be able to operate garage without requiring to define an alias. I propose /etc/garage/{garage-ca.crt,garage.crt,garage.key}, the Dockerfile should reflect this change
Add a first colored line named "Cluster Health" to garage status that would indicate the health of the cluster. The value could be OK (green), Node missing (orange), Restoring data (orange), Data unavailable (red). We could maybe add colors to node in the following and suggest an action to do (put the node back online, make it join the cluster, rebuild, etc.)
Report some information about space available in the cluster and per nodes, especially report nodes that could be a bottleneck due to their configuration VS space available on this partition.

For now, the only prospective feature I think of is:

Remove the need for configuring the number of tokens. We could have a sort of gossip averaging that try to regularly rebalance the number of tokens to keep it optimal according to new nodes and partitions sizes. Available space of a node could be still overriden by an operator, for example to create 0 size nodes.

These things are proposals, feel free to discuss them including why they could be bad ideas ^^. Also once a feature starts to be well defined, it might be better to create an independent issue to track it. See this issue as a place for idea incubation around operator friendliness of Garage.

**The aim of this issue is to make Garage's development more open and more transparent by clearly recording our early thinking.** Currently, Garage is a bit rough around the edges for operators. First, I would start with some concrete propositions to make it more operator friendly: - (**DONE**) Put default values to `--ca-cert`, `--client-cert` and `--client-key` to be able to operate garage without requiring to define an alias. I propose `/etc/garage/{garage-ca.crt,garage.crt,garage.key}`, the Dockerfile should reflect this change - Add a first colored line named "Cluster Health" to `garage status` that would indicate the health of the cluster. The value could be OK (green), Node missing (orange), Restoring data (orange), Data unavailable (red). We could maybe add colors to node in the following and suggest an action to do (put the node back online, make it join the cluster, rebuild, etc.) - Report some information about space available in the cluster and per nodes, especially report nodes that could be a bottleneck due to their configuration VS space available on this partition. For now, the only prospective feature I think of is: - Remove the need for configuring the number of tokens. We could have a sort of gossip averaging that try to regularly rebalance the number of tokens to keep it optimal according to new nodes and partitions sizes. Available space of a node could be still overriden by an operator, for example to create 0 size nodes. **These things are proposals, feel free to discuss them including why they could be bad ideas ^^. Also once a feature starts to be well defined, it might be better to create an independent issue to track it. See this issue as a place for idea incubation around operator friendliness of Garage.**