Add server/config package.#320
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a new configuration schema for the Bootz server, including Protobuf definitions, Bazel build rules, and generated Go code. The review feedback focuses on improving the Protobuf design and ensuring consistent Go package naming. Specifically, it is recommended to use the bytes type instead of Base64-encoded strings for binary data like certificates and keys to be more idiomatic and efficient. Additionally, the reviewer suggested adding option go_package to the proto file and updating the Bazel importpath to ensure the generated Go package name correctly reflects the intended structure.
Chounoki
force-pushed
the
pr3
branch
2 times, most recently
from
f673437 to
7ef981f
Compare
|
Is this for Monax? If so, I would like to see how this fits in with the existing protos or proposed workflow. Do you have any example configs for this? |
Yes, this is for Monax, so the vendors can use all real data like ownership_voucher. |
| // A binding configuration for Bootz server. | ||
| message Config { | ||
| // Bootz server address (IP:port). | ||
| string server_address = 1; |
There was a problem hiding this comment.
How will we know the server address of the Bootz server ahead of time? Some runtime environments (e.g. Google's) don't let you reserve a specific IP when bringing up SUT components. We will need another way to learn about the Bootz server address.
There was a problem hiding this comment.
Here is my understanding:
With Monax, each SUT will run in a container, which by default is assigned to a private network by Docker runtime. This private network is bridged (NAT'ed) by the host machine. (I.e. the host machine acts as the gateway of the SUT's private network).
https://docs.docker.com/engine/network/
This private network address doesn't really matter, because any external switch will only see the real IP address of the host machine on which the containers run. In other words, the host machine does port mapping from container private network to host public network.
The DHCP entries must use the real host IP address as Bootz server address. Similarly the image URI created by Bootz must also use the real host IP address, because to any external switches, only the real IP address of the host machine is visible (All IP addresses inside the private network of the containners are not visible externally).
There was a problem hiding this comment.
Monax isn't necessarily going to run on Docker containers. Some vendors may run these SUT components on bare-metal machines, and as I said above, it definitely wouldn't work in Borg. This is why we should find a solution that can be applied to any runtime/deployment environment.
There was a problem hiding this comment.
If the vendors are going to directly run the Monax test we provide to them, then each SUT will be inside a container just like the math example provided by Monax, you can see each SUT has its own Dockerfile and kubernetes.txtpb file.
If the vendors don't plan to run the Monax, but to use each service on bare machine without Monax, then they already know the IP of the machine which can be manually written into the textproto of this file by the vendors, and then Bootz server will be initialized from this textproto.
In either case, personally I think it is better and safer to ask the vendors to manually fill in the IP address of the machine on which they plan to run the test (regardless of Bootz server inside a container or not), since the machine can have multiple network interfaces and multiple IP addresses.
While Bootz calls net.Listen() function without specifying an IP, it will create a listener on each of those network interfaces and their IP addresses (https://pkg.go.dev/net#Listen), which means if we try to retrieve a single server IP address dynamically from Bootz runtime, we would get multiple IP addresses instead, and it would be a mess because we can only put one single Bootz IP into DHCP.
There was a problem hiding this comment.
I think my main point still stands: How will we run this Monax test on Borg if you have to know the IP address ahead of time? We need to think about all future cases, not just the immediate ones.
Monax at its core is a SUT manager. It's purpose is to automatically bring up, manage and provide access to SUT components without having to know the implementation/addressing details about them. It is responsible for figuring out the IP addresses of the SUT components it brings up, and then initiating the gRPC/HTTP/whatever connections to them. By introducing external fields that override this, you are potentially breaking that binding.
Regarding net.Listen() listening on all available IP addresses, I don't see that as a huge problem. Sure, the server will need to pick one address to provide in the DHCP lease but the beauty of Monax is that each vendor can customize that logic as much as they want. This is why I've put in a GetBootzURL() RPC to the Bootz controller: Each test implementation has full control over how the Bootz server determines what its correct address is.
| // Whether Streaming Bootz is supported or not. | ||
| bool streaming_supported = 5; | ||
| // Software image to be loaded on the chassis. | ||
| bootz.SoftwareImage intended_image = 6; |
There was a problem hiding this comment.
The Bootz SoftwareImage contains a download URL. How will we know that URL ahead of time? This URL should be generated by the ImageService SUT dynamically.
There was a problem hiding this comment.
I think the image URI must also use the real IP address of the host machine, which is known by the user doing the test beforehand.
In the worst case, Bootz can still override this IP if required (i.e. we can provide a function for Monax to call).
In that case, could you either edit the existing Monax protos under server/tests/proto or move the config there? The reason I ask is because this file is very similar to server/tests/proto/test.proto |
Besides Monax, this proto definition will be referenced by Even without doing any test (Monax or not), this proto definition is still required. |
I'm not sure how this config proto will be used in the production (non-test instance). Doesn't this config have to contain an exhaustive list of the inventory and other artifacts? In the real world, we fetch these from external systems. |
This config profo is only for the open source Bootz. This config proto itself as well as its consumers: Our internal Bootz will use a different version of |
I have created another PR of |
2 participants
No description provided.