Closed artek-koltun closed 1 year ago
the annotations looks ok, but creating new message
TcpController
should be discussed and in a separate PR
ok, I can remove TcpController
for a while.
However, I want we all to agree on where we want to keep pcie related part: in the existing NvmeRemoteControllerService
service or in a dedicated one within backend_nvme_pcie.proto
@jainvipin @sburla-marvell @GottliebNoam?
@artek-koltun yes, please remove TcpController
for now
and we will add have this discussion
See https://github.com/opiproject/opi-api/issues/198 and AIP-203
First, let's try to discuss where we are going to keep functionality to connect to local Nvme PCIe drives for JBOF case, since it would affect how to annotate the fields properly. I see an empty separate backend_nvme_pcie.proto, but also I see
NVME_TRANSPORT_PCIE
enum value inNvmeTransportType
in backend_nvme_tcp.proto file, which might imply reuse of the service for PCIe.Some assumptions/considerations in the first patch:
NvmeRemoteControllerService
for PCIeFabricsPath
message was created, which is part ofNvmePath
. It allows to annotate some fields asREQUIRED
for Nvme over fabrics transport types and have more consistent API. Otherwise, we need to keep themOPTIONAL
, since we can use the same API to connect to local PCIe.TcpController
was created to be consistent with previous item and isolateTCP
related optional parameters. This message is a part ofNvmeRemoteController
.