shimpa1
commented
2 months ago Good morning,
- provider address: akash16q46mn8tm7vtwn4h9rugr8ptch6shp2qda7y89
- provider URI name resolution works: yes
- provider ingress URI name resolution works: yes
- provider status endpoint works: yes
- provider GRPC status endpoint works: yes
- provider attributes set correctly: yes
- provider responding to orders: yes
- test deployment works: yes (without GPU)
- network functional: yes
Other remarks:
- The provider seems low on resources (CPU and RAM). Highly suggest using server hardware with much more resources. Suggestion: 32 cores, 128GB RAM or more.
- Provider storage is a RAID0 array with uneven drives:
sda 8:0 0 111.8G 0 disk ├─sda1 8:1 0 1G 0 part └─sda2 8:2 0 110.7G 0 part nvme1n1 259:0 0 953.9G 0 disk └─md0 9:0 0 1.4T 0 raid0 /etc/resolv.conf /etc/hostname /dev/termination-log /etc/hosts nvme0n1 259:1 0 465.8G 0 disk └─md0 9:0 0 1.4T 0 raid0 /etc/resolv.conf /etc/hostname /dev/termination-log /etc/hosts
This is bad practice on multiple levels: 1. RAID0 generally decreases stability by adding yet another SPOF. 2. RAID0 needs equally sized drives to work properly.
- The provider lists RTX 3090 which seems to be allocated on node2 however it's not available for lease. The GPU is also not currently leased:
curl -sk https://provider.gpu3090.ddns.net:8443/status | jq { "cluster": { "leases": 3, "inventory": { "active": [ { "cpu": 100, "gpu": 0, "memory": 100663296, "storage_ephemeral": 68157440 }, { "cpu": 100, "gpu": 0, "memory": 268435456, "storage_ephemeral": 268435456 }, { "cpu": 100, "gpu": 0, "memory": 100663296, "storage_ephemeral": 6291456 } ], "available": { "nodes": [ { "name": "node1", "allocatable": { "cpu": 4000, "gpu": 0, "memory": 12308656128, "storage_ephemeral": 224812917593 }, "available": { "cpu": 3080, "gpu": 0, "memory": 12087369728, "storage_ephemeral": 224812917593 } }, { "name": "node2", "allocatable": { "cpu": 8000, "gpu": 1, "memory": 8126627840, "storage_ephemeral": 1349073758432 }, "available": { "cpu": 5195, "gpu": 0, "memory": 3229161472, "storage_ephemeral": 1348194003168 } } ] } } }, "bidengine": { "orders": 0 }, "manifest": { "deployments": 0 }, "cluster_public_hostname": "provider.gpu3090.ddns.net", "address": "akash16q46mn8tm7vtwn4h9rugr8ptch6shp2qda7y89" }
This suggests that there's an issue either with the nVidia drivers, nVidia toolkit, or K8S plugin : Please refer to https://akash.network/docs/providers/build-a-cloud-provider/gpu-resource-enablement/
Please fix these issues and we can move on.
Shimpa
jakestanton2016
andy108369
Prerequisite Steps:
1. Make sure your provider has community provider attributes and your contact details (email, website):
Ref documentation:.
2. Make sure your provider *.ingress resolves to your provider IP (ideally worker node IP)
Example:
3. Please make sure your Akash provider doesn't block any Akash specific ports.
Audit Steps:
1. Title the issue: " [Provider Audit]: Provider Address" (e.g. "[Provider Audit]: provider.europlots.com")
2. Wait for response via comments. If no issues during provider Audit, process will be complete, provider should start bidding on leases, and Audit ticket will be closed.
3. If there are issues during the provider Audit, debug those issues, and Audit will be complete.
4. Audit Issue will be closed by core team member.
Leave contact information (optional)