push speed limit

[ZYWNB666]

System version: Ubuntu20.04

docker version: 27.5.1

docker compose version: 1.29.2

System memory: 256G

cpu core count: 144

Disk: nvme * 16

minio: 14+2ec

The above are my system performance parameters

My harbor uses Minio's s3 as the backend storage and starts harbor with docker-compose.yml. Currently, it takes 2 minutes and 30 seconds to upload a 3.5G image layer. We believe this speed does not meet expectations.

Question

Are there any other soft restrictions on harbor installed using docker?

[ZYWNB666]

This's my minio version.

$ minio  --version
minio version RELEASE.2024-12-18T13-15-44Z (commit-id=16f8cf1c52f0a77eeb8f7565aaf7f7df12454583)
Runtime: go1.23.4 linux/amd64
License: GNU AGPLv3 - https://www.gnu.org/licenses/agpl-3.0.html
Copyright: 2015-2024 MinIO, Inc.

[MinerYang]

Hi @ZYWNB666 ,

thanks for contacting with us. Generally , we don't set up any speed limit from our side. It depends on your proxy set up and the overall network conditions.

Best, Miner

[ZYWNB666]

Hi @ZYWNB666 ,

thanks for contacting with us. Generally , we don't set up any speed limit from our side. It depends on your proxy set up and the overall network conditions.

Best, Miner

Thank you for your reply. May I ask where I can find the configuration to increase the speed? My current hardware resources, including bandwidth, have no bottlenecks at all. Could it be because of a single copy of the harbor component?

[slcnx]

后端换一下看看效果，使用排除法看看

[reasonerjt]

@ZYWNB666

Thanks for reaching out to us.

I agree the suggestion by @slcnx

后端换一下看看效果，使用排除法看看

We wanna clarify whether this is caused by the overhead of Harbor. Could you setup OSS distribution pointing to the same minio as a backend storage and test the speed?

[ZYWNB666]

@ZYWNB666

Thanks for reaching out to us.

I agree the suggestion by @slcnx

后端换一下看看效果，使用排除法看看

We wanna clarify whether this is caused by the overhead of Harbor. Could you setup OSS distribution pointing to the same minio as a backend storage and test the speed?

Yes, I ruled out many options

1. lvm: 3.5G upload takes 2 min 30 seconds
2. minio: 3.5G upload takes 2 min 12 seconds
3. Bare disk nvme: 3.5G takes 2 min 2s This 3.5G layer for timing is one image layer. Other image layers have cache and are not included in the calculation

[wy65701436]

@ZYWNB666 can you setup a vanilla docker distribution(v2.8.3) with your storage? And try to compare the push scenario between harbor and docker distribution.

• If the difference is larger than expected, we can focus on the Harbor side to check for potential bottlenecks.
• If the difference is relatively small, I suspect the root cause is more likely related to your infra's performance.

[ZYWNB666]

@ZYWNB666 can you setup a vanilla docker distribution(v2.8.3) with your storage? And try to compare the push scenario between harbor and docker distribution.

• If the difference is larger than expected, we can focus on the Harbor side to check for potential bottlenecks.
• If the difference is relatively small, I suspect the root cause is more likely related to your infra's performance.

I don't understand. I'm currently pushing to harbor using the docker distribution and the docker push command

[wy65701436]

distribution

The Docker Distribution (v2.8.3) is just one component of Harbor, responsible for storing OCI artifacts and interacting with backend storage. Harbor, however, provides many additional features by integrating other components such as Harbor Core, Postgres, Redis, and more.

In your case, since the issue is related to push performance, we should try narrowing down the root cause. One approach is to set up a standalone Docker Distribution instance backed by your Minio and run the same push tests with a Docker client. If the performance results differ from what you observe with Harbor, it suggests that the overhead may come from Harbor’s additional components. From there, we can continue narrowing down which component is contributing to the performance issue.

[github-actions[bot]]

This issue is being marked stale due to a period of inactivity. If this issue is still relevant, please comment or remove the stale label. Otherwise, this issue will close in 30 days.