[GH-ISSUE #15708] "no such host" error during blob download for specific tags despite active proxy configuration #35776

Open
opened 2026-04-22 20:27:25 -05:00 by GiteaMirror · 0 comments
Owner

Originally created by @cubemac on GitHub (Apr 19, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15708

What is the issue?

Bug Report: Inconsistent DNS resolution when pulling specific model tags behind a proxy

Ollama Version: 0.21.0 OS: macOS (Apple Silicon M4 Pro) Network Environment: Restricted corporate network. All external traffic must go through an HTTP/HTTPS proxy. Internal DNS cannot resolve external hostnames.

Description:

There is an inconsistency in how Ollama handles DNS resolution for different model tags when a proxy is configured. While standard tags pull successfully, specific or longer tags (e.g., qwen3.6:35b-a3b-coding-nvfp4) fail with a no such host error.

Crucially, the failure often occurs after the manifest has been successfully retrieved and the blob download has initiated, suggesting that subsequent connection attempts or retries are bypassing the proxy's DNS resolution.

Steps to Reproduce

Configure proxy environment variables:
export HTTP_PROXY="http://:"
export HTTPS_PROXY="http://:"
export no_proxy="localhost,127.0.0.1"
Start the server: ollama serve
Pull a standard model: ollama pull qwen3.6:35b -> Success
Pull a specific tag: ollama pull qwen3.6:35b-a3b-coding-nvfp4 -> Fail
Observed Behavior

The pull for the specific tag starts, retrieving the manifest and initiating the first blob, but then fails immediately with a DNS error:

pulling manifest
pulling model: 0% ▏ ▏ 11 KB/ 21 GB
Error: max retries exceeded: Get "https://registry.ollama.ai/v2/library/...": dial tcp: lookup registry.ollama.ai on :53: no such host
It appears that while the initial request respects the proxy, certain subsequent calls (possibly for specific blobs or during multi-threaded downloads) attempt a local DNS lookup against the , which is unable to resolve external addresses.

Expected Behavior

Ollama should use the configured proxy for all network operations, including DNS resolution for every manifest and blob request, regardless of the tag length or complexity.

Additional Context

curl -I -x http://: https://registry.ollama.ai/v2/ works consistently in the same environment.
The issue is reproducible even when GODEBUG=netdns=go or netdns=cgo is set.
The fact that it fails after pulling the first few KB of the model blob indicates that the proxy was initially used, but then lost or bypassed for the next connection attempt.

Relevant log output


OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.21.0

Originally created by @cubemac on GitHub (Apr 19, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15708 ### What is the issue? Bug Report: Inconsistent DNS resolution when pulling specific model tags behind a proxy Ollama Version: 0.21.0 OS: macOS (Apple Silicon M4 Pro) Network Environment: Restricted corporate network. All external traffic must go through an HTTP/HTTPS proxy. Internal DNS cannot resolve external hostnames. Description: There is an inconsistency in how Ollama handles DNS resolution for different model tags when a proxy is configured. While standard tags pull successfully, specific or longer tags (e.g., qwen3.6:35b-a3b-coding-nvfp4) fail with a no such host error. Crucially, the failure often occurs after the manifest has been successfully retrieved and the blob download has initiated, suggesting that subsequent connection attempts or retries are bypassing the proxy's DNS resolution. Steps to Reproduce Configure proxy environment variables: export HTTP_PROXY="http://<proxy-host>:<port>" export HTTPS_PROXY="http://<proxy-host>:<port>" export no_proxy="localhost,127.0.0.1" Start the server: ollama serve Pull a standard model: ollama pull qwen3.6:35b -> Success Pull a specific tag: ollama pull qwen3.6:35b-a3b-coding-nvfp4 -> Fail Observed Behavior The pull for the specific tag starts, retrieving the manifest and initiating the first blob, but then fails immediately with a DNS error: pulling manifest pulling model: 0% ▏ ▏ 11 KB/ 21 GB Error: max retries exceeded: Get "https://registry.ollama.ai/v2/library/...": dial tcp: lookup registry.ollama.ai on <internal-dns-ip>:53: no such host It appears that while the initial request respects the proxy, certain subsequent calls (possibly for specific blobs or during multi-threaded downloads) attempt a local DNS lookup against the <internal-dns-ip>, which is unable to resolve external addresses. Expected Behavior Ollama should use the configured proxy for all network operations, including DNS resolution for every manifest and blob request, regardless of the tag length or complexity. Additional Context curl -I -x http://<proxy-host>:<port> https://registry.ollama.ai/v2/ works consistently in the same environment. The issue is reproducible even when GODEBUG=netdns=go or netdns=cgo is set. The fact that it fails after pulling the first few KB of the model blob indicates that the proxy was initially used, but then lost or bypassed for the next connection attempt. ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.21.0
GiteaMirror added the bug label 2026-04-22 20:27:25 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#35776