server : allow model downloads at model limit fix issue #26809

[chenmiaoming]

Overview

• Bug: When --models-max is already at capacity, POST /models fails with "model limit reached, try again later" (see Misc. bug: Downloading a new model should be possible even if it would exceed models-max #26809).
• Root cause: The under-lock capacity recheck in load(), intended to prevent concurrent overcommit, also rejects the download child process.
• Fix: Apply the capacity check only for SERVER_CHILD_MODE_NORMAL.
• Safety: Download child instances stay in the DOWNLOADING state, while is_running() only counts LOADED, LOADING, and SLEEPING instances (tools/server/server-models.h). Therefore, downloads do not consume model slots, and the child instance is cleaned up after the download exits.
• Test: Set models_max=1 and preload MODEL_B to occupy the only slot, then start a model download. Verify that the download succeeds and MODEL_B remains loaded.

Additional information

see #26809

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES I used GPT-5.6 Sol to help me with call-stack analysis ,I personally reviewed, understood, and tested the change

[ggml-gh-bot]

Hi @chenmiaoming, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• PR Template not respected: Please respect the template when creating a new pull request. Make sure to fill out all required sections.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.

[ngxson]

/bot review

[ggml-gh-bot]

❌ Code review failed.

Error: `git` exited with code 128