]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
server: fix deadlock in load_models() when erasing a finished download (#25358)
authorPascal <redacted>
Mon, 6 Jul 2026 17:26:06 +0000 (19:26 +0200)
committerGitHub <redacted>
Mon, 6 Jul 2026 17:26:06 +0000 (19:26 +0200)
commit9abce7473ad9e6870080caf08e2dcf8592e086b1
tree0d50b45eb36ce849003d268367a88116bbdd7233
parentcb295bf59663cd3577389315636772f4060bd1f5
server: fix deadlock in load_models() when erasing a finished download (#25358)

* server: fix deadlock in load_models() when erasing a finished download

The download monitoring thread acquires the models mutex on its way out,
but load_models() joined it from the erase loop while holding that mutex.
Join it outside the lock via threads_to_join like the other monitoring
threads.

* server: add default timeout to test requests

A hung server now fails the test after 10 minutes instead of stalling
the CI job for hours. Explicit timeouts are unchanged.
tools/server/server-models.cpp
tools/server/tests/unit/test_router.py
tools/server/tests/utils.py