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.
This commit is contained in:
Pascal
2026-07-06 19:26:06 +02:00
committed by GitHub
parent cb295bf596
commit 9abce7473a
3 changed files with 15 additions and 8 deletions

View File

@@ -523,6 +523,7 @@ void server_models::load_models() {
// collect all threads to join in one pass while the lock is held: // collect all threads to join in one pass while the lock is held:
// - monitoring threads from just-unloaded models (to_unload) // - monitoring threads from just-unloaded models (to_unload)
// - threads of finished downloads (DOWNLOADED), they acquire the mutex on exit
// - threads of already-UNLOADED models that are being removed from source // - threads of already-UNLOADED models that are being removed from source
std::vector<std::thread> threads_to_join; std::vector<std::thread> threads_to_join;
for (const auto & name : to_unload) { for (const auto & name : to_unload) {
@@ -535,6 +536,13 @@ void server_models::load_models() {
if (inst.meta.status == SERVER_MODEL_STATUS_DOWNLOADING) { if (inst.meta.status == SERVER_MODEL_STATUS_DOWNLOADING) {
continue; // downloading models are not from config sources, leave them alone continue; // downloading models are not from config sources, leave them alone
} }
if (inst.meta.status == SERVER_MODEL_STATUS_DOWNLOADED) {
// joining this thread under the lock deadlocks: it locks the mutex on its way out
if (inst.th.joinable()) {
threads_to_join.push_back(std::move(inst.th));
}
continue;
}
if (final_presets.find(name) == final_presets.end() && !inst.meta.is_running() && inst.th.joinable()) { if (final_presets.find(name) == final_presets.end() && !inst.meta.is_running() && inst.th.joinable()) {
threads_to_join.push_back(std::move(inst.th)); threads_to_join.push_back(std::move(inst.th));
} }
@@ -550,10 +558,8 @@ void server_models::load_models() {
if (it->second.meta.status == SERVER_MODEL_STATUS_DOWNLOADING) { if (it->second.meta.status == SERVER_MODEL_STATUS_DOWNLOADING) {
++it; // download thread is still busy, skip ++it; // download thread is still busy, skip
} else if (it->second.meta.status == SERVER_MODEL_STATUS_DOWNLOADED) { } else if (it->second.meta.status == SERVER_MODEL_STATUS_DOWNLOADED) {
// download finished, safe to erase // download finished, thread is joined above, safe to erase
if (it->second.th.joinable()) { GGML_ASSERT(!it->second.th.joinable());
it->second.th.join();
}
it = mapping.erase(it); it = mapping.erase(it);
} else if (final_presets.find(it->first) == final_presets.end()) { } else if (final_presets.find(it->first) == final_presets.end()) {
SRV_INF("(reload) removing model name=%s (no longer in source)\n", it->first.c_str()); SRV_INF("(reload) removing model name=%s (no longer in source)\n", it->first.c_str());

View File

@@ -314,7 +314,6 @@ def _wait_for_sse_event(collected: list, event_type: str, model: str, timeout: i
return False return False
@pytest.mark.skip(reason="sse_thread sometimes hangs on GH actions, to be investigated")
def test_router_download_model(): def test_router_download_model():
"""Case 1: download a model, verify SSE events and GET /models.""" """Case 1: download a model, verify SSE events and GET /models."""
global server global server
@@ -358,7 +357,6 @@ def test_router_download_model():
assert MODEL_DOWNLOAD_ID in ids, f"{MODEL_DOWNLOAD_ID} not found in /models after download" assert MODEL_DOWNLOAD_ID in ids, f"{MODEL_DOWNLOAD_ID} not found in /models after download"
@pytest.mark.skip(reason="sse_thread sometimes hangs on GH actions, to be investigated")
def test_router_delete_model(): def test_router_delete_model():
"""Case 2: delete the downloaded model, verify it disappears from GET /models.""" """Case 2: delete the downloaded model, verify it disappears from GET /models."""
global server global server

View File

@@ -31,6 +31,9 @@ import wget
DEFAULT_HTTP_TIMEOUT = 60 DEFAULT_HTTP_TIMEOUT = 60
# per-request timeout, a hung server fails the test instead of stalling the CI for hours
DEFAULT_REQUEST_TIMEOUT = 600
class ServerResponse: class ServerResponse:
headers: dict headers: dict
@@ -330,7 +333,7 @@ class ServerProcess:
path: str, path: str,
data: dict | Any | None = None, data: dict | Any | None = None,
headers: dict | None = None, headers: dict | None = None,
timeout: float | None = None, timeout: float | None = DEFAULT_REQUEST_TIMEOUT,
) -> ServerResponse: ) -> ServerResponse:
url = f"http://{self.server_host}:{self.server_port}{path}" url = f"http://{self.server_host}:{self.server_port}{path}"
parse_body = False parse_body = False
@@ -389,7 +392,7 @@ class ServerProcess:
path: str, path: str,
data: dict | None = None, data: dict | None = None,
headers: dict | None = None, headers: dict | None = None,
timeout: float | None = None, timeout: float | None = DEFAULT_REQUEST_TIMEOUT,
) -> dict: ) -> dict:
stream = data.get('stream', False) stream = data.get('stream', False)
if stream: if stream: