Public issue detail

Runaway CPU investigation for llama-server: timer churn at mt_find

llama-server is stuck in a likely timer churn loop: 8.42% of sampled CPU passed through mt_find, with repeated thread backtraces show 22 thread(s) around 0x00007f03399b14be in ?? () from target:<path> and 1 thread(s) around 0x00007f03399b14be in ?? () from target:<path>.

highscore: 106reports: 1

Last seen: 2026-07-09 01:53 UTC. Public JSON: /v1/issues/019f4473-347f-75d0-8ee4-58cc8558c55b

Technical snapshot

This is the clearest retained userspace thread cluster Fixer captured while the process was spinning.

Representative thread backtrace

  • Command: /usr/local/lib/ollama/llama-server --model /usr/share/ollama/.ollama/models/blobs/sha256-797b70c4edf85907fe0a49eb85811256f65fa0f7bf52166b147fd16be2be4662 --port 33163 --host 127.0.0.1 --no-webui --offline -c 512 -np 1 --log-verbosity 4 --no-log-prefix --no-log-timestamps --flash-attn auto --embedding -b 512 -ub 512 -t 8 --context-shift --keep 4
  • Why Fixer classified it this way: Timer-related syscalls or symbols dominate the sample, which suggests a wakeup timer is firing too aggressively.
  • Thread summary: thread backtraces show 22 thread(s) around 0x00007f03399b14be in ?? () from target:<path> and 1 thread(s) around 0x00007f03399b14be in ?? () from target:<path>
  • Contention signals: gdb-stderr: gdb: warning: Couldn't determine a path for the index cache directory.
  • Repeated loop: futex -> futex -> futex
  • Top syscalls: futex x1241, <... futex resumed>) = -1 ETIMEDOUT x192, <... futex resumed>) = -1 EAGAIN x26, close x25
  • Kernel: 6.17.10+deb14-amd64
  • Distribution: debian
0x00007f03399b14be in ?? () from target:<path>
0x00007f03399a5c84 in ?? () from target:<path>
0x00007f03399a629c in ?? () from target:<path>
0x00007f03399a8828 in pthread_cond_wait () from target:<path>
0x00007f0339cbaaa7 in std::condition_variable::wait(std::unique_lock<std::mutex>&) () from target:<path>++.so.6
0x00007f033a65c90b in std::thread::_State_impl<std::thread::_Invoker<std::tuple<common_log::resume()::{lambda()#1}> > >::_M_run() () from target:<path>
0x00007f0339cec446 in ?? () from target:<path>++.so.6
0x00007f03399a9289 in ?? () from target:<path>

Possible duplicates

These are suggestions based on sanitized trigram similarity plus structured fields like package, subsystem, classification, and wait site. They are not auto-merged.

Runaway CPU investigation for ollama: timer churn at fdget

possible duplicate

ollama is stuck in a likely timer churn loop: 48.28% of sampled CPU passed through fdget, with repeated thread backtraces show 1 thread(s) around 0x000055c5c39ed143 in ?? () and 1 thread(s) around 0x000055c5c39ed143 in ?? ().

highscore: 106reports: 2successful triagesimilarity: 86%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-07-03 18:33 UTC. Public page: /issues/019de4bd-5222-7312-bfda-fe64185d7301. Public JSON: /v1/issues/019de4bd-5222-7312-bfda-fe64185d7301

Runaway CPU investigation for ollama: timer churn at __get_user_8

possible duplicate

ollama is stuck in a likely timer churn loop: 33.33% of sampled CPU passed through __get_user_8, with repeated thread backtraces show 1 thread(s) around 0x000055c5c39ed143 in ?? () and 1 thread(s) around 0x000055c5c39ed143 in ?? ().

highscore: 106reports: 1similarity: 85%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-06-27 06:58 UTC. Public page: /issues/019e7555-afaf-7610-9965-6b375b9aa4e9. Public JSON: /v1/issues/019e7555-afaf-7610-9965-6b375b9aa4e9

Runaway CPU investigation for llama-server: unknown userspace loop at mt_find

possible duplicate

llama-server is stuck in a likely unclassified userspace loop: 11.72% of sampled CPU passed through mt_find, with repeated thread backtraces show 22 thread(s) around 0x00007f1cf9ca74be in sem_trywait () from target:<path> and 1 thread(s) around 0x00007f1cf9ca74be in sem_trywait () from target:<path>.

highscore: 106reports: 1similarity: 85%

Why this looks related: same subsystem, same target, same wait site

Last seen: 2026-07-10 03:42 UTC. Public page: /issues/019f427c-2154-77c0-818c-eded7911a06a. Public JSON: /v1/issues/019f427c-2154-77c0-818c-eded7911a06a

Runaway CPU investigation for ollama: timer churn at __pollwait

possible duplicate

ollama is stuck in a likely timer churn loop: 25.00% of sampled CPU passed through __pollwait, with repeated thread backtraces show 1 thread(s) around 0x000055c5c39ed143 in ?? () and 1 thread(s) around 0x000055c5c39ed143 in ?? ().

highscore: 106reports: 1successful triagesimilarity: 83%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-07-04 10:34 UTC. Public page: /issues/019d5a4d-2a99-7f50-8b1c-afedc974159d. Public JSON: /v1/issues/019d5a4d-2a99-7f50-8b1c-afedc974159d

Runaway CPU investigation for ollama: timer churn at __hrtimer_setup

possible duplicate

ollama is stuck in a likely timer churn loop: 25.00% of sampled CPU passed through __hrtimer_setup, with repeated thread backtraces show 1 thread(s) around 0x000055c5c39ed143 in ?? () and 1 thread(s) around 0x000055c5c397a50e in ?? ().

highscore: 106reports: 1similarity: 83%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-07-01 04:38 UTC. Public page: /issues/019d85a7-37c6-7671-8000-ffbb4d39e629. Public JSON: /v1/issues/019d85a7-37c6-7671-8000-ffbb4d39e629

Runaway CPU investigation for ollama: timer churn at __schedule

possible duplicate

ollama is stuck in a likely timer churn loop: 25.00% of sampled CPU passed through __schedule, with repeated thread backtraces show 1 thread(s) around 0x000055c5c39ed143 in ?? () and 1 thread(s) around 0x000055c5c39ed143 in ?? ().

highscore: 106reports: 1similarity: 83%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-07-03 07:58 UTC. Public page: /issues/019d923e-9063-7b13-831f-2476cfee31b3. Public JSON: /v1/issues/019d923e-9063-7b13-831f-2476cfee31b3

Published attempts

No public attempts have been published for this issue yet.

Quarantine stays on by default so the public queue is harder to spam.