Ollama: Scheduler Reliability Overhaul
Two related pull requests fix scheduler bugs that could leave dead or evicted model runners stuck in a broken state, while a separate fix corrects flawed scoring logic in the vision test suite.
Duration: PT2M31S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-08-02T13:00:55Z
- Audio duration: PT2M31S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good day, and welcome to Ollama's briefing for August 2nd, 2026.
Today's clear signal comes from the scheduler. Two pull requests from raymondginger2018-sudo tackle the same underlying problem: runners getting stuck in a state that looks healthy but isn't.
PR 17516 addresses a case where the llama-server process behind a loaded model crashes or wedges after loading finishes. The scheduler wasn't checking whether the process had actually exited, so "ollama ps" and health checks would report everything as fine, even though the model was dead. Requests would hang until…
PR 17515 fixes a related deadlock. When a runner is marked for eviction, the scheduler sets its session duration to zero so it unloads once idle. But a concurrent request taking the fast path was overwriting that duration, effectively resurrecting a runner that was supposed to go away — and leaving the pending…
Together these two PRs point to the same theme: the scheduler's bookkeeping around runner lifecycle — loaded, evicting, dead — wasn't airtight under concurrency or process failure. Anyone debugging hangs or stuck models in recent builds should know these gaps are now closed.
Separately, PR 17514 fixes a scoring…
Nearby episodes from Ollama
- Speculative Decoding and the Cloud Model Nudge
- API Compatibility and Model Correctness Push
- Speculative Decoding Gains and a Lint Lockdown
- Cleaning Up Concurrency and Cutting Experimental Code
- Trust Your Cache, Fix Your Logs
- Parsing Bugs and Process Hardening
- Weekly Recap - New Model Support & Concurrency Hardening
- Closing the Gaps Between Local Config and Cloud Compatibility