When you have ever stared at hundreds of traces of integration check logs questioning which of the sixteen log recordsdata truly comprises your bug, you aren’t alone — and Google now has information to show it.
A group of Google researchers launched Auto-Diagnose, an LLM-powered instrument that robotically reads the failure logs from a damaged integration check, finds the basis trigger, and posts a concise prognosis immediately into the code assessment the place the failure confirmed up. On a handbook analysis of 71 real-world failures spanning 39 distinct groups, the instrument accurately recognized the basis trigger 90.14% of the time. It has run on 52,635 distinct failing exams throughout 224,782 executions on 91,130 code modifications authored by 22,962 distinct builders, with a ‘Not useful’ fee of simply 5.8% on the suggestions acquired.
The issue: integration exams are a debugging tax
Integration exams confirm that a number of elements of a distributed system truly talk to one another accurately. The exams Auto-Diagnose targets are airtight practical integration exams: exams the place a complete system underneath check (SUT) — usually a graph of speaking servers — is introduced up inside an remoted setting by a check driver, and exercised towards enterprise logic. A separate Google survey of 239 respondents discovered that 78% of integration exams at Google are practical, which is what motivated the scope.
Diagnosing integration check failures confirmed up as one of many prime 5 complaints in EngSat, a Google-wide survey of 6,059 builders. A follow-up survey of 116 builders discovered that 38.4% of integration check failures take greater than an hour to diagnose, and eight.9% take greater than a day — versus 2.7% and 0% for unit exams.
The foundation trigger is structural. Check driver logs normally floor solely a generic symptom (a timeout, an assertion). The precise error lives someplace inside one of many SUT part logs, typically buried underneath recoverable warnings and ERROR-level traces that aren’t truly the trigger.
How Auto-Diagnose works
When an integration check fails, a pub/sub occasion triggers Auto-Diagnose. The system collects all check driver and SUT part logs at degree INFO and above — throughout information facilities, processes, and threads — then joins and kinds them by timestamp right into a single log stream. That stream is dropped right into a immediate template together with part metadata.
The mannequin is Gemini 2.5 Flash, referred to as with temperature = 0.1 (for near-deterministic, debuggable outputs) and primep = 0.8. Gemini was not fine-tuned on Google’s integration check information; that is pure immediate engineering on a general-purpose mannequin.
The immediate itself is essentially the most instructive a part of this analysis. It walks the mannequin by an express step-by-step protocol: scan log sections, learn part context, find the failure, summarize errors, and solely then try a conclusion. Critically, it consists of arduous adverse constraints — for instance: if the logs don’t comprise traces from the part that failed, don’t draw any conclusion.
The mannequin’s response is post-processed right into a markdown discovering with ==Conclusion==, ==Investigation Steps==, and ==Most Related Log Traces== sections, then posted as a remark in Critique, Google’s inside code assessment system. Every cited log line is rendered as a clickable hyperlink.
Numbers from manufacturing
Auto-Diagnose averages 110,617 enter tokens and 5,962 output tokens per execution, and posts findings with a p50 latency of 56 seconds and p90 of 346 seconds — quick sufficient that builders see the prognosis earlier than they’ve switched contexts.
Critique exposes three suggestions buttons on a discovering: Please repair (utilized by reviewers), Useful, and Not useful (each utilized by authors). Throughout 517 whole suggestions experiences from 437 distinct builders, 436 (84.3%) have been “Please repair” from 370 reviewers — by far the dominant interplay, and an indication that reviewers are actively asking authors to behave on the diagnoses. Amongst dev-side suggestions, the helpfulness ratio (H / (H + N)) is 62.96%, and the “Not useful” fee (N / (PF + H + N)) is 5.8% — effectively underneath Google’s 10% threshold for preserving a instrument stay. Throughout 370 instruments that publish findings to Critique, Auto-Diagnose ranks #14 in helpfulness, placing it within the prime 3.78%.
The handbook analysis additionally surfaced a helpful facet impact. Of the seven instances the place Auto-Diagnose failed, 4 have been as a result of check driver logs weren’t correctly saved on crash, and three have been as a result of SUT part logs weren’t saved when the part crashed — each actual infrastructure bugs, reported again to the related groups. In manufacturing, round 20 ‘extra data is required‘ diagnoses have equally helped floor infrastructure points.
Key Takeaways
- Auto-Diagnose hit 90.14% root-cause accuracy on a handbook analysis of 71 real-world integration check failures spanning 39 groups at Google, addressing an issue 6,059 builders ranked amongst their prime 5 complaints within the EngSat survey.
- The system runs on Gemini 2.5 Flash with no fine-tuning — simply immediate engineering. A pub/sub set off collects logs throughout information facilities and processes, joins them by timestamp, and sends them to the mannequin at temperature 0.1 and primep 0.8.
- The immediate is engineered to refuse fairly than guess. Onerous adverse constraints drive the mannequin to reply with “extra data is required” when proof is lacking — a deliberate trade-off that forestalls hallucinated root causes and even helped floor actual infrastructure bugs in Google’s logging pipeline.
- In manufacturing since Could 2025, Auto-Diagnose has run on 52,635 distinct failing exams throughout 224,782 executions on 91,130 code modifications from 22,962 builders, posting findings in a p50 of 56 seconds — quick sufficient that engineers see the prognosis earlier than switching contexts.
Try the Pre-Print Paper right here. Additionally, be at liberty to comply with us on Twitter and don’t overlook to hitch our 130k+ ML SubReddit and Subscribe to our E-newsletter. Wait! are you on telegram? now you’ll be able to be part of us on telegram as effectively.
Must associate with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and so on.? Join with us
