{ "schema_version": "0.6.0", "paper_id": "an-adaptive-logging-system-als-enhancing-software-logging-with-reinforcement-learning-techniques", "page_url": "https://naser.github.io/research-publications/papers/an-adaptive-logging-system-als-enhancing-software-logging-with-reinforcement-learning-techniques/", "title": "An Adaptive Logging System (ALS): Enhancing Software Logging with Reinforcement Learning Techniques", "title_variants": [], "authors": [ "Amirmahdi Khosravi Tabrizi", "Naser Ezzati-Jivan", "Francois Tetreault" ], "author_details": [ { "name": "Amirmahdi Khosravi Tabrizi", "orcid": "https://orcid.org/0000-0002-6498-9319", "profile_url": "https://dblp.org/pid/344/3160.html" }, { "name": "Naser Ezzati-Jivan", "orcid": "https://orcid.org/0000-0003-1435-6297", "profile_url": "https://naser.github.io/" }, { "name": "Francois Tetreault", "orcid": "https://orcid.org/0000-0003-1975-1501", "profile_url": "https://dblp.org/pid/288/0256.html" } ], "publication": { "year": 2024, "venue": "ACM/SPEC International Conference on Performance Engineering (ICPE)", "type": "conference paper", "publication_date": "2024-05-07", "online_date": "2024-05-07", "print_date": "2024-05-07", "volume": null, "issue": null, "pages": "37-47", "article_number": null, "publisher": "ACM", "issn": [], "isbn": [], "crossref_type": "proceedings-article" }, "publication_type": "conference paper", "status": "published_with_public_full_text", "canonical_source_url": "https://doi.org/10.1145/3629526.3645033", "source_record_id": "an-adaptive-logging-system-als-enhancing-software-logging-with-reinforcement-learning-techniques-26544da898", "identifiers": { "doi": "10.1145/3629526.3645033" }, "abstract": null, "abstract_source": "ICPE proceedings PDF reviewed; abstract paraphrased for this catalog.", "abstract_available": false, "scholar_eligibility": { "eligible": false, "basis": "not-eligible", "note": "The page is a discovery record; it does not claim Google Scholar article-host eligibility." }, "description": "ALS uses source-code features and reinforcement learning to recommend which Python functions to log and which log levels to use for performance-bug diagnosis.", "evidence_level": "full-text-reviewed", "evidence": { "source_basis": "full-text-reviewed", "coverage": "material paper sections", "summary_origin": "AI-assisted catalog editorial summary", "review_status": "catalog-reviewed; paper-author approval pending", "verified_on": "2026-08-09", "sources": [ { "note": "ICPE ALS PDF: Python AST, GitHub/PyGithub/git, 12 static features, bug classes, Stable-Baselines, and DQN/A2C/PPO selection" }, { "note": "ICPE ALS PDF: 17 Apache/Django projects, 93,484 functions, 4,235 files, eight-episode evaluation, and reward stability findings" }, { "note": "ICPE ALS PDF: Python/dynamic-feature/agent-scope limitations and stated future work" }, { "note": "Local PDF hash verified in pdf-evidence/extraction-manifest.json" } ] }, "summary": { "core_contribution": "ALS uses source-code features and reinforcement learning to recommend which Python functions to log and which log levels to use for performance-bug diagnosis.", "problem": "Logging can provide essential postmortem evidence but adds code, storage, and runtime cost. Existing log-placement systems are tied to fixed objectives and show weak cross-project transfer; resource-constrained environments need adaptable performance-bug logging (pp. 1-3).", "method": "ALS extracts 12 static Python source features covering loops, synchronization issues, and Django ORM API misuses. Python AST FunctionDef nodes are collected from cloned GitHub repositories; five discrete actions select not-log, Trace, Debug, Info, or Warn. Error/Fatal are excluded because the policy uses static features. A deterministic RL environment uses rewards from -4 to +2; DQN, A2C, and PPO are evaluated through Stable-Baselines/OpenAI Baselines (pp. 3-8).", "findings": "The dataset contains 17 projects (11 Apache, 6 Django), 4,235 Python files, and 93,484 functions: 56,755 training functions/1,421 files and 36,729 testing functions/2,814 files (p. 7, Table 2). All three methods show positive reward trends by the end of the first of eight episodes; DQN is reported as the most stable in training and cross-project evaluation (pp. 8-9). No accuracy, logging-overhead, or production-quality number is reported.", "limitations": "Python-only static features; no dynamic CPU/memory terms; one agent handles both placement and level; function selection does not specify complete log content; project/algorithm coverage and reward design limit generalization. Production/Ciena deployment is future work, not completed evaluation.", "future_work": "Multi-language AST/LLM/Jina features, CPU/memory reward terms, dual-agent placement/level control, integration with LLM/RLHF and logging frameworks, and deployment/evaluation in Ciena's operational context (pp. 9-10)." }, "tags": [ "observability", "performance-engineering", "machine-learning", "trace-reduction" ], "keywords": [ "adaptive logging", "ALS", "reinforcement learning", "log placement", "log level selection", "performance bugs", "Python AST", "Stable-Baselines", "DQN", "A2C", "PPO", "Apache", "Django", "cross-project evaluation", "Ciena", "source-code features" ], "versions": [ { "id": "published-version", "label": "Published version", "relation": "version-of-record", "title": "An Adaptive Logging System (ALS): Enhancing Software Logging with Reinforcement Learning Techniques", "url": "https://doi.org/10.1145/3629526.3645033", "pdf_url": null, "status": "published", "canonical_for_citation": true }, { "id": "public-full-text", "label": "Public full text", "relation": "source-record", "title": "An Adaptive Logging System (ALS): Enhancing Software Logging with Reinforcement Learning Techniques", "url": "https://research.spec.org/icpe_proceedings/2024/proceedings/p37.pdf", "pdf_url": "https://research.spec.org/icpe_proceedings/2024/proceedings/p37.pdf", "status": "public_full_text", "canonical_for_citation": false } ], "access": { "status": "published_with_public_full_text", "note": "The DOI is the canonical citation target; the public full-text or author-manuscript link is external and the PDF is not redistributed here.", "license": null }, "resources": { "code": null, "data": null, "slides": null, "demo": null }, "citation_guidance": { "when_to_cite": "Cite this paper when your work uses or compares an RL formulation in which source-function observations select both whether to log and which verbosity level to use.", "points": [ "For an RL formulation in which source-function observations select both whether to log and which verbosity level to use.", "For the 12-feature PerformanceBugLoggingObjective covering loops, synchronization, and Django ORM misuse.", "For cross-project reward-curve evaluation of DQN, A2C, and PPO over 17 Apache/Django projects, with DQN's reported stability.", "Do not cite this paper for measured production logging overhead or production deployment; those remain future work." ], "canonical_version_id": "published-version" }, "provenance": { "metadata_verified_on": "2026-08-09", "metadata_source": [ "ICPE ALS PDF: Python AST, GitHub/PyGithub/git, 12 static features, bug classes, Stable-Baselines, and DQN/A2C/PPO selection", "ICPE ALS PDF: 17 Apache/Django projects, 93,484 functions, 4,235 files, eight-episode evaluation, and reward stability findings", "ICPE ALS PDF: Python/dynamic-feature/agent-scope limitations and stated future work", "Local PDF hash verified in pdf-evidence/extraction-manifest.json" ], "summary_written_by": "AI-assisted", "summary_verified_by": "full-text-grounded catalog review; author approval pending", "linked_preprint_record": null, "author_order_note": null }, "batch": { "phase": 2, "batch_label": "expanded forty-paper release", "status": "included_in_expanded_catalog", "selected_at": "2026-08-09" } }