LTTng, Linux’s powerful tracing toolkit, turns 20 but remains outside the mainline kernel. Despite adoption by IBM, Sony, and hyperscalers, Linus Torvalds rejects bulk upstreaming—demanding unified infrastructure. Explore why LTTng’s future hinges on incremental merging.
Two Decades of LTTng Adoption
The Linux Trace Toolkit Next Generation (LTTng) celebrates its 20th anniversary this year, having become a mission-critical tracing solution for hyperscalers, Fortune 500 companies, and Linux administrators.
Industry giants like IBM, Sony, and Siemens rely on LTTng for real-time system debugging, performance analysis, and kernel-level observability.
Despite its widespread adoption, LTTng’s kernel modules remain outside the mainline Linux kernel—even after four failed upstreaming attempts. A fifth push began this week, but early resistance from Linus Torvalds suggests history may repeat itself.
The Upstreaming Challenge: Why LTTng Struggles to Enter the Mainline Kernel
Past Attempts and Roadblocks
Four previous efforts to merge LTTng into the kernel stalled due to:
75,000+ lines of new code, requiring extensive review.
Debates over whether to integrate with Linux Perf or maintain independence.
Concerns about duplicating tracing infrastructure (LTTng vs. ftrace).
The Latest Proposal: Bulk Upstreaming vs. Incremental Merging
Mathieu Desnoyers, LTTng’s creator, recently floated a new RFC on the Linux Kernel Mailing List (LKML), proposing two paths:
Bulk Upstreaming – A single, large-scale merge to preserve user-space ABI compatibility and replace existing out-of-tree modules.
Staging Area Submission – Gradual refinement in the kernel’s staging tree before full integration.
However, Torvalds swiftly dismissed bulk merging, arguing:
*"If people haven’t unified tracing models in 20 years, merging LTTng won’t magically fix it. Show me incremental, shared infrastructure—or keep it out-of-tree."*
Why Linus Torvalds Rejects LTTng’s Upstreaming Push
Torvalds’ objections center on:
Avoiding redundant tracing systems (LTTng vs. ftrace).
No evidence of future unification between competing models.
Historical precedent of stalled collaboration.
His response was blunt:
"This isn’t a discussion. Either unify tracing incrementally, or stay out-of-tree. I don’t believe in fairy tales."
What’s Next for LTTng?
Short-Term Outlook
Corporate adoption won’t decline—LTTng remains essential for real-time tracing.
No near-term mainline inclusion unless proponents demonstrate incremental, unified progress.
Long-Term Possibilities
Collaboration with Perf subsystem could ease integration.
Staging tree refinement may improve acceptance.
Key Takeaways
✅ LTTng is industry-proven but remains out-of-tree.
⚠️ Torvalds demands unification—no exceptions.
🔮 Future upstreaming depends on incremental, shared infrastructure.
FAQ: LTTng and Linux Kernel Tracing
Q: Why hasn’t LTTng been merged yet?
A: Code size, duplication concerns, and lack of consensus on unified tracing.
Q: Can enterprises still use LTTng effectively?
A: Absolutely—many hyperscalers rely on out-of-tree modules without issue.
Q: What’s the best alternative to LTTng?
A: Ftrace and BPF are mainline options, but lack LTTng’s real-time capabilities.
Nenhum comentário:
Postar um comentário