Tracking, Not Tracing, Linux Thread Activity For Complete System Visibility - Tanel Poder

  Рет қаралды 711

eBPF & Cilium Community

eBPF & Cilium Community

Күн бұрын

Пікірлер: 1
@wolpumba4099
@wolpumba4099 2 ай бұрын
*Tracking, Not Tracing, Linux Thread Activity for Complete System Visibility* * *0:10** Introduction:* Tanel Poder introduces the concept of Linux thread state tracking as a way to gain complete system visibility from a demand-side perspective, focusing on the activities of active threads. * *0:16** Demand-Side Visibility:* The goal is to understand what threads are doing and why they might be off-CPU, providing insight beyond traditional system utilization metrics. * *0:47** Process Snapper & Fact Table Analogy:* Tanel highlights his older tool, Process Snapper, which samples proc files to show thread activity. This data can be considered a fact table in a data warehouse, allowing analysis of thread behavior over time. * *1:09** Xtop & Extended Task State Array:* Xtop, a newer eBPF-based tool, is introduced. It utilizes an "extended task state array" to store custom thread metadata, including application-level context like system calls and stack traces. * *3:41** Decoupled Layers:* The system employs two decoupled layers: eBPF programs that maintain the extended task state array by updating thread state information, and a sampling layer that periodically captures snapshots of this state. * *4:50** Tracking vs. Tracing:* The approach emphasizes "tracking" over "tracing." Instead of logging every event, only the latest state and relevant context are stored, providing an overview without the overhead of full tracing. * *6:34** Benefits of Tracking:* This method offers a balance between system-wide visibility and the ability to drill down into individual threads and their off-CPU behavior, as demonstrated with examples of MySQL threads performing fsync operations. * *7:32** Scheduler Hooking & Latency Analysis:* Tanel discusses the potential for hooking into scheduler events to track thread wake-up relationships and analyze latencies between different components, such as application threads and database or file system operations. * *8:57** Future Plans & Call for Contributions:* Future plans include incorporating distributed tracing IDs and building a V3 of the tool using modern libbpf. A call for contributions is made to help develop the open-source project further. * *9:38** Conclusion:* Tanel encourages viewers to try out the tools, provide feedback, and reach out with any questions. I used gemini-1.5-pro-exp-0801 on rocketrecap dot com to summarize the transcript. Cost (if I didn't use the free tier): $0.06 Input tokens: 14677 Output tokens: 486
Continuous Profiling With Open Telemetry - Florian Lehner
6:11
eBPF & Cilium Community
Рет қаралды 314
eCHO Episode 155: Demystifying the CNI by Writing One From Scratch
49:02
eBPF & Cilium Community
Рет қаралды 400
Quilt Challenge, No Skills, Just Luck#Funnyfamily #Partygames #Funny
00:32
Family Games Media
Рет қаралды 25 МЛН
Long Nails 💅🏻 #shorts
00:50
Mr DegrEE
Рет қаралды 18 МЛН
bpftime: Userspace eBPF Runtime for Network and Observability - Yusheng Zheng
9:48
eCHO Episode 158: Complete view into system performance with xcapture-ebpf
1:02:05
Why Your Computer Has Probably Not Been Hacked
8:19
Ask Leo!
Рет қаралды 340
Cluster Mesh
7:18
Isovalent
Рет қаралды 3,9 М.
Securing Containers on Embedded Platforms Using eBPF - Reingard Kugler
10:25
eBPF & Cilium Community
Рет қаралды 124
MNAR Selection Modeling in Blimp
18:40
Blimp Video Series
Рет қаралды 26
eCHO Episode 161: We watched all the Cilium KubeCon Sessions so you don't have to
1:05:38
Quilt Challenge, No Skills, Just Luck#Funnyfamily #Partygames #Funny
00:32
Family Games Media
Рет қаралды 25 МЛН