Automated performance debugging of production applications

Patent metadata

Field	Value
Publication number	US-10915425-B2
Application number	US-201616326217-A
Country	US
Kind code	B2
Filing date	Sep 9, 2016
Priority date	Sep 9, 2016
Publication date	Feb 9, 2021
Grant date	Feb 9, 2021

Abstract

Performance anomalies in production applications can be analyzed to determine the dynamic behavior over time of hosting processes on the same or different computers. Problematic call sites (call sites that are performance bottlenecks or that are causing hangs) can be identified. Instead of relying on static code analysis and development phase load testing to identify a performance bottleneck or application hang, a lightweight sampling strategy collects predicates representing key performance data in production scenarios. Performance predicates provide information about the subject (e.g., what the performance issue is, what caused the performance issue, etc.). The data can be fed into a model based on a decision tree to identify critical threads running the problematic call sites. The results along with the key performance data can be used to build a call graph prefix binary tree for analyzing call stack patterns. Data collection, analysis and visualizations of result can be performed.

First claim

What is claimed:

1. A computing device comprising:
a processor; and
a memory connected to the processor; the processor configured to:
provide a program module that collects key performance data for an application executing in a production environment using a lightweight sampling strategy;
using a trained machine learning model, analyze the key performance data to identify a particular critical thread of the application, wherein the trained machine learning model has been trained using training cases to designate threads as critical or normal based at least on a corresponding likelihood that the threads have call stack information relating to causes of performance anomalies;
analyze a call stack pattern of the particular critical thread; and
provide analysis results comprising debug information for the application executing in the production environment.

2. The computing device of claim 1 , wherein the key performance data comprises data from execution of the application.

3. The computing device of claim 1 , wherein the lightweight sampling strategy comprises injecting a data collection agent into a process executing the application.

4. The computing device of claim 1 , wherein the trained machine learning model comprises a decision tree model, and the processor is further configured to:
feed the key performance data into the decision tree model to identify a plurality of critical threads of the application, individual critical threads running call sites causing a performance slowdown of the application.

5. The computing device of claim 1 , wherein the processor is further configured to:
identify busy threads between any two points in time within a sampling duration; and
input the key performance data for the busy threads to the trained machine learning model.

6. The computing device of claim 1 , wherein the processor is further configured to:
perform a ranking of a plurality of busy threads of the application,
wherein the ranking of the plurality of busy threads is used by the trained machine learning model to distinguish the particular critical thread from a normal thread of the application.

7. The computing device of claim 1 , wherein the processor is further configured to:
determine respective call stack lengths of multiple threads of the application,
wherein respective call stack lengths are used by the trained machine learning model to distinguish the particular critical thread from a normal thread of the application.

8. The computing device of claim 1 , wherein the processor is further configured to:
determine whether multiple threads of the application include non-framework call sites,
wherein the presence or absence of non-framework call sites is used by the trained machine learning model to distinguish the particular critical thread from a normal thread of the application.

9. The computing device of claim 8 , wherein the particular critical thread starts and ends with framework call sites and has a non-framework call site in between the framework call sites.

10. The computing device of claim 1 , wherein the processor is further configured to:
determine whether multiple threads of the application include non-framework call sites in consecutive samples,
wherein the presence or absence of non-framework call sites in the consecutive samples is used by the trained machine learning model to distinguish the particular critical thread from a normal thread of the application.

11. A method comprising configuring a processor of a computing device to:
receive key performance data for automated debugging of a production application;
analyze the key performance data to identify a critical thread, the critical thread a plurality of call sites, at least one of the plurality of call sites causing a performance slowdown of the production application;
perform an analysis of the critical thread using a call graph prefix binary tree reflecting call stack patterns of the critical thread with respect to the plurality of call sites; and
provide analysis results, the analysis results reflecting the analysis of the critical thread.

12. The method of claim 11 , further comprising configuring the processor to:
identify busy threads between any two points in time within a sampling duration; and
select the critical thread based at least on a ranking of the busy threads.

13. The method of claim 11 , further comprising configuring the processor to:
identify the critical thread using a decision tree model trained to distinguish critical from non-critical threads.

14. The method of claim 13 , further comprising configuring the processor to:
train the decision tree model using an existing set of training cases.

15. The method of claim 11 , further comprising configuring the processor to:
collect the key performance data using a lightweight sampling technique comprising:
create a handle to a process in which the production application executes,
inject a thread into the process,
inject a data collection agent into the process to collect the key performance data.

16. A computing device comprising:
a processor; and
a memory;
the memory connected to the processor; the processor configured to:
receive key performance data for automated debugging of a production application;
analyze the key performance data to identify a critical thread having a plurality of call sites;
perform an analysis of the critical thread using a call graph tree reflecting call stack patterns of the critical thread with respect to the plurality of call sites, the call graph tree having nodes representing individual call sites of the critical thread; and
based at least on the analysis, provide analysis results identifying a particular call site of the production application associated with a performance slowdown.

17. The computing device of claim 16 , the processor further configured to:
identify the critical thread using a decision tree based model.

18. The computing device of claim 16 , the processor further configured to:
provide a CPU utilization view displaying a busy threads ranking.

19. The computing device of claim 16 , the processor further configured to:
provide a call stack pattern view that identifies a hot spot in the production application where the performance slowdown occurs.

20. The computing device of claim 16 , the processor further configured to:
output critical thread data for the critical thread.

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Wang Yawei

Classifications

• H04L43/022 Primary

  by sampling · CPC title

• G06F11/0778

  Dumping, i.e. gathering error/state information after a fault for later diagnosis · CPC title

• H04L43/00

  Arrangements for monitoring or testing data switching networks · CPC title

• H04L43/0817

  by checking functioning · CPC title

• H04L43/04

  Processing captured monitoring data, e.g. for logfile generation · CPC title