Insight
From APIDesign
| Line 5: | Line 5: | ||
[[Insight]] is obviously great fit for tracing. The close co-operation with Pavol Loffay from Jaeger - the industry leading implementation of OpenTelemetry solution - has resulted in a medium post about [https://medium.com/jaegertracing/t-trace-agent-like-instrumentation-for-graalvm-d8b79b755e71 agent-like instrumentation for GraalVM]. | [[Insight]] is obviously great fit for tracing. The close co-operation with Pavol Loffay from Jaeger - the industry leading implementation of OpenTelemetry solution - has resulted in a medium post about [https://medium.com/jaegertracing/t-trace-agent-like-instrumentation-for-graalvm-d8b79b755e71 agent-like instrumentation for GraalVM]. | ||
| - | Instrumentation has always been part of [[Truffle]] system. However, prior to invention of [[Insight]] it was necessary to write your instruments in [[Java]]. Generating a [[Truffle]]-based language implementation compiled by [[native image]] may take a while. Recompiling everytime you want to change your instrument, slows you down significantly. That's not a problem for [[GraalVM]] [[Insight]] - all the hooks are written in dynamic | + | Instrumentation has always been part of [[Truffle]] system. However, prior to invention of [[Insight]] it was necessary to write your instruments in [[Java]]. Generating a [[Truffle]]-based language implementation compiled by [[native image]] may take a while. Recompiling everytime you want to change your instrument, slows you down significantly. That's not a problem for [[GraalVM]] [[Insight]] - all the hooks are written in dynamic languages and compiled on the fly into the appropriate user code locations. There is no limit on what these hooks can do - [[Insight]] are fully [[Turing complete]] code snippets capable to do everything the surrounding programming language can. |
A promising direction of research is usage of [[Insight]] as a harness for post-mortem debugging. Apply watchpoints dynamically, capture the state of execution whenever a program goes though them and replay the recorded stream in advanced tools like [[Chrome]] dev tools or [[NetBeans]]. | A promising direction of research is usage of [[Insight]] as a harness for post-mortem debugging. Apply watchpoints dynamically, capture the state of execution whenever a program goes though them and replay the recorded stream in advanced tools like [[Chrome]] dev tools or [[NetBeans]]. | ||
Revision as of 05:23, 11 July 2020
GraalVM Insight has been introduced in version 19.3 of GraalVM. It is a multipurpose, flexible tool that allows to dynamically apply complex insights gathering hooks on already deployed applications running at full speed. Insight further blurs the difference between various DevOps tasks - code once, apply your insights anytime, anywhere! See Insight.md.
Insight builds on the polyglot nature of GraalVM. One can write Insights in any language - e.g. JavaScript, Ruby, etc. and apply them to any other language including also C, C++, Fortran, etc. The capabilities of Insight are improved with every GraalVM release. Version 20.0 for example improved embedding of Insight into custom applications. Version 20.1 added access to local variables of C and other Sulong based programs. Version 20.2 allows to modify local variables and write insight scripts in Python.
Insight is obviously great fit for tracing. The close co-operation with Pavol Loffay from Jaeger - the industry leading implementation of OpenTelemetry solution - has resulted in a medium post about agent-like instrumentation for GraalVM.
Instrumentation has always been part of Truffle system. However, prior to invention of Insight it was necessary to write your instruments in Java. Generating a Truffle-based language implementation compiled by native image may take a while. Recompiling everytime you want to change your instrument, slows you down significantly. That's not a problem for GraalVM Insight - all the hooks are written in dynamic languages and compiled on the fly into the appropriate user code locations. There is no limit on what these hooks can do - Insight are fully Turing complete code snippets capable to do everything the surrounding programming language can.
A promising direction of research is usage of Insight as a harness for post-mortem debugging. Apply watchpoints dynamically, capture the state of execution whenever a program goes though them and replay the recorded stream in advanced tools like Chrome dev tools or NetBeans.
Follow
