error scope on a computational grid douglas thain university of wisconsin 4 march 2002
TRANSCRIPT
Error Scope on a Computational
GridDouglas Thain
University of Wisconsin4 March 2002
Overview
We have added a Java Universe to Condor. (More from Todd.)
Adding this code forced us to think about the fundamental problem of coupling systems and representing errors.
A lesson: One must consider the scope of an error as well as its detail.
Java for Scientific Computing
Java is emerging as a tool for large scale (Grande) scientific computing.• More accessible to domain scientists.• Simplified porting.• Faster development, debugging.
User communities are forming:• ACM Java Grande Conference• The Java Grande Forum
The Hype:
Java:•“Write once, run anywhere!”
Condor:•“Submit once, run everywhere!”
The Grid:•Uniform, dependable, consistent,
pervasive, and inexpensive computing.
The Reality:
Coupling systems is not trivial! The easy part:
• Putting java in front of the program name.
The tricky parts:• Dealing with unexpected events!
– Bad java installation.Bad java installation.– Unavailable file system.Unavailable file system.– Temporary resource exhaustion.Temporary resource exhaustion.
Architecture
Execution:• User just specifies “java” universe.• Execution site gives details of JVM.
I/O:• Know all of your files?
– Condor transfers whole files for you.Condor transfers whole files for you.
• Need online I/O?– Link program with Chirp I/O Library.Link program with Chirp I/O Library.– Execution site provides proxy to home site.Execution site provides proxy to home site.
startershadow
HomeFile
System
Execution SiteSubmission Site
startershadow
HomeFile
System
Execution SiteSubmission Site
JVM
Fork
startershadow
HomeFile
System
Execution SiteSubmission Site
JVM
Fork
The Job
JVM
Fork
startershadow
HomeFile
System
I/O Library
The Job
I/O Server I/O Proxy
Secure Remote I/O
Local System Calls Local I/O(Chirp)
Execution SiteSubmission Site
Initial Experience
Bad news: Nearly any unexpected failure would cause the job to be returned to the user:• Out of memory at execution site.• Java misconfigured at execution site.• I/O proxy can’t initialize.• Home file system offline.
What do Users Want?
This was correct in a certain sense:• The information was true.• But, still frustrating.
Users want to know when their program fails by design (NullPointerException,) but not if it fails due to the environment.
What Did We Do Wrong?
We thought that we were very careful to propagate errors:• I/O errors: server->proxy->library->job• JVM exit code: JVM->starter->home
But, we failed to draw a distinction:• Errors that are a natural property of the
program.• Errors that were an incidental result of
the environment.
Scope and Detail
The scope of an error is the portion of the system that it invalidates.
The detail of an error describes its philosophical cause.
An error must be delivered according to the handler that manages its scope.
Examples
Detail Scope Handler
Program exited normally. Program User
Null pointer exception. Program User
Out of memory. Remote Machine
Condor
Home file system offline. Home Machine
Condor
Program image corrupt. Job User
An Example
With this understanding, we reconsidered many elements of the Java Universe.
One example:• The JVM exit code is not a useful result.• It gives results that ignore error scope.
Solution:• Trap the program exit at a higher level.• Report the result and scope on a
separate channel.
JVM Exit Code
Detail Scope Handler Exit Code
Program exited normally.
Program User (x)
Null pointer exception. Program User 1
Out of memory. Remote Machine
Condor 1
Home file system offline.
Home Machine
Condor 1
Program image corrupt. Job User 1
JVM
startershadow
HomeFile
System
Wrapper
I/O Library
The Job
ResultFile
JVM Result
ProgramResult
orError and
Scope
Starter Result +Program Result
JVM
starter
shadow
HomeFile
System
Wrapper
I/O Library
The Job
ResultFile
JVM Result
I/O Proxy
Errors of Larger Scope
Errors InsideProgram Scope
Conclusion
We started building the Java Universe with some naive assumptions about errors.
On encountering practical difficulties, we thought more abstractly about errors and developed the notion of scope and detail.
By routing errors according to their scope, we made the system more robust and usable.
Details in an upcoming paper.
Deeper Problems Systems have deep semantic differences
that cross multiple functions. Consider this self-cleaning program:
• Open a file.• Delete the file.• Close the file.
Works on UNIX, fails on WinNT. Can we really provide a uniform interface?
More Info:
Demo on Wednesday Morning• Room 3381 CS anytime
The Condor Project:• http://www.cs.wisc.edu/condor
These slides:• http://www.cs.wisc.edu/~thain
Douglas Thain• [email protected]
Questions now?