Memory error-talk
0 likes562 views
The document discusses methods for handling 'Out of Memory' errors, particularly in Eclipse, emphasizing the need for recovery techniques such as the 'hedge' technique, which involves pre-allocating memory to help manage low memory situations. It highlights the difficulties in implementing these strategies due to potential memory inconsistencies and compiler behaviors. The author suggests that while hedge recovery can be effective, improvements are necessary in language specifications and compiler behavior to ensure safer memory management.
![One Problem Leads To Another (2 of 3)
H So we pre-allocate some memory:
void performAction()
{
start();
int[] space = new int[1000];
// Point A
try {
doIt();
} finally {
// Point B
space = null;
cleanup();
}
}
H But what if the compiler . . .
– moves the allocation later (B)?
– moves the deallocation earlier (A)?
ECOOP EHWS Handling OutOfMemoryError 11](https://image.slidesharecdn.com/memory-error-talk-140714045514-phpapp01/85/Memory-error-talk-12-320.jpg)
![One Problem Leads To Another (3 of 3)
H Fake uses force early allocation.
H Fake tests force late deallocation.
void performAction()
{
start();
int[] space = new int[1000];
space[45] = 1+space[fact(6)];
try {
doIt();
} finally {
if (space[45] > space[44]) {
space = null;
cleanup();
}
}
}
H We have obfuscated our program.
ECOOP EHWS Handling OutOfMemoryError 12](https://image.slidesharecdn.com/memory-error-talk-140714045514-phpapp01/85/Memory-error-talk-13-320.jpg)
Memory error-talk
- 1. Handling “Out of Memory” Errors John Tang Boyland ECOOP EHWS, July 25, 2005
- 2. Handling “Out Of Memory” Errors H Current advice for handling OutOfMemoryError – Use to find memory limits. – Don’t! There’s nothing you can do. – . . . but Eclipse does! H The “Hedge” technique. – allocate a large hedge; – free when recovering from low memory condition. H Desiderata – Language Specification: What is safe if memory is low? – Compilers: Don’t move allocation later or deallocation earlier. – Runtime: per-thread memory restrictions. ECOOP EHWS Handling OutOfMemoryError 1
- 3. An OutOfMemoryError occurs ? Exception thrown here H Recovery difficult because of “low memory” condition. H When exception is thrown, last request is not fulfilled. ECOOP EHWS Handling OutOfMemoryError 2
- 4. Measuring Available Memory Exception thrown and caught Object allocated and discarded H A loop: – Try allocating a HUGE array; – Catch the exception and try again with a smaller amount. – Repeat until no exception is thrown. H A rough underestimate of available memory. (More accurate than Runtime.freeMemory().) ECOOP EHWS Handling OutOfMemoryError 3
- 5. Reasoning About OutOfMemoryError H Could occur at any time: – even in code “proven” not to raise an exception; – OutOfMemoryError is a subclass of Error, (in principle) “unpredictable” and “unpreventable” errors. H . . . well, almost any time: – if memory needed (allocation, boxing, concatenation); – if stack needed (call, local var. frame); – if exception created (NPE, ArrayStoreException, etc). H Typical advice: don’t try to handle it. H Alternate advice: use soft/weak references. ECOOP EHWS Handling OutOfMemoryError 4
- 6. A “Real” Program Must Handle the Error H Almost no realistic program can provably avoid running out of memory. H For example: Eclipse – uses more memory if more files are being edited; – memory is used by many different parts (GUI, Compiler, as- sistance, markers etc) H Crashing on OOME is unacceptable: – user’s work is lost, and – workbench left (perhaps) in inconsistent state, but – logging errors or saving files taken memory; H The error must be handled. ECOOP EHWS Handling OutOfMemoryError 5
- 7. Handling OutOfMemoryError in Eclipse (1 of 2) H Eclipse catches OOME and displays warning dialog: – but memory is low; – dialog appears after emergency exit fails; – otherwise only error messages on Unix stdout. H Eclipse 3.1 uses a larger max heap size than previously – Normally degradation (thrashing) long precedes OOME. – Artificially lowering the heap size gets previous behavior. ECOOP EHWS Handling OutOfMemoryError 6
- 8. Handling OutOfMemoryError in Eclipse (2 of 2) Exception in thread "...JavaReconciler" java.lang.OutOfMemoryError Exception in thread "...JavaReconciler" java.lang.OutOfMemoryError Exception in thread "...JavaReconciler" java.lang.OutOfMemoryError Error while logging event loop exception: java.lang.OutOfMemoryError: Java heap space Logging exception: java.lang.OutOfMemoryError: Java heap space Error while informing user about event loop exception: java.lang.OutOfMemoryError: Java heap space Dialog open exception: java.lang.OutOfMemoryError: Java heap space Fatal error happened during workbench emergency close. java.lang.OutOfMemoryError: Java heap space Unhandled event loop exception Reason: Java heap space H Then dialog brought up. ECOOP EHWS Handling OutOfMemoryError 7
- 9. The “Hedge” Technique H Pre-allocate a large area (the “hedge”); H When OutOfMemoryError happens, release it; H After recovery re-allocate hedge. Recovery time Hedge ECOOP EHWS Handling OutOfMemoryError 8
- 10. Difficulties Using the Hedge Technique H Need to overestimate memory required for recovery; H Interrupted computation may leave data inconsistent; H finally clauses before recovery may re-throw OOME; H Error may be thrown in thread other than the “guilty” one; H Compiler may move allocation later or deallocation earlier; H Cannot be made automatic. (see next slides) ECOOP EHWS Handling OutOfMemoryError 9
- 11. One Problem Leads To Another (1 of 3) H To avoid corruption, we introduce “finally”: void performAction() { start(); doIt(); cleanup(); } © void performAction() { start(); try { doIt(); } finally { cleanup(); } } H But what if cleanup needs memory (heap/stack) ? ECOOP EHWS Handling OutOfMemoryError 10
- 12. One Problem Leads To Another (2 of 3) H So we pre-allocate some memory: void performAction() { start(); int[] space = new int[1000]; // Point A try { doIt(); } finally { // Point B space = null; cleanup(); } } H But what if the compiler . . . – moves the allocation later (B)? – moves the deallocation earlier (A)? ECOOP EHWS Handling OutOfMemoryError 11
- 13. One Problem Leads To Another (3 of 3) H Fake uses force early allocation. H Fake tests force late deallocation. void performAction() { start(); int[] space = new int[1000]; space[45] = 1+space[fact(6)]; try { doIt(); } finally { if (space[45] > space[44]) { space = null; cleanup(); } } } H We have obfuscated our program. ECOOP EHWS Handling OutOfMemoryError 12
- 14. Placing Hedge Recovery H At outer level + few code changes; + lock state clear; - work undone; H Close to allocation + recovery fast; - state unclear; H If automatic, then how is recovery invoked? – at error point, then re-entrancy problems; – elsewhere, then finally is still an issue. ECOOP EHWS Handling OutOfMemoryError 13
- 15. Experiences With Hedge Recovery H Importing Java Into Internal Representation: – Must persist in “eras”; – As few eras as possible; – No easy way to use weak/soft references; H Converted JDK 1.4.2 provided source – 4500 source files; – 12 hours; – 11 OutOfMemoryErrors generated; – (300 Mb max heap on Solaris x86). H Avoided threading issues (single-threaded code). ECOOP EHWS Handling OutOfMemoryError 14
- 16. Conclusions H Hedge recovery can work. Perhaps Eclipse could use it. H Hedge recovery would be safer if: – Language specified what operations need memory; – Compilers don’t move allocation/deallocation past try-finally boundaries; – Threads had own memory restrictions. H Thrashing is a good alternative for interactive programs. ECOOP EHWS Handling OutOfMemoryError 15