RFR JDK-8170089: nsk/jdi/EventSet/resume/resume008: ERROR: suspendCounts don't match for : Common-Cleaner
Gary Adams
gary.adams at oracle.com
Wed Aug 22 17:05:16 UTC 2018
On 8/6/18, 3:16 PM, Chris Plummer wrote:
> On 8/6/18 11:41 AM, Gary Adams wrote:
>> On 8/6/18, 1:56 PM, Chris Plummer wrote:
>>> On 8/6/18 4:16 AM, Gary Adams wrote:
>>>> On 8/3/18, 6:38 PM, Chris Plummer wrote:
>>>>> Hi Gary,
>>>>>
>>>>> Overall it looks good.
>>>>>
>>>>> Is the EventHandler.isDisconnected() check needed?
>>>> This just follows the pattern used in other calls to setValue.
>>> I'm not seeing any other examples of this. Can you point me to them?
>>> Isn't it expected that you will always be connected, and it will
>>> only be disconnected if there is something very wrong with the
>>> execution of the test? Not producing an error in that case could
>>> actually be misleading, causing the test to fail with some sort of
>>> state related error rather than some sort of exception indicating it
>>> was disconnected.
>> The best examples of checking EventHandler.isDisconnected()
>> can be seen in the implementation of shouldRunAfterBreakPoint()
>> See test/hotspot/jtreg/vmTestbase/nsk/share/jdi/TestDebuggerType1.java
>>
>> It's used in the loop waiting for the breakpoint event to be observed,
>> and in the getValue() fetching of the next "instruction" indicating
>> testing is completed.
> Well, that's just 2 uses of isDisconnected() out of the 200+
> get/setValue() calls. I can see its use in the loop, since it is used
> to force the exit of the loop when disconnected (rather than waiting
> for timeout). The one before the getValue() call is more like your
> use, and I don't see the need in this case either. What's to prevent
> becoming disconnected between the isDisconnected() and the
> get/setValue() call?
Just following up on this loose end after vacation ...
I agree that there is nothing preventing the connection being terminated
between the time isDisconnected() is checked and the call to setValue()
being made. I also don't see any harm in including the isDisconnected()
check here. If you think the test is improved by removing the check,
I'll make those changes, post a fresh webrev and repeat the testing.
>
> Chris
>>>
>>> Chris
>>>> No point in attempting the operation, if you know the
>>>> connection was lost. An exception at this point could
>>>> be misleading, if some other error has already occurred.
>>>>>
>>>>> In resume008a.java you removed a lot of empty lines. In some
>>>>> places it's fine, but the lines at 132 and 134 should have
>>>>> remained. Also, for the ones that were ok to remove, I don't see
>>>>> you doing the same thing in the other files. I think probably it's
>>>>> best to be consistent, and for this webrev probably best not to do
>>>>> them since it distracts too much from the important changes.
>>>> The original bug was reported against resume008, so more time was
>>>> spent in that
>>>> particular test, including some line wrapping changes. I will
>>>> restore the blank lines
>>>> you mentioned before producing a final patch. The other tests had
>>>> observed failures
>>>> also during testing. Applying the same change fixed those failures
>>>> as well.
>>>>>
>>>>> Seems like there is a lot of abstraction that could have been done
>>>>> with these tests to share a lot of code, but since so far that
>>>>> hasn't been done, probably not a good idea to start doing that
>>>>> with this fix. Do you think it's worth filing an enhancement
>>>>> request for?
>>>> Reformatting or refactoring these older tests would be at best a P5.
>>>> I don't think it's worth filing a bug, but as we fix bugs in these
>>>> test it's
>>>> worth some minimal amount of cleanup while we are in the code.
>>>>>
>>>>> thanks,
>>>>>
>>>>> Chris
>>>>>
>>>>> On 8/3/18 11:04 AM, Gary Adams wrote:
>>>>>> Here is an updated webrev with the alternate solution implemented
>>>>>> for
>>>>>> resume 1 to 10. The debugger sets testCase variable in the debuggee
>>>>>> when each test case completes in the debugger. By having the
>>>>>> debuggee
>>>>>> wait for the debugger to complete with test case 0, it avoids the
>>>>>> interference
>>>>>> that occurs by proceeding to the breakpoint set in
>>>>>> MethodForCommunication
>>>>>> before the debugger has compared expected suspend counts.
>>>>>>
>>>>>> Webrev:
>>>>>> http://cr.openjdk.java.net/~gadams/8170089/webrev.01/index.html
>>>>>>
>>>>>> On 7/17/18, 11:33 AM, Gary Adams wrote:
>>>>>>> A race condition exists between the debugger and the debuggee.
>>>>>>>
>>>>>>> The first test thread is started with SUSPEND_NONE policy set.
>>>>>>> While processing the thread start event the debugger captures
>>>>>>> an initial set of thread suspend counts and resumes the
>>>>>>> debuggee vm. If the debuggee advances quickly it reaches
>>>>>>> the breakpoint set for methodForCommunication. Since the breakpoint
>>>>>>> carries with it SUSPEND_ALL policy, when the debugger captures a
>>>>>>> second
>>>>>>> set of suspend counts, it will not match the expected counts for
>>>>>>> a SUSPEND_NONE scenario.
>>>>>>>
>>>>>>> The proposed fix introduces a yield in the debuggee test thread
>>>>>>> run method
>>>>>>> to allow the debugger to get the expected sampled values.
>>>>>>>
>>>>>>> Issue: https://bugs.openjdk.java.net/browse/JDK-8170089
>>>>>>> Webrev: http://cr.openjdk.java.net/~gadams/8170089/webrev.00/
>>>>>>>
>>>>>>>
>>>>>>> test/hotspot/jtreg/vmTestbase/nsk/share/jdi/TestDebuggerType1.java:
>>>>>>> ...
>>>>>>> 186 private void
>>>>>>> setCommunicationBreakpoint(ReferenceType refType, String
>>>>>>> methodName) {
>>>>>>> 187 Method method = debuggee.methodByName(refType,
>>>>>>> methodName);
>>>>>>> 188 Location location = null;
>>>>>>> 189 try {
>>>>>>> 190 location = method.allLineLocations().get(0);
>>>>>>> 191 } catch (AbsentInformationException e) {
>>>>>>> 192 throw new Failure(e);
>>>>>>> 193 }
>>>>>>> 194 bpRequest = debuggee.makeBreakpoint(location);
>>>>>>> 195
>>>>>>>
>>>>>>> 196 bpRequest.setSuspendPolicy(EventRequest.SUSPEND_ALL);
>>>>>>>
>>>>>>> 197 bpRequest.putProperty("number", "zero");
>>>>>>> 198 bpRequest.enable();
>>>>>>> 199
>>>>>>> 200 eventHandler.addListener(
>>>>>>> 201 new EventHandler.EventListener() {
>>>>>>> 202 public boolean eventReceived(Event
>>>>>>> event) {
>>>>>>> 203 if (event instanceof
>>>>>>> BreakpointEvent && bpRequest.equals(event.request())) {
>>>>>>> 204 synchronized(eventHandler) {
>>>>>>> 205 display("Received
>>>>>>> communication breakpoint event.");
>>>>>>> 206 bpCount++;
>>>>>>> 207 eventHandler.notifyAll();
>>>>>>> 208 }
>>>>>>> 209 return true;
>>>>>>> 210 }
>>>>>>> 211 return false;
>>>>>>> 212 }
>>>>>>> 213 }
>>>>>>> 214 );
>>>>>>> 215 }
>>>>>>>
>>>>>>>
>>>>>>> test/hotspot/jtreg/vmTestbase/nsk/jdi/EventSet/resume/resume008.java:
>>>>>>>
>>>>>>> ...
>>>>>>> 140 display("......--> vm.suspend();");
>>>>>>> 141 vm.suspend();
>>>>>>> 142
>>>>>>> 143 display(" getting : Map<String,
>>>>>>> Integer> suspendsCounts1");
>>>>>>> 144
>>>>>>> 145 Map<String, Integer> suspendsCounts1 =
>>>>>>> new HashMap<String, Integer>();
>>>>>>> 146 for (ThreadReference threadReference :
>>>>>>> vm.allThreads()) {
>>>>>>> 147 suspendsCounts1.put(threadReference.name(),
>>>>>>> threadReference.suspendCount());
>>>>>>> 148 }
>>>>>>> 149 display(suspendsCounts1.toString());
>>>>>>> 150
>>>>>>> 151 display(" eventSet.resume;");
>>>>>>> 152 eventSet.resume();
>>>>>>> 153
>>>>>>> 154 display(" getting : Map<String,
>>>>>>> Integer> suspendsCounts2");
>>>>>>>
>>>>>>> This is where the breakpoint is encountered before the second
>>>>>>> set of suspend counts is acquired.
>>>>>>>
>>>>>>> 155 Map<String, Integer> suspendsCounts2 =
>>>>>>> new HashMap<String, Integer>();
>>>>>>> 156 for (ThreadReference threadReference :
>>>>>>> vm.allThreads()) {
>>>>>>> 157 suspendsCounts2.put(threadReference.name(),
>>>>>>> threadReference.suspendCount());
>>>>>>> 158 }
>>>>>>> 159 display(suspendsCounts2.toString());
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
More information about the serviceability-dev
mailing list