So the touch method will spin its wheels for anything up to a second before we've registered an actual 'touch'.
This isn't so bad as files aren't touch that often, but it's a lot more problematic in the case of SegmentInfos.write(), where a new file is created, then copied over the old one. The testDeleteReaderWriteConflict testcase fails due to this. The old file and the new have the same timestamp, and so the staleness check fails. How should this be fixed?
Also, won't the fix for FSDirectory mean there's a window where other threads could end up with bad info, if we're taking up to a second to touch a file?
On Tuesday, September 30, 2003, at 01:12 PM, Dmitry Serebrennikov wrote:
Sounds like the same error that I fixed in the RAMDirectory last week. I didn't think that FSDirectory would have this problem because I thought that the OS filesystem would take care of this. Apparently it does not.
Hani, please take a look at the RAMDirectory touchFile method (line 148). It's basically the same fix you proposed except that it has some commented out monitoring code and it uses Thread.sleep(0,1) to release the CPU while waiting. In my measurements I was seeing 5-10 loops with the Thread.sleep call (regardless of the actual sleep duration that I specified), and around 10000 loops without a sleep call. But of course I was on a different hardware/OS combo and probably a different VM.
Hani, will you be able to try a variant of this fix for the FSDirectory?
I'm currious to see what kinds of results will observed on OS X.
Thanks. Dmitry.
Hani Suleiman wrote:
Ok, I figured out what's going on.
I suspect this will only manifest itself on faster machines (I'm on a dual 1.42ghz G4)
What happens basically is that the the timer resolution on file.lastModified isn't accurate enough, so when a file is touched, it returns the same value for lastModified. This in turn causes the segment staleness checks in IndexReader to fail (around line 276, lastModified(directory)==segmentInfosAge), hence the testcase fails.
I believe this is a bug, but I can't come up with a satisfactory workaround. My solution is to modify FSDirectory.touchFile(String name) to have the following block;
long oldModified = file.lastModified(); do { file.setLastModified(System.currentTimeMillis()); } while(oldModified==file.lastModified());
This ensures that a touch ensures the file is considered 'fresher' that it was before, which makes everything happy. It's a bit of a hack though, so I'm not sure of how you guys would prefer handling this, take the performance hit with the extra lastModified checks, if make it an even bigger hack by checking for OSX, or something far more clever and not detrimental to performance ;)
Hani
On Tuesday, September 30, 2003, at 12:42 AM, Hani Suleiman wrote:
here you go, happens to me on linux too (with JDK 1.4.0)
OSX using latest 1.4.1
[junit] ------------- ---------------- ---------------
[junit] Testsuite: org.apache.lucene.index.TestIndexReader
[junit] Tests run: 5, Failures: 2, Errors: 0, Time elapsed: 2.883 sec
[junit] Testcase: testDeleteReaderWriterConflict(org.apache.lucene.index.TestIndexReade r) : FAILED
[junit] Delete allowed on an index reader with stale segment information
[junit] junit.framework.AssertionFailedError: Delete allowed on an index reader with stale segment information
[junit] at org.apache.lucene.index.TestIndexReader.testDeleteReaderWriterConflic t( TestIndexReader.java:271)
[junit] at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
[junit] at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl. ja va:39)
[junit] at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces so rImpl.java:25)
[junit] Testcase: testDeleteReaderReaderConflict(org.apache.lucene.index.TestIndexReade r) : FAILED
[junit] Delete allowed from a stale index reader
[junit] junit.framework.AssertionFailedError: Delete allowed from a stale index reader
[junit] at org.apache.lucene.index.TestIndexReader.testDeleteReaderReaderConflic t( TestIndexReader.java:397)
[junit] at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
[junit] at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl. ja va:39)
[junit] at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces so rImpl.java:25)
On Tuesday, September 30, 2003, at 12:37 AM, Erik Hatcher wrote:
Hani,
I don't see this test failure on OS X. Thanks to your patch on the RussianAnalyzer, all my tests now successfully pass. Could you give us the stack trace for the problem you're seeing?
Erik
On Monday, September 29, 2003, at 11:28 PM, Hani Suleiman wrote:
I'm seeing a fairly disturbing testcase failure on OSX in TestIndexReader.testDeleteReaderWriterConflict and TestIndexReader.testDeleteReaderReaderConflict
This only happens on OSX though, is this a known issue? Any have an ideas of what sort of thing is going wrong before I dig in further?
Hani
------------------------------------------------------------------- --
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
-------------------------------------------------------------------- -
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
