https://issues.apache.org/bugzilla/show_bug.cgi?id=48760
--- Comment #11 from Stefan Kendall <arcanef...@gmail.com> 2010-02-18 20:02:19 UTC --- gfuser9...@yahoo.co.uk has presented a reasoned explanation of what is happening below. Apparently, this is a problem that needs to be backported from Glassfish. I've copied the e-mail below: Now, really the REAL bug fix in Tomcat you need to port back from Glassfish is actually this http://fisheye5.cenqua.com/changelog/glassfish?cs=SJSAS91_FCS_BRANCH%3Asultal%3A20080904082511 This is the actual root cause for 45601 which is not updated back to Tomcat bug entry in the late 2008 but this is the RCA and is fixed in SUN Glassfish Application server (where they share the same common TC code). -------------------------- Here's the salient point: -------------------------- 0. You notice the issue happens in DefaultServlet copyRange as all your submitters submit - The code all do something like this InputStream resourceInputStream = cacheEntry.resource.streamContent(); 1. So they all take a FileResource/Resource from a cache entry (which is looked up from some cache) 2. Now We know that this is a FileResource (static resource) 3. Finally most of the code will have this reader = new InputStreamReader(resourceInputStream, fileEncoding); copyRange(reader,....) <-- do something reader.close(); 4. Now, you notice that reader.close() will CLOSE the underlying FileInputStream when it is run. 5. Hence, there is a SUBTLE race in FileResource.streamContent where a TWO or more threads get the SAME InputStream. This is due to the FACT that multiple requests can get hold of the same cached entry and hence due to the BAD way the FileResource streamContent stuff the InputStream to a temp instance variable.. Due to MT scheduling (it is possible to get a MT schedule where two threads gets the some common InputStream.) [Please see TC code to see why this can happen] Testing? ------------ It is easy to ARTIFICIALLY reproduce & to confirm You can slow things down in the FileResource.streamContent() by putting delays and etc (or also printing the resulting InputStream in the copy while doing two or more parallel wget for the same FileResource of filesize of some large MB to see. -- Configure bugmail: https://issues.apache.org/bugzilla/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
