The iPlanet Portal Server 3.0 Gateway is a heavily multithreaded TCP/IP server Java application that is delivered on Solaris 2.6, 7, and 8. In our latest service pack, we need to upgrade the Solaris production JVM used from 1.2.2_09 to at least 1.2.2_15. However in doing so, the process quickly hangs under a load test.
During the hang, a pstack output shows that the JVM is in the process of garbage collection:
----------------- lwp# 5 / thread# 5 --------------------
fef9c07c lwp_cond_wait (ff356080, ff356090, fecf1a38)
fef93604 _lwp_cond_reltimedwait (ff356080, ff356090, fecf1a38, 0, 0, fecf1a30) + c
fef936d4 _lwp_cond_timedwait (2f98ad58, 3f05b8f9, fecf1aa8, ff356090, ff356080, fecf1aa0) + 98
ff247860 sysCondvarTimedWaitGCOnly (ff356080, 2f98ad58, 0, ff356090, 3b9aca00, 0) + 194
ff222bc0 suspendAllMutators (1, ff356108, ff3560f0, ff356090, ff33f688, ff351380) + 234
ff191150 stopWorldAndDoCurThread (1, ff195038, e2d413dc, ff35b000, e2d413dc, 2740d) + 180
ff19164c GCThread_run (ff3515c8, ff3515b8, ff33e464, 1, 1b7e0, 1fd70) + 60
ff245724 _start (0, ff351380, ff33fc00, fec, ff350000, ff35b408) + 23c
ff00b730 _thread_start (ff351380, 0, 0, 0, 0, 0) + 40
However, the GC never proceeds beyond this point.
This problem has been reproduced using both 1.2.2_15 and 1.2.2_16 (the latest at this time). Migration to a 1.3 or 1.4 revision JVM is not an option at this time due to the unacceptable QA overhead that would be incurred by the necessary recertification of the product.
During the hang, a pstack output shows that the JVM is in the process of garbage collection:
----------------- lwp# 5 / thread# 5 --------------------
fef9c07c lwp_cond_wait (ff356080, ff356090, fecf1a38)
fef93604 _lwp_cond_reltimedwait (ff356080, ff356090, fecf1a38, 0, 0, fecf1a30) + c
fef936d4 _lwp_cond_timedwait (2f98ad58, 3f05b8f9, fecf1aa8, ff356090, ff356080, fecf1aa0) + 98
ff247860 sysCondvarTimedWaitGCOnly (ff356080, 2f98ad58, 0, ff356090, 3b9aca00, 0) + 194
ff222bc0 suspendAllMutators (1, ff356108, ff3560f0, ff356090, ff33f688, ff351380) + 234
ff191150 stopWorldAndDoCurThread (1, ff195038, e2d413dc, ff35b000, e2d413dc, 2740d) + 180
ff19164c GCThread_run (ff3515c8, ff3515b8, ff33e464, 1, 1b7e0, 1fd70) + 60
ff245724 _start (0, ff351380, ff33fc00, fec, ff350000, ff35b408) + 23c
ff00b730 _thread_start (ff351380, 0, 0, 0, 0, 0) + 40
However, the GC never proceeds beyond this point.
This problem has been reproduced using both 1.2.2_15 and 1.2.2_16 (the latest at this time). Migration to a 1.3 or 1.4 revision JVM is not an option at this time due to the unacceptable QA overhead that would be incurred by the necessary recertification of the product.
- relates to
-
JDK-4881610 JVM Panic on 1.2.2_15
- Closed