On Tue, Aug 07, 2012 at 03:01:56AM -0700, Anand Avati wrote: > I cannot find a code path where synctask_get() happens without a > synctask_set() happening in the same thread. Did you manage to find one? Here is the kind of trace I can get in rebalance.log, after the first synctask_new invocation (self is pthread_self() return value). [syncop.c:216:synctask_new] 0-: self = 0xb8000000, newtask = 0xb7f02400 [syncop.c:291:synctask_switchto] 0-: self = 0xb9e00000, synctask_set(0xb7f02400) [syncop.c:120:synctask_wrap] 0-: self = 0xb8000000, synctask_get() = 0x0, old_task = 0xb7f02400 And here of course it crashe, the reason is obvious. I can work it around by using old_task for the tésk in :synctask_wrap(), but I get other nasty crashes around swapcontext(), which suggests the work around does not fix the root cause. By the way, why use swapcontext()? There was no simplier way of doing it? -- Emmanuel Dreyfus manu@xxxxxxxxxx
