[PVFS-developers] a bug of 1.6.2?

[songdm@lenovo.com]

Hi,all,

When I run a HA test on my cluster, I found my test program waiting for a
long time if I killed one of iods. But I run the test for the second time
without the iod restart, such a phenomenon didn't occured. I tried to trace
pvfsd, and it blocked bvrecv() (pvfs1.6.2/lib/mgrcomm.c: line 197). It's a
new function. It seems that this function will read mack and iods info from
socket. But if mgr return a error( ie.mack->status<0 ),it could only
recieve a mack. No iods info would be returned from mgr. So it will fall
into the "while" loop and can't return anything. Are my analysis correct ?
I have rewrote some code of send_mreq_saddr() according code in 1.5.8,
throwing off bvrecv(). Now it seemed worked.

Best regards,

Song Dongmei
E-mail:songdm@lenovo.com