rasmgr segfaults when restarting rasserver

[Dimitar Misev]

This is bug probably introduced by the p2p patch changeset:90fa2688a862f227e7f52b0423041c185f1eed34

Seems to happen when the countdown/timeout is run out and a rasserver is to be restarted. To test run source:systemtest/testcases_petascope/test_wcps it will fail around the 48/49th test if the rasmgr.conf is with default countdown parameters.

Workaround: set countdown and timeout in rasmgr.conf to some large numbers.

Segfault stack trace:

Program received signal SIGSEGV, Segmentation fault.
0x0000000000423811 in MasterComm::askOutpeer (this=0x6c1b80 <masterCommunicator>, peer=1,
    outmsg=0x7fff64b70a40 "POST peerrequest HTTP/1.1\r\nAccept: text/plain\r\nUserAgent: RasMGR/1.0\r\nAuthorization: ras  ras rasguest:8e70a429be359b6dace8b5b2500dedb0\r\nContent-length: 47\r\n\r\nhifi RASBASE RNP ro 2130706689:1369740188"...) at rasmgr_master_nb.cc:877
877        struct hostent *hostinfo = gethostbyname(config.outpeers[peer]);
(gdb) bt
#0  0x0000000000423811 in MasterComm::askOutpeer (this=0x6c1b80 <masterCommunicator>, peer=1,
    outmsg=0x7fff64b70a40 "POST peerrequest HTTP/1.1\r\nAccept: text/plain\r\nUserAgent: RasMGR/1.0\r\nAuthorization: ras  ras rasguest:8e70a429be359b6dace8b5b2500dedb0\r\nContent-length: 47\r\n\r\nhifi RASBASE RNP ro 2130706689:1369740188"...) at rasmgr_master_nb.cc:877
#1  0x00000000004231fd in MasterComm::getFreeServer (this=0x6c1b80 <masterCommunicator>, fake=false, frompeer=false) at rasmgr_master_nb.cc:802
#2  0x000000000042178f in MasterComm::processRequest (this=0x6c1b80 <masterCommunicator>, currentJob=...) at rasmgr_master_nb.cc:482
#3  0x00000000004206e2 in MasterComm::processJob (this=0x6c1b80 <masterCommunicator>, currentJob=...) at rasmgr_master_nb.cc:214
#4  0x00000000004200ce in MasterComm::Run (this=0x6c1b80 <masterCommunicator>) at rasmgr_master_nb.cc:165
#5  0x000000000040d066 in main (argc=1, argv=0x7fff64b71548, envp=0x7fff64b71558) at rasmgr_main.cc:172 

[ungarj]

FYI, ticket #395 seems to be affected by this issue as well.

[mrusu]

Yes, hopefully the fix for this ticket solves the other problem as well, as they seem to be both caused by the same bug in rasmgr's new logic.