Re: [HTCondor-users] Starter exited with status -1073740940

["John (TJ) Knoeller" <johnkn@xxxxxxxxxxx>]

That error code is 0xC0000374  which is STATUS_HEAP_CORRUPTION defined 
in ntstatus.h.

Logging of the priv change only happens after the change was successful 
so the crash is
whatever happens next after this line.

c:\condor\execute\dir_18128\userdir\src\condor_starter.v6.1\basestarter.cpp:1789

The next thing to happen is a
  mkdir the working directory
  write machine ad and job ad into the working directory
  set acls are on the working directory
  if (ENCRYPT_EXECUTE_DIRECTORY)  Load ADVAPI32.dll to get 
EncryptFile() function and use it to encrypt the working dir
  chdir to working dir
  dprintf( D_FULLDEBUG, "Done moving to directory \"%s\"\n", 
WorkingDir.Value() );

So since you aren't seeing "Done moving to directory...",  the problem 
must happen as we are setting up the working directory.

Can you tell how far into the process we got?
was the working directory made?
were ads written to it?
do you have encryption enabled?

None of this code ever calls exit, so the exit must be happening down 
inside some library.

-tj


On 7/10/2014 1:21 PM, Ben Cotton wrote:
> I'm running HTCondor 8.2.1 in a small cluster on AWS and I'm having a
> hard time getting my Windows jobs to run. The Windows execute node is
> Server 2k8 R2 (which HTCondor identifies as Windows 7). The job
> matches, appears to start, but then the condor_starter.exe dies. The
> StartLog records:
>
> ïïïïïïï07/10/14 17:48:47 condor_read() failed: recv(fd=1012) returned
> -1, errno = 10054 , reading 5 bytes from <127.0.0.1:50882>.
> 07/10/14 17:48:47 IO: Failed to read packet header
> 07/10/14 17:48:47 Closing job ClassAd update socket from starter.
> 07/10/14 17:48:47 Starter pid 336 exited with status -1073740940
>
>  From the StarterLog:
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_HOSTNAME) Daemon client (shadow)
> address determined: name: "ip-10-151-7-218.ec2.internal", pool:
> "NULL", alias: "NULL", addr: "<10.151.7.218:48140?noUDP>"
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_ALWAYS) Communicating with
> shadow <10.151.7.218:48140?noUDP>
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_ALWAYS) Submitting machine is
> "ip-10-151-7-218.ec2.internal"
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_SYSCALLS) Doing
> CONDOR_register_starter_info
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_NETWORK) condor_write(fd=604
> <10.151.7.218:59144>,,size=515,timeout=300,flags=0,non_blocking=0)
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_NETWORK) condor_read(fd=604
> <10.151.7.218:59144>,,size=5,timeout=300,flags=0,non_blocking=0)
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_NETWORK) condor_read(fd=604
> <10.151.7.218:59144>,,size=8,timeout=300,flags=0,non_blocking=0)
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_ALWAYS) setting the orig job
> name in starter
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_ALWAYS) setting the orig job iwd
> in starter
> 07/10/14 17:48:47 (fd:7) (pid:336) (D_PRIV) PRIV_CONDOR -->
> PRIV_CONDOR at c:\condor\execute\dir_18128\userdir\src\condor_starter.v6.1\basestarter.cpp:1789
>
> And then it goes poof. I see on the MagicNumbers page[1] that negative
> statuses might mean "Possibly missing libraries or missing functions
> in libraries on Windows. Try running from the command line to see if
> you get any errors." I tried running from the command line and got no
> output, error or otherwise. The other daemons seem to be fine. Any
> ideas what's going on here?
>
> [1] https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=MagicNumbers
>
>
> Thanks,
> BC