CMS UNIX Signal Handlers

Complete:

Unix Signal Handler Service

The Unix signal handler service is automatically initialized early in any cmsRun job. There are no configuration parameters for this service so this note is primarily "For Your Information" except for one important detail for developers who may choose to define a new processing thread.

The POSIX standard states that if multiple threads indicate their willingness to accept delivery of some particular Unix signal, the result is undefined behavior: the signal will be delivered to some thread but it cannot be predicted a priori which one. As a consequence of this, we have adopted a policy to require that

• only the main thread - the one that includes the main program - will accept signal delivery, and,
• all other threads must refuse them.

Developers who define additional threads must set the signal mask to refuse signal delivery immediately after the new thread is created. There is a small set of free functions in FWCore/Utilities to manipulate signal masks. The interface for these is defined in FWCore/Utilities/interface/UnixSignalHandlers.h

We note that Root is in the main thread at present and that it has provided functions to handle 15 of the 31 defined signals (excluding the 32 real-time signals available on Linux systems). We have added a handler for SIGUSR2 to tell the CMS.EventProcessor to stop processing on completion of the current event, close open files and shut down cleanly rather than go on to the next event.

Signal names, numbers and their description

The attached table shows the list of signal names, signal numbers and their description. Where a handler is defined - either by Root or by the framework - we list how CMSSW responds. When there is no user defined response, the system default is taken.

/* Signal number definitions.  Linux version.
 * Copyright (C) 1995,1996,1997,1998,1999,2003 Free Software Foundation, Inc.
 */

System   System   Description                Root                    Root  Response
Name     Value                               Name                    Value

SIGHUP     1      Hangup                                                   System default
SIGINT     2      Interrupt                  kSigInterrupt              6  Exit
SIGQUIT    3      Quit                       kSigQuit                   5  Dump and exit
SIGILL     4      Illegal instruction        kSigIllegalInstruction     4  Dump and exit
SIGTRAP    5      Trace trap                                               System default
SIGABRT    6      Abort                                                    System default
SIGIOT     6      IOT trap                                                 System default
SIGBUS     7      BUS error                  kSigBus                    0  Dump and exit
SIGFPE     8      Floating-point exception   kSigFloatingException     11  Dump and exit
SIGKILL    9      Kill, unblockable                                        System default
SIGUSR1   10      User-defined signal 1      kSigUser1                 13  Exit
SIGSEGV   11      Segmentation violation     kSigSegmentationViolation  1  Dump and exit
SIGUSR2   12      User-defined signal 2      kSigUser2                 14  Clean exit
SIGPIPE   13      Broken pipe                kSigPipe                   3  Exit
SIGALRM   14      Alarm clock                kSigAlarm                  8  Exit
SIGTERM   15      Termination                kSigTermination           12  Exit
SIGSTKFLT 16      Stack fault.                                             System default
SIGCLD    SIGCHLD Same as SIGCHLD                                          Ignore
SIGCHLD   17      Child status has changed   kSigChild                  9  Ignore
SIGCONT   18      Continue                                                 System default
SIGSTOP   19      Stop, unblockable                                        System default
SIGTSTP   20      Keyboard stop                                            Sytsem default
SIGTTIN   21      Background read from tty                                 System default
SIGTTOU   22      Background write to tty                                  System default
SIGURG    23      Urgent condition on socket kSigUrgent                10  Ignore
SIGXCPU   24      CPU limit exceeded                                       System default
SIGXFSZ   25      File size limit exceeded                                 System default
SIGVTALRM 26      Virtual alarm clock                                      System default
SIGPROF   27      Profiling alarm clock                                    System default
SIGWINCH  28      Window size change         kSigWindowChanged          7  Ignore
SIGPOLL   SIGIO   Pollable event occurred                                  System default
SIGIO     29      I/O now possible                                         System default
SIGPWR    30      Power failure restart                                    System default
SIGSYS    31      Bad system call.           kSigSystem                 2  Ignore
SIGUNUSED 31

_NSIG     65      Biggest signal number + 1 (including real-time signals).

Notes

1. Dumps may or may not be accompanied by a stack trace. When no stack trace was produced, one can always be obtained with gdb provided both the executable and the core.nnnn files are still available. This will probably not be true for farm or grid jobs.
2. Exit is equivalent to calling exit(); and will almost surely leave the last event in a mangled state. Open files are closed but the last "record" will probably be incomplete.
3. Clean exit means finish up the current event, write output files and shut down cleanly instead of going on to the next event.

Review Status

Editor/Review and date	Comments
JohnMarraffino - 19 Feb 2006	Page created
JennyWilliams - 20 Feb 2006	Twiki editing for swguide consistency

Responsible: ChrisDJones
Last reviewed by: MostRecentReviewer