ATP - Environment Variables

  • ATP_ENABLED If set to 1, ATP is enabled. If undefined or set to 0, ATP is not enabled at runtime. If enabled, ATP will produce an overview of the crashed program and write it to standard error. This overview provides rank information, signal that caused the crash, and crash location and assertion, if available. Default: Site-specific, or else undefined and thus not enabled.

  • ATP_OUTPUT_STYLE ATP will produce an overview of the crashed program and write it to standard error. This overview provides rank information, signal that caused the crash, and crash location and assertion, if available. By default, ATP will also produce the top frames of any crashing threads in the job. This variable can be set to configure different output styles:

    • crashing : The default, print top crashing frames.

    • minimal : Only print signal information and no frames.

    • all : Print a formatted tree view of all threads.

    • machine : Print an unformatted list of all threads.

    Default: “crashing”.

  • ATP_HOLD_TIME If set to a non-zero value, ATP will pause for the provided number of minutes after detecting a crash. The job will then be held in a stopped state, so that a debugger, such as gdb4hpc, may attach for further debugging. Default: 0 minutes.

  • ATP_IGNORE_SIGTERM If set to a non-zero value, disable ATP processing for the receipt of SIGTERM. Default: 0, SIGTERM processing enabled.

  • ATP_MAX_ANALYSIS_TIME After sending a crash analysis request to the ATP backend process, the ATP frontend process will wait the given number of seconds for crash analysis to occur. If this timeout expires, ATP will assume that the backend process was unsuccessful, and continue with job termination. Default: 300 seconds.

  • ATP_MAX_CORES After crash analysis has completed, ATP will select a subset of ranks from which to dump corefiles. The maximum number of such files is limited by this variable. While no core dumps will be produced if Linux core dumping is not enabled, if it is enabled, ATP-specific core dumping can be disabled by setting ATP_MAX_CORES to zero. If ATP routinely hits this limit, consider using ATP_USER_LIB_INTERFACE_PREFIXES to further prune the stack backtrace tree and reduce the number of core files produced. Default: 20 core files.

  • ATP_CORE_HOSTS ATP will dump cores for all ranks on the listed hosts instead of using the automatic core selection algorithm. Default: unset.

  • ATP_CORE_RANKS ATP will dump cores for the listed ranks instead of using the automatic core selection algorithm. Default: unset.

  • ATP_CORE_FILE_DIRECTORY ATP will write corefiles from the selected subset of ranks to the given directory. Default: current directory.

  • ATP_USER_LIB_INTERFACE_PREFIXES A space-separated list of routine name prefixes that defines library interface names. Routines beginning with these prefixes are considered leaf nodes and pruned from the stack backtrace tree. This enables users to reduce the set of processes included in core dumps and focus on the processes of interest. Default: not set.

  • ATP_GDB_BINARY Path to the GDB binary to be used for program analysis in place of the internal ATP analysis implementation. Set to 0 to disable GDB detection and analysis. Default: not set.

  • ATP_REFRESH_DWARF_INFO CCE version 16 has a known issue where GDB may not immediately recognize Fortran debug info. If you are missing line numbers in ATP output while running with a CCE 16 Fortran program, set ATP_REFRESH_DWARF_INFO=1 to force a refresh of the line number information. Default: not set.