Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avextract.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avextract.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avextract.c (revision 34654) @@ -137,5 +137,5 @@ // this does all the work of re-packaging the command, calling it on the remote machines, then loading in the results if (PARALLEL && !HOST_ID) { - int status = HostTableParallelOps (argc, argv, RESULT_FILE, 0, VERBOSE); + int status = HostTableParallelOps (argc, argv, RESULT_FILE, TRUE, 0, VERBOSE); dbFreeFields (fields, Nfields); Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avmatch.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avmatch.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/avmatch.c (revision 34654) @@ -95,5 +95,5 @@ // I need to pass the RA & DEC vectors to the remote clients... - int status = HostTableParallelOps (argc, argv, RESULT_FILE, RAvec->Nelements, VERBOSE); + int status = HostTableParallelOps (argc, argv, RESULT_FILE, TRUE, RAvec->Nelements, VERBOSE); if (vec) free (vec); Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/dvo_host_utils.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/dvo_host_utils.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/dvo_host_utils.c (revision 34654) @@ -79,10 +79,14 @@ } -// bundle the arguments into a command and pass to dvo_client. this implementation -// expects there to be a result file from the clients, and to load this into vectors in -// the main shell. 'Nelements' is a temp hack : for most commands, the result vectors are -// concatenated, but for avmatch, the vectors are merged by index into a pre-known -// length. this is probably not a solution to a general problem.. -int HostTableParallelOps (int argc, char **argv, char *ResultFile, int Nelements, int VERBOSE) { +// bundle the arguments into a command and pass to dvo_client. + +// the normal ending step expects there to be a result file from the clients, and to load +// this into vectors in the main shell. 'Nelements' is a temp hack : for most commands, +// the result vectors are concatenated, but for avmatch, the vectors are merged by index +// into a pre-known length. this is probably not a solution to a general problem.. + +// an alternative ending step ignores the result files and instead saves the names into +// the list 'result:n' for the user to access as desired +int HostTableParallelOps (int argc, char **argv, char *ResultFile, int ReadVectors, int Nelements, int VERBOSE) { int i; @@ -153,9 +157,113 @@ } + // create the result file list + char name[256]; + snprintf (name, 256, "RESULT_FILE:n"); + set_int_variable (name, table->Nhosts); + snprintf (name, 256, "RESULT_DATA:n"); + set_int_variable (name, table->Nhosts); + snprintf (name, 256, "RESULT_STATUS:n"); + set_int_variable (name, table->Nhosts); + // load fields from file int Nvec = 0; Vector **vec = NULL; for (i = 0; i < table->Nhosts; i++) { + + snprintf (name, 256, "RESULT_FILE:%d", i); + set_str_variable (name, table->hosts[i].results); + + // DATA : 0 (unread), 1 (read) + snprintf (name, 256, "RESULT_DATA:%d", i); + set_int_variable (name, 0); + + // STATUS : 0 (normal exit), -1 (crash), N (failure exit status) + snprintf (name, 256, "RESULT_STATUS:%d", i); + set_int_variable (name, table->hosts[i].status); + if (table->hosts[i].status) continue; + + if (ReadVectors) { + int Ninvec = 0; + Vector **invec = ReadVectorTableFITS (table->hosts[i].results, "RESULT", &Ninvec); + if (!invec) { + // failed to read the file, now what? + gprint (GP_ERR, "failed to read remote result file : %s\n", table->hosts[i].results); + free (table->hosts[i].results); + table->hosts[i].results = NULL; + continue; + } + free (table->hosts[i].results); + table->hosts[i].results = NULL; + set_int_variable (name, 1); // result file has been read + + if (Nelements == 0) { + vec = MergeVectors (vec, &Nvec, invec, Ninvec); + if (vec != invec) { + FreeVectorArray (invec, Ninvec); + } + } else { + vec = MergeVectorsByIndex (vec, &Nvec, invec, Ninvec, Nelements); + FreeVectorArray (invec, Ninvec); + } + } + } + + // write vectors to a table (this is used by parallel dvo operations, but can be used elsewhere) + if (ResultFile) { + int status = WriteVectorTableFITS (ResultFile, "RESULT", vec, Nvec, FALSE, NULL); + if (!status) { + gprint (GP_ERR, "failed to write result file %s\n", ResultFile); + return FALSE; + } + } + + for (i = 0; i < Nvec; i++) { + AssignVector (vec[i], vec[i]->name, ANYVECTOR, TRUE); + } + free (vec); + + free (table); + return TRUE; +} + +// re-gather the remote results files: this can be used in case one of the clients failed, +// and has since been re-run +int HostTableReloadResults (char *uniquer, int VERBOSE) { + + int i; + + // load the list of hosts + SkyTable *sky = GetSkyTable(); + if (!sky) { + gprint (GP_ERR, "failed to load sky table for database\n"); + return FALSE; + } + + char *CATDIR = GetCATDIR (); + if (!CATDIR) { + gprint (GP_ERR, "failed to get CATDIR for database\n"); + return FALSE; + } + + HostTable *table = HostTableLoad (CATDIR, sky->hosts); + if (!table) { + gprint (GP_ERR, "ERROR: failure reading Host Table %s for database %s\n", sky->hosts, CATDIR); + return FALSE; + } + + // load fields from file + int Nvec = 0; + Vector **vec = NULL; + for (i = 0; i < table->Nhosts; i++) { + // ensure that the paths are absolute path names + char *tmppath = abspath (table->hosts[i].pathname, DVO_MAX_PATH); + free (table->hosts[i].pathname); + table->hosts[i].pathname = tmppath; + + // need to save the results filename with the uniquer + // XXX a bit of a waste (but only 1024 * 60 bytes or so + ALLOCATE (table->hosts[i].results, char, DVO_MAX_PATH); + snprintf (table->hosts[i].results, DVO_MAX_PATH, "%s/dvo.results.%s.fits", table->hosts[i].pathname, uniquer); int Ninvec = 0; @@ -171,21 +279,9 @@ table->hosts[i].results = NULL; - if (Nelements == 0) { - vec = MergeVectors (vec, &Nvec, invec, Ninvec); - if (vec != invec) { - FreeVectorArray (invec, Ninvec); - } - } else { - vec = MergeVectorsByIndex (vec, &Nvec, invec, Ninvec, Nelements); + // fprintf (stderr, "%s : %d\n", table->hosts[i].pathname, invec[0]->Nelements); + + vec = MergeVectors (vec, &Nvec, invec, Ninvec); + if (vec != invec) { FreeVectorArray (invec, Ninvec); - } - } - - // write vectors to a table (this is used by parallel dvo operations, but can be used elsewhere) - if (ResultFile) { - int status = WriteVectorTableFITS (ResultFile, "RESULT", vec, Nvec, FALSE, NULL); - if (!status) { - gprint (GP_ERR, "failed to write result file %s\n", ResultFile); - return FALSE; } } @@ -199,70 +295,2 @@ return TRUE; } - -// re-gather the remote results files: this can be used in case one of the clients failed, -// and has since been re-run -int HostTableReloadResults (char *uniquer, int VERBOSE) { - - int i; - - // load the list of hosts - SkyTable *sky = GetSkyTable(); - if (!sky) { - gprint (GP_ERR, "failed to load sky table for database\n"); - return FALSE; - } - - char *CATDIR = GetCATDIR (); - if (!CATDIR) { - gprint (GP_ERR, "failed to get CATDIR for database\n"); - return FALSE; - } - - HostTable *table = HostTableLoad (CATDIR, sky->hosts); - if (!table) { - gprint (GP_ERR, "ERROR: failure reading Host Table %s for database %s\n", sky->hosts, CATDIR); - return FALSE; - } - - // load fields from file - int Nvec = 0; - Vector **vec = NULL; - for (i = 0; i < table->Nhosts; i++) { - // ensure that the paths are absolute path names - char *tmppath = abspath (table->hosts[i].pathname, DVO_MAX_PATH); - free (table->hosts[i].pathname); - table->hosts[i].pathname = tmppath; - - // need to save the results filename with the uniquer - // XXX a bit of a waste (but only 1024 * 60 bytes or so - ALLOCATE (table->hosts[i].results, char, DVO_MAX_PATH); - snprintf (table->hosts[i].results, DVO_MAX_PATH, "%s/dvo.results.%s.fits", table->hosts[i].pathname, uniquer); - - int Ninvec = 0; - Vector **invec = ReadVectorTableFITS (table->hosts[i].results, "RESULT", &Ninvec); - if (!invec) { - // failed to read the file, now what? - gprint (GP_ERR, "failed to read remote result file : %s\n", table->hosts[i].results); - free (table->hosts[i].results); - table->hosts[i].results = NULL; - continue; - } - free (table->hosts[i].results); - table->hosts[i].results = NULL; - - // fprintf (stderr, "%s : %d\n", table->hosts[i].pathname, invec[0]->Nelements); - - vec = MergeVectors (vec, &Nvec, invec, Ninvec); - if (vec != invec) { - FreeVectorArray (invec, Ninvec); - } - } - - for (i = 0; i < Nvec; i++) { - AssignVector (vec[i], vec[i]->name, ANYVECTOR, TRUE); - } - free (vec); - - free (table); - return TRUE; -} Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mextract.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mextract.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mextract.c (revision 34654) @@ -108,5 +108,5 @@ // this does all the work of re-packaging the command, calling it on the remote machines, then loading in the results if (PARALLEL && !HOST_ID) { - int status = HostTableParallelOps (argc, argv, RESULT_FILE, 0, VERBOSE); + int status = HostTableParallelOps (argc, argv, RESULT_FILE, TRUE, 0, VERBOSE); dbFreeFields (fields, Nfields); Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mmatch.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mmatch.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/mmatch.c (revision 34654) @@ -124,5 +124,5 @@ // call the remote client - int status = HostTableParallelOps (targc, targv, RESULT_FILE, 0, VERBOSE); + int status = HostTableParallelOps (targc, targv, RESULT_FILE, TRUE, 0, VERBOSE); if (vec) free (vec); Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/remote.c =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/remote.c (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/dvo/remote.c (revision 34654) @@ -12,4 +12,10 @@ remove_argument (N, &argc, argv); VERBOSE = TRUE; + } + + int ReadVectors = TRUE; + if ((N = get_argument (argc, argv, "-skip-result"))) { + remove_argument (N, &argc, argv); + ReadVectors = FALSE; } @@ -43,5 +49,5 @@ // strip of the 'remote' and send the remaining arguments to the remote machine - int status = HostTableParallelOps (argc - 1, &argv[1], NULL, 0, VERBOSE); + int status = HostTableParallelOps (argc - 1, &argv[1], NULL, ReadVectors, 0, VERBOSE); return status; } Index: /branches/eam_branches/ipp-20120905/Ohana/src/opihi/include/dvoshell.h =================================================================== --- /branches/eam_branches/ipp-20120905/Ohana/src/opihi/include/dvoshell.h (revision 34653) +++ /branches/eam_branches/ipp-20120905/Ohana/src/opihi/include/dvoshell.h (revision 34654) @@ -100,5 +100,5 @@ int HostTableLaunchJobs PROTO((HostTable *table, char *basecmd, char *options, int VERBOSE)); -int HostTableParallelOps PROTO((int argc, char **argv, char *ResultFile, int Nelements, int VERBOSE)); +int HostTableParallelOps PROTO((int argc, char **argv, char *ResultFile, int ReadVectors, int Nelements, int VERBOSE)); int HostTableReloadResults PROTO((char *uniquer, int VERBOSE));