Index: /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.setphot.parallel.dvo =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.setphot.parallel.dvo (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.setphot.parallel.dvo (revision 33303) @@ -158,5 +158,5 @@ # run setphot on the serial db and check that the images now match the expected values - exec setphot -update -ubercal testzpt.fits -D CATDIR $catdir + exec setphot -reset -update -ubercal testzpt.fits -D CATDIR $catdir for i 0 mjd_uc[] ckexposure $catdir mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i setphot_uc @@ -178,8 +178,9 @@ # run setphot on the parallel db and check that the images now match the expected values - exec setphot -parallel -update -ubercal testzpt.fits -D CATDIR $catdir.p0 + exec setphot -reset -parallel -update -ubercal testzpt.fits -D CATDIR $catdir.p0 # bring the detections back to the local db - exec dvodist -in $catdir.p0 >& log.dvodist.in + # exec dvodist -in $catdir.p0 >& log.dvodist.in + # XXX this should now not be needed # check the results @@ -190,14 +191,8 @@ ckexposure $catdir.p0 mjd_nc[$i] zpt_nc[$i] exptime_nc[$i] airmass_nc[$i] $filt_nc:$i setphot_nc end - - # push the detections back out to the parallel db - exec dvodist -out $catdir.p0 >& log.dvodist.in # run relphot on the parallel db and check that the images now match the expected values exec relphot -parallel g,r,i -v -region 9.5 10.5 19.5 20.5 -D CATDIR $catdir.p0 -D STAR_TOOFEW 1 -D SIGMA_LIM 0.07 -statmode WT_MEAN -cloud-limit 0.5 -keep-ubercal -D IMAGE_OFFSET 0.5 -update >& log.relphot.p0 - # bring the detections back to the local db - exec dvodist -in $catdir.p0 >& log.dvodist.in - for i 0 mjd_uc[] ckexposure $catdir.p0 mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i relphot @@ -208,4 +203,50 @@ tapDONE +end + +macro qt + + ## SER + + # run setphot on the serial db and check that the images now match the expected values + exec setphot -reset -update -ubercal testzpt.fits -D CATDIR $catdir + for i 0 mjd_uc[] + ckexposure $catdir mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i setphot_uc + end + for i 0 mjd_nc[] + ckexposure $catdir mjd_nc[$i] zpt_nc[$i] exptime_nc[$i] airmass_nc[$i] $filt_nc:$i setphot_nc + end + + # run relphot on the serial db and check that the images now match the expected values + exec relphot g,r,i -v -region 9.5 10.5 19.5 20.5 -D CATDIR $catdir -D STAR_TOOFEW 1 -D SIGMA_LIM 0.07 -statmode WT_MEAN -cloud-limit 0.5 -keep-ubercal -D IMAGE_OFFSET 0.5 -update >& log.relphot.s0 + for i 0 mjd_uc[] + ckexposure $catdir mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i relphot + end + for i 0 mjd_nc[] + ckexposure $catdir mjd_nc[$i] zpt_nc[$i] exptime_nc[$i] airmass_nc[$i] $filt_nc:$i relphot + end + + ## PAR + + # run setphot on the parallel db and check that the images now match the expected values + exec setphot -reset -parallel -update -ubercal testzpt.fits -D CATDIR $catdir.p0 + + # check the results + for i 0 mjd_uc[] + ckexposure $catdir.p0 mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i setphot_uc + end + for i 0 mjd_nc[] + ckexposure $catdir.p0 mjd_nc[$i] zpt_nc[$i] exptime_nc[$i] airmass_nc[$i] $filt_nc:$i setphot_nc + end + + # run relphot on the parallel db and check that the images now match the expected values + exec relphot -parallel g,r,i -v -region 9.5 10.5 19.5 20.5 -D CATDIR $catdir.p0 -D STAR_TOOFEW 1 -D SIGMA_LIM 0.07 -statmode WT_MEAN -cloud-limit 0.5 -keep-ubercal -D IMAGE_OFFSET 0.5 -update >& log.relphot.p0 + + for i 0 mjd_uc[] + ckexposure $catdir.p0 mjd_uc[$i] zpt_uc[$i] exptime_uc[$i] airmass_uc[$i] $filt_uc:$i relphot + end + for i 0 mjd_nc[] + ckexposure $catdir.p0 mjd_nc[$i] zpt_nc[$i] exptime_nc[$i] airmass_nc[$i] $filt_nc:$i relphot + end end Index: /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/include/dvo.h =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/include/dvo.h (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/include/dvo.h (revision 33303) @@ -579,4 +579,5 @@ // functions for parallel DVO HostTable *HostTableLoad (char *catdir, char *rootname); +int HostTableWaitJobs (HostTable *table, char *file, int lineno); # endif Index: /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/src/HostTable.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/src/HostTable.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/libdvo/src/HostTable.c (revision 33303) @@ -1,3 +1,5 @@ # include "dvo.h" +# include +# include void InitHosts (HostInfo *hosts, int Nhosts, int NHOSTS) { @@ -114,2 +116,71 @@ return table; } + +// wait for all children to complete, report output to stdout +int HostTableWaitJobs (HostTable *table, char *file, int lineno) { + + int i; + + // we have launched table->Nhosts jobs; wait for all of them to complete... + // if one (N) failed to launch, we will get an ECHILD error from the last (N) calls + int done = FALSE; + for (i = 0; !done && (i < table->Nhosts); i++) { + int status = 0; + + // XXX we'll need to pass in WNOHANG and keep retrying if we want to have a timeout... + int pid = waitpid (-1, &status, 0); + if (!pid) { + // this should only occur if we called waitpid with the WNOHANG option + fprintf (stderr, "programming error (1)? %s %d", file, lineno); + exit (2); + } + if (pid == -1) { + switch (errno) { + case ECHILD: + done = TRUE; + break; + default: + fprintf (stderr, "programming error (2)? %s %d", file, lineno); + exit (2); + } + } + + // find the host which has finished + int j; + int found = FALSE; + for (j = 0; j < table->Nhosts; j++) { + if (table->hosts[j].pid != pid) continue; + found = TRUE; + // check on the status of this and report any output? + fprintf (stderr, "job finished for %s (%d)\n", table->hosts[j].hostname, pid); + // read the stderr and stdout + IOBuffer buffer; + InitIOBuffer (&buffer, 100); + EmptyIOBuffer (&buffer, 100, table->hosts[j].stdio[1]); + fprintf (stderr, "--- stdout from %s ---\n", table->hosts[j].hostname); + write (STDOUT_FILENO, buffer.buffer, buffer.Nbuffer); + fprintf (stderr, "\n"); + + InitIOBuffer (&buffer, 100); + EmptyIOBuffer (&buffer, 100, table->hosts[j].stdio[2]); + fprintf (stderr, "--- stderr from %s ---\n", table->hosts[j].hostname); + write (STDOUT_FILENO, buffer.buffer, buffer.Nbuffer); + fprintf (stderr, "\n"); + if (WIFEXITED(status)) { + fprintf (stderr, "normal completion, exit status is %d\n", WEXITSTATUS(status)); + if (WEXITSTATUS(status)) { + fprintf (stderr, "job failed on %s\n", table->hosts[j].hostname); + exit (1); + } + } else { + fprintf (stderr, "job exited abnormally on %s\n", table->hosts[j].hostname); + exit (1); + } + } + if (!found) { + fprintf (stderr, "failed to matched finished job to known host!\n"); + exit (2); + } + } + return TRUE; +} Index: /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyFromHostLocation.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyFromHostLocation.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyFromHostLocation.c (revision 33303) @@ -86,4 +86,16 @@ valid: + + // the output (tgtname) could exist and be a link to the same file; delete it if it exists + status = stat (tgtname, &filestat); + if (!status) { + // a tgtfile already exists; get rid of it + if (unlink (tgtname)) { + fprintf (stderr, "failed to remove CATDIR file %s\n", tgtname); + success = FALSE; + continue; + } + } + // read srcname, write to tgtname, get MD5 sum for srcname as we go: if (!get_md5_with_copy (srcname, tgtname, srcDigest)) { Index: /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyToHostLocation.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyToHostLocation.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyToHostLocation.c (revision 33303) @@ -11,4 +11,5 @@ struct stat filestat; char srcname[1024]; + char newname[1024]; char tgtname[1024]; char outdir[1024]; @@ -131,6 +132,28 @@ skylist->regions[i]->hostFlags |= DATA_ON_TGT; for (j = 0; success && (j < 4); j++) { + sprintf (tgtname, "%s/%s.%s", host->pathname, skylist->regions[i]->name, extname[j]); sprintf (srcname, "%s/%s.%s", catdir, skylist->regions[i]->name, extname[j]); - if (!DEBUG && DELETE_ORIGINAL) unlink (srcname); + if (!DEBUG && DELETE_ORIGINAL) { + unlink (srcname); + } else { + sprintf (newname, "%s/%s.%s.bck", catdir, skylist->regions[i]->name, extname[j]); + status = stat (newname, &filestat); + if (!status) { + // a backup file already exists; get rid of it + if (unlink (newname)) { + fprintf (stderr, "failed to remove old backup file %s\n", newname); + continue; + } + } + if (rename (srcname, newname)) { + fprintf (stderr, "failed to rename %s to %s\n", srcname, newname); + continue; + } + char *absname = abspath (tgtname, 1024); + if (symlink (absname, srcname)) { + fprintf (stderr, "failed to make symlink from %s to %s\n", absname, srcname); + } + free (absname); + } } } Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h (revision 33303) @@ -314,4 +314,2 @@ int CopyAverageTiny (AverageTiny *averageT, Average *average); int CopyMeasureTiny (MeasureTiny *measureT, Measure *measure); - -int HostTableWaitJobs (HostTable *table, char *file, int lineno); Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ImageSubset.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ImageSubset.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ImageSubset.c (revision 33303) @@ -43,5 +43,5 @@ if (!gfits_load_header (f, &theader)) { fclose (f); - return (NULL); + return NULL; } Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/args.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/args.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/args.c (revision 33303) @@ -394,4 +394,10 @@ } + KEEP_UBERCAL = FALSE; + if ((N = get_argument (argc, argv, "-keep-ubercal"))) { + remove_argument (N, &argc, argv); + KEEP_UBERCAL = TRUE; + } + /* define time */ TimeSelect = FALSE; Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/load_catalogs.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/load_catalogs.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/load_catalogs.c (revision 33303) @@ -121,4 +121,5 @@ // options / arguments that can affect relphot_client -load: // VERBOSE, VERBOSE2 + // KEEP_UBERCAL // RESET (-reset) // TimeSelect -time @@ -138,4 +139,5 @@ if (VERBOSE2) { snprintf (tmpline, 1024, "%s -vv", command); strcpy (command, tmpline); } if (RESET) { snprintf (tmpline, 1024, "%s -reset", command); strcpy (command, tmpline); } + if (KEEP_UBERCAL) { snprintf (tmpline, 1024, "%s -keep-ubercal", command); strcpy (command, tmpline); } if (DophotSelect) { snprintf (tmpline, 1024, "%s -dophot %d", command, DophotValue); strcpy (command, tmpline); } if (ImagSelect) { snprintf (tmpline, 1024, "%s -instmag %f %f", command, ImagMin, ImagMax); strcpy (command, tmpline); } Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/reload_catalogs.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/reload_catalogs.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/reload_catalogs.c (revision 33303) @@ -1,5 +1,3 @@ # include "relphot.h" -# include -# include # define DEBUG 1 @@ -157,4 +155,5 @@ if (VERBOSE2) { snprintf (tmpline, 1024, "%s -vv", command); strcpy (command, tmpline); } if (RESET) { snprintf (tmpline, 1024, "%s -reset", command); strcpy (command, tmpline); } + if (KEEP_UBERCAL) { snprintf (tmpline, 1024, "%s -keep-ubercal", command); strcpy (command, tmpline); } if (AreaSelect) { snprintf (tmpline, 1024, "%s -area %f %f %f %f", command, AreaXmin, AreaXmax, AreaYmin, AreaYmax); strcpy (command, tmpline); } if (TimeSelect) { @@ -198,71 +197,2 @@ return (TRUE); } - -// wait for all children to complete, report output to stdout -int HostTableWaitJobs (HostTable *table, char *file, int lineno) { - - int i; - - // we have launched table->Nhosts jobs; wait for all of them to complete... - // if one (N) failed to launch, we will get an ECHILD error from the last (N) calls - int done = FALSE; - for (i = 0; !done && (i < table->Nhosts); i++) { - int status = 0; - - // XXX we'll need to pass in WNOHANG and keep retrying if we want to have a timeout... - int pid = waitpid (-1, &status, 0); - if (!pid) { - // this should only occur if we called waitpid with the WNOHANG option - fprintf (stderr, "programming error (1)? %s %d", file, lineno); - exit (2); - } - if (pid == -1) { - switch (errno) { - case ECHILD: - done = TRUE; - break; - default: - fprintf (stderr, "programming error (2)? %s %d", file, lineno); - exit (2); - } - } - - // find the host which has finished - int j; - int found = FALSE; - for (j = 0; j < table->Nhosts; j++) { - if (table->hosts[j].pid != pid) continue; - found = TRUE; - // check on the status of this and report any output? - fprintf (stderr, "job finished for %s (%d)\n", table->hosts[j].hostname, pid); - // read the stderr and stdout - IOBuffer buffer; - InitIOBuffer (&buffer, 100); - EmptyIOBuffer (&buffer, 100, table->hosts[j].stdio[1]); - fprintf (stderr, "--- stdout from %s ---\n", table->hosts[j].hostname); - write (STDOUT_FILENO, buffer.buffer, buffer.Nbuffer); - fprintf (stderr, "\n"); - - InitIOBuffer (&buffer, 100); - EmptyIOBuffer (&buffer, 100, table->hosts[j].stdio[2]); - fprintf (stderr, "--- stderr from %s ---\n", table->hosts[j].hostname); - write (STDOUT_FILENO, buffer.buffer, buffer.Nbuffer); - fprintf (stderr, "\n"); - if (WIFEXITED(status)) { - fprintf (stderr, "normal completion, exit status is %d\n", WEXITSTATUS(status)); - if (WEXITSTATUS(status)) { - fprintf (stderr, "job failed on %s\n", table->hosts[j].hostname); - exit (1); - } - } else { - fprintf (stderr, "job exited abnormally on %s\n", table->hosts[j].hostname); - exit (1); - } - } - if (!found) { - fprintf (stderr, "failed to matched finished job to known host!\n"); - exit (2); - } - } - return TRUE; -} Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/include/setphot.h =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/include/setphot.h (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/include/setphot.h (revision 33303) @@ -19,14 +19,18 @@ /* global variables set in parameter file */ -char ImageCat[256]; -char CATDIR[256]; +# define MAX_PATH_LENGTH 1024 +char ImageCat[MAX_PATH_LENGTH]; +char *CATDIR; int HOST_ID; char *HOSTDIR; char *IMAGES; int VERBOSE; +int RESET; int UBERCAL; // load the supplied ubercal zero point fits table (with flat-field corrections) int NO_METADATA; // the supplied ubercal data has no descriptive metadata -int VERBOSE; int UPDATE; +int PARALLEL; +int PARALLEL_MANUAL; +int PARALLEL_SERIAL; /***** prototypes ****/ Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/ImageSubset.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/ImageSubset.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/ImageSubset.c (revision 33303) @@ -39,10 +39,15 @@ // load data for this header - if (!gfits_load_header (f, &theader)) return (NULL); + if (!gfits_load_header (f, &theader)) { + fclose (f); + return NULL; + } // read the fits table bytes - if (!gfits_fread_ftable_data (f, &ftable, FALSE)) return (NULL); - - // XXX close the fits file here + if (!gfits_fread_ftable_data (f, &ftable, FALSE)) { + fclose (f); + return NULL; + } + fclose (f); // a bit annoying : we read the entire block of data (180M), then extract the columns, then set the image structure values. @@ -153,4 +158,7 @@ gfits_free_table (&ftable); + fclose (f); + fflush (f); + return TRUE; } Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot.c (revision 33303) @@ -31,4 +31,36 @@ if ((N = get_argument (argc, argv, "-v"))) { VERBOSE = TRUE; + remove_argument (N, &argc, argv); + } + + RESET = FALSE; + if ((N = get_argument (argc, argv, "-reset"))) { + RESET = TRUE; + remove_argument (N, &argc, argv); + } + + PARALLEL = FALSE; + if ((N = get_argument (argc, argv, "-parallel"))) { + PARALLEL = TRUE; + remove_argument (N, &argc, argv); + } + // this is a test mode : rather than launching the remote jobs and waiting for completion, + // setphot will simply list the remote command and wait for the user to signal completion + PARALLEL_MANUAL = FALSE; + if ((N = get_argument (argc, argv, "-parallel-manual"))) { + PARALLEL = TRUE; // -parallel-manual implies -parallel + PARALLEL_MANUAL = TRUE; + remove_argument (N, &argc, argv); + } + // this is a test mode : rather than launching the setphot_client jobs remotely, they are + // run in serial via 'system' + PARALLEL_SERIAL = FALSE; + if ((N = get_argument (argc, argv, "-parallel-serial"))) { + if (PARALLEL_MANUAL) { + fprintf (stderr, "ERROR: cannot mix -parallel-manual and -parallel-serial\n"); + exit (1); + } + PARALLEL = TRUE; // -parallel-serial implies -parallel + PARALLEL_SERIAL = TRUE; remove_argument (N, &argc, argv); } @@ -82,5 +114,11 @@ if (VERBOSE) fprintf (stderr, "loaded config file: %s\n", file); - GetConfig (config, "CATDIR", "%s", 0, CATDIR); + // force CATDIR to be absolute (so parallel mode will work) + char *tmpcatdir = NULL; + ALLOCATE (tmpcatdir, char, MAX_PATH_LENGTH); + GetConfig (config, "CATDIR", "%s", 0, tmpcatdir); + CATDIR = abspath (tmpcatdir, MAX_PATH_LENGTH); + free (tmpcatdir); + sprintf (ImageCat, "%s/Images.dat", CATDIR); Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot_client.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot_client.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/initialize_setphot_client.c (revision 33303) @@ -33,4 +33,10 @@ } + RESET = FALSE; + if ((N = get_argument (argc, argv, "-reset"))) { + RESET = TRUE; + remove_argument (N, &argc, argv); + } + UBERCAL = FALSE; if ((N = get_argument (argc, argv, "-ubercal"))) { @@ -61,11 +67,10 @@ if (!HOSTDIR) goto usage; - CATDIR[0] = 0; if ((N = get_argument (argc, argv, "-catdir"))) { remove_argument (N, &argc, argv); - strcpy (CATDIR, argv[N]); + CATDIR = strcreate (argv[N]); remove_argument (N, &argc, argv); } - if (!CATDIR[0]) goto usage; + if (!CATDIR) goto usage; IMAGES = NULL; @@ -81,5 +86,5 @@ usage: - fprintf (stderr, "USAGE: setphot_client -hostname (hostname) -catdir (catdir) -hostdir (hostdir) -images (images) [options]\n"); + fprintf (stderr, "USAGE: setphot_client -hostID (hostID) -catdir (catdir) -hostdir (hostdir) -images (images) [options]\n"); fprintf (stderr, " options:\n"); fprintf (stderr, " -v : verbose mode\n"); Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/match_zpts_to_images.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/match_zpts_to_images.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/match_zpts_to_images.c (revision 33303) @@ -38,4 +38,12 @@ // sort the zpts sort_zpts_by_time (zpts, Nzpts); + + if (RESET) { + for (i = 0; i < Nimage; i++) { + image[i].Mcal = 0.0; + image[i].dMcal = NAN; + image[i].flags &= ~ID_IMAGE_PHOTOM_NOCAL; // clear the NOCAL flag + } + } NImatch = 0; // matched images Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/setphot.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/setphot.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/setphot.c (revision 33303) @@ -21,6 +21,7 @@ zpts = load_zpt_ubercal (argv[1], &Nzpts, &flatcorrTable); - char flatcorrfile[64]; - snprintf (flatcorrfile, 64, "%s/flatcorr.fits", CATDIR); + char flatcorrfile[MAX_PATH_LENGTH]; + int size = snprintf (flatcorrfile, MAX_PATH_LENGTH, "%s/flatcorr.fits", CATDIR); + assert (size < MAX_PATH_LENGTH); FlatCorrectionSave(&flatcorrTable, flatcorrfile); // XXX should this program save any old copy of this file? Index: /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/update_dvo_setphot.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/update_dvo_setphot.c (revision 33302) +++ /branches/eam_branches/ipp-20111122/Ohana/src/uniphot/src/update_dvo_setphot.c (revision 33303) @@ -14,5 +14,4 @@ SkyTableSetFilenames (sky, CATDIR, "cpt"); -# define PARALLEL 1 if (PARALLEL && !HOST_ID) { update_dvo_setphot_parallel (sky, image, Nimage); @@ -105,24 +104,49 @@ for (i = 0; i < table->Nhosts; i++) { + // ensure that the paths are absolute path names + char *tmppath = abspath (table->hosts[i].pathname, MAX_PATH_LENGTH); + free (table->hosts[i].pathname); + table->hosts[i].pathname = tmppath; + char command[1024]; - snprintf (command, 1024, "setphot_client -test -hostname %s -catdir %s -hostdir %s -images %s", - table->hosts[i].hostname, CATDIR, table->hosts[i].pathname, imageFile); + snprintf (command, 1024, "setphot_client -hostID %d -catdir %s -hostdir %s -images %s", + table->hosts[i].hostID, CATDIR, table->hosts[i].pathname, imageFile); fprintf (stderr, "command: %s\n", command); - // launch the job on the remote machine (no handshake) - int errorInfo = 0; - int pid = rconnect ("ssh", table->hosts[i].hostname, command, table->hosts[i].stdio, &errorInfo, FALSE); - if (!pid) { - if (DEBUG) fprintf (stderr, "failure to start %s (error %d)\n", table->hosts[i].hostname, errorInfo); - exit (1); + char tmpline[1024]; + if (VERBOSE) { snprintf (tmpline, 1024, "%s -v", command); strcpy (command, tmpline); } + if (RESET) { snprintf (tmpline, 1024, "%s -reset", command); strcpy (command, tmpline); } + if (UPDATE) { snprintf (tmpline, 1024, "%s -update", command); strcpy (command, tmpline); } + if (UBERCAL) { snprintf (tmpline, 1024, "%s -ubercal", command); strcpy (command, tmpline); } + + if (PARALLEL_MANUAL) continue; + + if (PARALLEL_SERIAL) { + int status = system (command); + if (status) { + fprintf (stderr, "ERROR running setphot_client\n"); + exit (2); + } + } else { + // launch the job on the remote machine (no handshake) + int errorInfo = 0; + int pid = rconnect ("ssh", table->hosts[i].hostname, command, table->hosts[i].stdio, &errorInfo, FALSE); + if (!pid) { + if (DEBUG) fprintf (stderr, "failure to start %s (error %d)\n", table->hosts[i].hostname, errorInfo); + exit (1); + } + table->hosts[i].pid = pid; // save for future reference } - table->hosts[i].pid = pid; // save for future reference - - // check that all hosts started OK? } - // watch for stdout / stderr from those jobs... - // wait for all of them to complete... + // wait for the remote jobs to be completed + if (PARALLEL_MANUAL) { + fprintf (stderr, "run the setphot_client commands above. when these are done, hit return\n"); + getchar(); + } + if (!PARALLEL_MANUAL && !PARALLEL_SERIAL) { + HostTableWaitJobs (table, __FILE__, __LINE__); + } return (TRUE);