Index: /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.parallel.dvo =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.parallel.dvo (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/addstar/test/relphot.parallel.dvo (revision 33301) @@ -132,7 +132,7 @@ # run relphot on the parallel db and check that the images now match the expected values - exec relphot -parallel-serial 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 - - exec dvodist -in $catdir.p0 + 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 + + exec dvodist -in $catdir.p0 >& log.dvodist.in for i 0 mjd_nc[] @@ -432,5 +432,5 @@ $catdir = $tmp1/$tmp2 - exec rsync -auv $catdir/ $catdir.p0/ + exec rsync -auv $catdir/ $catdir.p0/ >& log.rsync mkdir $catdir.d1 mkdir $catdir.d2 @@ -442,5 +442,5 @@ exec echo "3 $hostname $catdir.d3" >> $catdir.p0/HostTable.dat - exec dvodist -out $catdir.p0 -end - + exec dvodist -out $catdir.p0 >& log.dvodist.out +end + Index: /branches/eam_branches/ipp-20111122/Ohana/src/libohana/include/ohana.h =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/libohana/include/ohana.h (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/libohana/include/ohana.h (revision 33301) @@ -264,4 +264,5 @@ int check_dir_access PROTO((char *path, int verbose)); int check_file_exec PROTO((char *filename)); +char *abspath PROTO((char *oldpath, int maxlength)); /* in glockfile.c */ Index: /branches/eam_branches/ipp-20111122/Ohana/src/libohana/src/findexec.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/libohana/src/findexec.c (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/libohana/src/findexec.c (revision 33301) @@ -404,2 +404,29 @@ return TRUE; } + +// ohana-based equivalent to 'realpath'. realpath does not guarantee an absolute path (eg on Solaris), +// just a 'cannonical path'. I don't care about resolving out links, just ensuring an absolute path +// also, realpath is a bit ambiguous on the maxlength argument +char *abspath (char *oldpath, int maxlength) { + + if (!oldpath) return NULL; + + if (oldpath[0] == '/') { + char *newpath = strcreate (oldpath); + return newpath; + } + + char *cwd = getcwd(NULL, maxlength); + if (cwd == NULL) { + // XXX need an error reporting function... + fprintf (stderr, "error getting cwd (longer than %d chars?)\n", maxlength); + return NULL; + } + + char *newpath = NULL; + + int Nbytes = strlen(cwd) + strlen(oldpath) + 2; + ALLOCATE (newpath, char, Nbytes); + snprintf (newpath, Nbytes, "%s/%s", cwd, oldpath); + return newpath; +} Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/include/relphot.h (revision 33301) @@ -71,13 +71,14 @@ /* global variables set in parameter file */ -char ImageCat[256]; -char ImageTemplate[256]; -char CatTemplate[256]; -char GSCFILE[256]; -char CATDIR[256]; +# define MAX_PATH_LENGTH 1024 +char ImageCat[MAX_PATH_LENGTH]; +char ImageTemplate[MAX_PATH_LENGTH]; +char CatTemplate[MAX_PATH_LENGTH]; +char GSCFILE[MAX_PATH_LENGTH]; +char *CATDIR; char CATMODE[16]; /* raw, mef, split, mysql */ char CATFORMAT[16]; /* internal, elixir, loneos, panstarrs */ -char CameraConfig[256]; -char SKY_TABLE[256]; +char CameraConfig[MAX_PATH_LENGTH]; +char SKY_TABLE[MAX_PATH_LENGTH]; int SKY_DEPTH; /** XXX EAM : depth of catalog tables, fix usage */ @@ -314,2 +315,3 @@ int CopyMeasureTiny (MeasureTiny *measureT, Measure *measure); +int HostTableWaitJobs (HostTable *table, char *file, int lineno); Index: /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ConfigInit.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ConfigInit.c (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/ConfigInit.c (revision 33301) @@ -33,6 +33,12 @@ GetConfig (config, "IMAGE_GOOD_FRACTION", "%lf", 0, &IMAGE_GOOD_FRACTION); + // 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); + GetConfig (config, "GSCFILE", "%s", 0, GSCFILE); - GetConfig (config, "CATDIR", "%s", 0, CATDIR); ScanConfig (config, "CATMODE", "%s", 0, CATMODE); ScanConfig (config, "CATFORMAT", "%s", 0, CATFORMAT); 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 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/load_catalogs.c (revision 33301) @@ -108,24 +108,11 @@ } - char *cwd; - if ((cwd = getcwd(NULL, 1024)) == NULL) { - fprintf (stderr, "error getting cwd (longer than 1024 chars?)\n"); - exit (1); - } - int i; for (i = 0; i < table->Nhosts; i++) { - // if table->hosts[i].pathname is a relative path (pathname[0] is not '/'), then - // we need to prepend our current path - - if (table->hosts[i].pathname[0] != '/') { - int Nbytes = strlen(cwd) + strlen(table->hosts[i].pathname) + 2; - char *newpath = NULL; - ALLOCATE (newpath, char, Nbytes); - snprintf (newpath, Nbytes, "%s/%s", cwd, table->hosts[i].pathname); - free (table->hosts[i].pathname); - table->hosts[i].pathname = newpath; - } + // 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 catalogFile[512]; @@ -190,37 +177,5 @@ } if (!PARALLEL_MANUAL && !PARALLEL_SERIAL) { - // we have launched table->Nhosts jobs; wait for all of them to complete... - for (i = 0; i < table->Nhosts; i++) { - int status = 0; - int pid = waitpid (-1, &status, 0); - - // XXX we'll need to pass in WNOHANG and keep retrying if we want to have a timeout... - // find the host which has finished - int j; - for (j = 0; j < table->Nhosts; j++) { - if (table->hosts[j].pid != pid) continue; - // check on the status of this and report any output? - fprintf (stderr, "job finished for %s (%d)\n", table->hosts[j].hostname, pid); - 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); - } - } - // 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"); - } - } + HostTableWaitJobs (table, __FILE__, __LINE__); } 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 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/reload_catalogs.c (revision 33301) @@ -1,3 +1,5 @@ # include "relphot.h" +# include +# include # define DEBUG 1 @@ -132,4 +134,9 @@ int i; 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]; @@ -179,5 +186,4 @@ } table->hosts[i].pid = pid; // save for future reference - // check that all hosts started OK? } } @@ -186,9 +192,77 @@ fprintf (stderr, "run the relphot_client commands above. when these are done, hit return\n"); getchar(); - } else { - // watch for stdout / stderr from those jobs... - // wait for all of them to complete... - } - + } + if (!PARALLEL_MANUAL && !PARALLEL_SERIAL) { + HostTableWaitJobs (table, __FILE__, __LINE__); + } 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/relphot/src/relphot_client.c =================================================================== --- /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/relphot_client.c (revision 33300) +++ /branches/eam_branches/ipp-20111122/Ohana/src/relphot/src/relphot_client.c (revision 33301) @@ -22,7 +22,15 @@ // load the current sky table (layout of all SkyRegions) SkyTable *sky = SkyTableLoadOptimal (CATDIR, NULL, NULL, TRUE, -1, VERBOSE); + if (!sky) { + fprintf (stderr, "ERROR running loading sky table from %s\n", CATDIR); + exit (2); + } SkyTableSetFilenames (sky, CATDIR, "cpt"); SkyList *skylist = SkyListByPatch (sky, -1, &UserPatch); + if (!skylist) { + fprintf (stderr, "ERROR setting up skylist for %s\n", CATDIR); + exit (2); + } switch (MODE) { @@ -30,7 +38,12 @@ int Ncatalog; Catalog *catalog = load_catalogs (skylist, &Ncatalog, HOST_ID, HOSTDIR); + if (!catalog) { + fprintf (stderr, "ERROR loading catalogs from %s\n", CATDIR); + exit (2); + } BrightCatalog *bcatalog = BrightCatalogMerge (catalog, Ncatalog); if (!BrightCatalogSave (BCATALOG, bcatalog)) { - abort(); + fprintf (stderr, "ERROR saving bright catalog from %s\n", CATDIR); + exit (2); } break; @@ -41,4 +54,8 @@ off_t Nimage; ImageSubset *image = ImageSubsetLoad (IMAGES, &Nimage); + if (!image) { + fprintf (stderr, "ERROR loading image subset %s\n", CATDIR); + exit (2); + } // save the available image information in the static array in ImageOps.c