diff --git a/config.h.in b/config.h.in index c5261ac4..121c551f 100644 --- a/config.h.in +++ b/config.h.in @@ -6,6 +6,12 @@ /* Whether we are using a broken srun */ #undef BROKEN_SRUN +/* Colon-separated list of potential back-end cache directories */ +#undef CACHEPATHS + +/* Back-end directory for communication and housekeeping */ +#undef COMMPATH + /* Define if were using biter for client/server communication */ #undef COMM_BITER @@ -137,9 +143,6 @@ /* Default mode for slurm launch */ #undef SLURMLAUNCH_ENABLED -/* The default local directory for Spindle */ -#undef SPINDLE_LOC - /* The default colon-separated list of directories that Spindle will not cache files out of */ #undef SPINDLE_LOCAL_PREFIX diff --git a/configure b/configure index 126b8b42..ac8d1434 100755 --- a/configure +++ b/configure @@ -848,6 +848,9 @@ enable_maintainer_mode with_default_port with_default_num_ports with_localstorage +with_cachepaths +with_cachepath +with_commpath with_default_local_prefix with_testrm with_rm @@ -1590,7 +1593,13 @@ Optional Packages: --with-default-numports=NUM Number of TCP/IP ports to scan for Spindle server communication - --with-localstorage=DIR Directory on back-ends for storing relocated files + --with-localstorage=DIR (obsolete) + Use --with-cachepaths and --with-commpath instead. + --with-cachepaths=DIR Colon-separated list of potential back-end cache + directories +, + --with-compath=DIR Back-end directory for communication and + housekeeping --with-default-local-prefix=DIRS Colon-seperated list of directories that Spindle will not cache files out of @@ -16664,9 +16673,29 @@ fi # Check whether --with-localstorage was given. if test "${with_localstorage+set}" = set; then : - withval=$with_localstorage; SPINDLE_LOC=${withval} + withval=$with_localstorage; as_fn_error $? "requested obsolete option --with-localstorage. Use --with-cachepaths and --with-commpath instead." "$LINENO" 5 +fi + + +# Check whether --with-cachepaths was given. +if test "${with_cachepaths+set}" = set; then : + withval=$with_cachepaths; CACHEPATHS=${withval} +else + CACHEPATHS=$DEFAULT_LOC +fi + + +# Check whether --with-cachepath was given. +if test "${with_cachepath+set}" = set; then : + withval=$with_cachepath; as_fn_error $? "use --with-cachepaths=DIRS (plural) instead of --with-cachepath=DIR to specify one or more cache paths" "$LINENO" 5 +fi + + +# Check whether --with-commpath was given. +if test "${with_commpath+set}" = set; then : + withval=$with_commpath; COMMPATH=${withval} else - SPINDLE_LOC=$DEFAULT_LOC + COMMPATH=$DEFAULT_LOC fi @@ -16674,7 +16703,7 @@ fi if test "${with_default_local_prefix+set}" = set; then : withval=$with_default_local_prefix; SPINDLE_LOCAL_PREFIX=${withval} else - SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$SPINDLE_LOC" + SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$COMMPATH" fi @@ -16694,7 +16723,12 @@ _ACEOF cat >>confdefs.h <<_ACEOF -#define SPINDLE_LOC "$SPINDLE_LOC" +#define COMMPATH "$COMMPATH" +_ACEOF + + +cat >>confdefs.h <<_ACEOF +#define CACHEPATHS "$CACHEPATHS" _ACEOF diff --git a/configure.common.ac b/configure.common.ac index ae2a64df..baa9c0a9 100644 --- a/configure.common.ac +++ b/configure.common.ac @@ -18,17 +18,30 @@ AC_ARG_WITH(default-num-ports, [NUM_COBO_PORTS=${withval}], [NUM_COBO_PORTS=$DEFAULT_NUM_COBO_PORTS]) AC_ARG_WITH(localstorage, - [AS_HELP_STRING([--with-localstorage=DIR],[Directory on back-ends for storing relocated files])], - [SPINDLE_LOC=${withval}], - [SPINDLE_LOC=$DEFAULT_LOC]) + [AS_HELP_STRING([--with-localstorage=DIR (obsolete)],[Use --with-cachepaths and --with-commpath instead.])], + [AC_MSG_ERROR(requested obsolete option --with-localstorage. Use --with-cachepaths and --with-commpath instead.)], + []) +AC_ARG_WITH(cachepaths, + [AS_HELP_STRING([--with-cachepaths=DIR],[Colon-separated list of potential back-end cache directories])], + [CACHEPATHS=${withval}], + [CACHEPATHS=$DEFAULT_LOC]) +AC_ARG_WITH(cachepath, + [[],[]], + [AC_MSG_ERROR(use --with-cachepaths=DIRS (plural) instead of --with-cachepath=DIR to specify one or more cache paths)], + []) +AC_ARG_WITH(commpath, + [AS_HELP_STRING([--with-compath=DIR],[Back-end directory for communication and housekeeping])], + [COMMPATH=${withval}], + [COMMPATH=$DEFAULT_LOC]) AC_ARG_WITH(default-local-prefix, [AS_HELP_STRING([--with-default-local-prefix=DIRS],[Colon-seperated list of directories that Spindle will not cache files out of])], [SPINDLE_LOCAL_PREFIX=${withval}], - [SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$SPINDLE_LOC"]) + [SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$COMMPATH"]) AC_DEFINE_UNQUOTED([SPINDLE_PORT],[$SPINDLE_PORT],[The default port for Spindle]) AC_DEFINE_UNQUOTED([NUM_COBO_PORTS],[$NUM_COBO_PORTS],[Number of ports for COBO to search for an open port]) AC_DEFINE_UNQUOTED([SPINDLE_MAX_PORT],[$(($SPINDLE_PORT + $NUM_COBO_PORTS - 1))],[The maximum port value]) -AC_DEFINE_UNQUOTED([SPINDLE_LOC],"[$SPINDLE_LOC]",[The default local directory for Spindle]) +AC_DEFINE_UNQUOTED([COMMPATH],"[$COMMPATH]",[Back-end directory for communication and housekeeping]) +AC_DEFINE_UNQUOTED([CACHEPATHS],"[$CACHEPATHS]",[Colon-separated list of potential back-end cache directories]) AC_DEFINE_UNQUOTED([SPINDLE_LOCAL_PREFIX],"[$SPINDLE_LOCAL_PREFIX]",[The default colon-separated list of directories that Spindle will not cache files out of]) TESTRM=unknown diff --git a/containers/spindle-flux-ubuntu/Dockerfile b/containers/spindle-flux-ubuntu/Dockerfile index 3af607a8..57badbfc 100644 --- a/containers/spindle-flux-ubuntu/Dockerfile +++ b/containers/spindle-flux-ubuntu/Dockerfile @@ -5,6 +5,8 @@ FROM fluxrm/flux-sched:${flux_sched_version} AS builder ARG replicas=4 ENV workers=${replicas} USER root +ENV TMPDIR=/tmp +RUN echo 'TMPDIR="/tmp"' >> /etc/environment RUN DEBIAN_FRONTEND="noninteractive" apt-get update \ && apt-get -qq install -y --no-install-recommends \ diff --git a/containers/spindle-flux-ubuntu/scripts/build_spindle.sh b/containers/spindle-flux-ubuntu/scripts/build_spindle.sh index 9257f85e..a63b5d8d 100755 --- a/containers/spindle-flux-ubuntu/scripts/build_spindle.sh +++ b/containers/spindle-flux-ubuntu/scripts/build_spindle.sh @@ -4,7 +4,7 @@ set -euxo pipefail mkdir -p /home/${USER}/Spindle-build cd /home/${USER}/Spindle-build -/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=flux --enable-flux-plugin --with-localstorage=/tmp CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" +/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=flux --enable-flux-plugin --with-cachepaths=/tmp/commpath/cachepath --with-commpath=/tmp/commpath CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" make -j$(nproc) make install diff --git a/containers/spindle-serial-ubuntu/Dockerfile b/containers/spindle-serial-ubuntu/Dockerfile index 3070596e..62c0cbf5 100644 --- a/containers/spindle-serial-ubuntu/Dockerfile +++ b/containers/spindle-serial-ubuntu/Dockerfile @@ -1,6 +1,8 @@ ARG ubuntu_version=noble FROM ubuntu:${ubuntu_version} USER root +ENV TMPDIR=/tmp +RUN echo 'TMPDIR="/tmp"' >> /etc/environment RUN DEBIAN_FRONTEND="noninteractive" apt-get update \ # install latest pkg utils: diff --git a/containers/spindle-serial-ubuntu/scripts/build_spindle.sh b/containers/spindle-serial-ubuntu/scripts/build_spindle.sh index 37b6491a..1022ba14 100755 --- a/containers/spindle-serial-ubuntu/scripts/build_spindle.sh +++ b/containers/spindle-serial-ubuntu/scripts/build_spindle.sh @@ -4,7 +4,7 @@ set -euxo pipefail mkdir -p /home/${USER}/Spindle-build cd /home/${USER}/Spindle-build -/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=serial --with-localstorage=/tmp CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" +/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=serial --with-cachepaths=/tmp/commpath/cachepath --with-commpath=/tmp/commpath CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" make -j$(nproc) make install diff --git a/containers/spindle-slurm-ubuntu/base/Dockerfile b/containers/spindle-slurm-ubuntu/base/Dockerfile index d4724276..d6eee9f2 100644 --- a/containers/spindle-slurm-ubuntu/base/Dockerfile +++ b/containers/spindle-slurm-ubuntu/base/Dockerfile @@ -1,6 +1,8 @@ ARG UBUNTU_VERSION=noble FROM ubuntu:${UBUNTU_VERSION} USER root +ENV TMPDIR=/tmp +RUN echo 'TMPDIR="/tmp"' >> /etc/environment RUN apt-get update \ && DEBIAN_FRONTEND="noninteractive" apt-get -qq install -y --no-install-recommends \ diff --git a/containers/spindle-slurm-ubuntu/testing-plugin/Dockerfile b/containers/spindle-slurm-ubuntu/testing-plugin/Dockerfile index 7b66a155..951480f1 100644 --- a/containers/spindle-slurm-ubuntu/testing-plugin/Dockerfile +++ b/containers/spindle-slurm-ubuntu/testing-plugin/Dockerfile @@ -2,6 +2,8 @@ ARG BASE_VERSION=latest FROM ghcr.io/llnl/spindle-slurm-base:${BASE_VERSION} ARG replicas=4 ENV workers=${replicas} +ENV TMPDIR=/tmp +RUN echo 'TMPDIR="/tmp"' >> /etc/environment ARG BUILD_ROOT=containers/spindle-slurm-ubuntu/testing-plugin diff --git a/containers/spindle-slurm-ubuntu/testing-plugin/scripts/build_spindle.sh b/containers/spindle-slurm-ubuntu/testing-plugin/scripts/build_spindle.sh index 2a252b32..2b36be90 100755 --- a/containers/spindle-slurm-ubuntu/testing-plugin/scripts/build_spindle.sh +++ b/containers/spindle-slurm-ubuntu/testing-plugin/scripts/build_spindle.sh @@ -3,7 +3,7 @@ set -euxo pipefail mkdir -p /home/${USER}/Spindle-build cd /home/${USER}/Spindle-build -/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm-plugin --enable-slurm-plugin --with-localstorage=/tmp CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" +/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm-plugin --enable-slurm-plugin --with-cachepaths=/tmp/commpath/cachepath --with-commpath=/tmp/commpath CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" make -j$(nproc) make install diff --git a/containers/spindle-slurm-ubuntu/testing-srun/scripts/build_spindle.sh b/containers/spindle-slurm-ubuntu/testing-srun/scripts/build_spindle.sh index acf1ef0b..125fe2eb 100755 --- a/containers/spindle-slurm-ubuntu/testing-srun/scripts/build_spindle.sh +++ b/containers/spindle-slurm-ubuntu/testing-srun/scripts/build_spindle.sh @@ -3,7 +3,7 @@ set -euxo pipefail mkdir -p /home/${USER}/Spindle-build cd /home/${USER}/Spindle-build -/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm --with-localstorage=/tmp CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" +/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm --with-cachepaths=/tmp/commpath/cachepath --with-commpath=/tmp/commpath CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" make -j$(nproc) make install diff --git a/containers/spindle-slurm-ubuntu/testing/Dockerfile b/containers/spindle-slurm-ubuntu/testing/Dockerfile index 91d6901c..99768535 100644 --- a/containers/spindle-slurm-ubuntu/testing/Dockerfile +++ b/containers/spindle-slurm-ubuntu/testing/Dockerfile @@ -2,6 +2,8 @@ ARG BASE_VERSION=latest FROM ghcr.io/llnl/spindle-slurm-base:${BASE_VERSION} ARG replicas=4 ENV workers=${replicas} +ENV TMPDIR=/tmp +RUN echo 'TMPDIR="/tmp"' >> /etc/environment ARG BUILD_ROOT=containers/spindle-slurm-ubuntu/testing diff --git a/containers/spindle-slurm-ubuntu/testing/scripts/build_spindle.sh b/containers/spindle-slurm-ubuntu/testing/scripts/build_spindle.sh index 6943e49a..7fcb48e3 100755 --- a/containers/spindle-slurm-ubuntu/testing/scripts/build_spindle.sh +++ b/containers/spindle-slurm-ubuntu/testing/scripts/build_spindle.sh @@ -3,7 +3,7 @@ set -euxo pipefail mkdir -p /home/${USER}/Spindle-build cd /home/${USER}/Spindle-build -/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm --with-rsh-launch --with-rsh-cmd=/usr/bin/ssh --with-localstorage=/tmp CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" +/home/${USER}/Spindle/configure --prefix=/home/${USER}/Spindle-inst --enable-sec-munge --with-rm=slurm --with-rsh-launch --with-rsh-cmd=/usr/bin/ssh --with-cachepaths=/tmp/commpath/cachepath --with-commpath=/tmp/commpath CFLAGS="-O2 -g" CXXFLAGS="-O2 -g" make -j$(nproc) make install diff --git a/src/client/beboot/spindle_bootstrap.c b/src/client/beboot/spindle_bootstrap.c index 9c82c061..98f91913 100644 --- a/src/client/beboot/spindle_bootstrap.c +++ b/src/client/beboot/spindle_bootstrap.c @@ -53,7 +53,7 @@ static int rankinfo[4]={-1,-1,-1,-1}; number_t number; static int use_cache; static unsigned int cachesize; -static char *location, *number_s, *orig_location, *symbolic_location; +static char *commpath, *number_s, *symbolic_commpath; static char **cmdline; static char *executable; static char *client_lib; @@ -92,7 +92,7 @@ extern char *realize(char *path); static int establish_connection() { debug_printf2("Opening connection to server\n"); - ldcsid = client_open_connection(location, number); + ldcsid = client_open_connection(commpath, number); if (ldcsid == -1) return -1; @@ -108,14 +108,18 @@ static void setup_environment() { char rankinfo_str[256]; snprintf(rankinfo_str, 256, "%d %d %d %d %d", ldcsid, rankinfo[0], rankinfo[1], rankinfo[2], rankinfo[3]); - + char *connection_str = NULL; if (opts & OPT_RELOCAOUT) connection_str = client_get_connection_string(ldcsid); + char *chosen_parsed_cachepath = NULL; + send_cachepath_query( ldcsid , NULL, &chosen_parsed_cachepath); + assert( chosen_parsed_cachepath ); + setenv("LD_AUDIT", client_lib, 1); - setenv("LDCS_LOCATION", location, 1); - setenv("LDCS_ORIG_LOCATION", orig_location, 1); + setenv("LDCS_COMMPATH", commpath, 1); + setenv("LDCS_CHOSEN_PARSED_CACHEPATH", chosen_parsed_cachepath, 1); setenv("LDCS_NUMBER", number_s, 1); setenv("LDCS_RANKINFO", rankinfo_str, 1); if (connection_str) @@ -161,7 +165,8 @@ static int parse_cmdline(int argc, char *argv[]) daemon_args[i - 3] = NULL; } - symbolic_location = argv[i++]; + symbolic_commpath = argv[i++]; + i++; // Skip over candidate_cachepaths. number_s = argv[i++]; number = (number_t) strtoul(number_s, NULL, 0); opts_s = argv[i++]; @@ -174,7 +179,7 @@ static int parse_cmdline(int argc, char *argv[]) return 0; } -static void launch_daemon(char *location) +static void launch_daemon(char *commpath) { /*grand-child fork, then execv daemon. By grand-child forking we ensure that the app won't get confused by seeing an unknown process as a child. */ @@ -184,12 +189,12 @@ static void launch_daemon(char *location) char unique_file[MAX_PATH_LEN+1]; char buffer[32]; - result = spindle_mkdir(location); + result = spindle_mkdir(commpath); if (result == -1) { debug_printf("Exiting due to spindle_mkdir error\n"); exit(-1); } - snprintf(unique_file, MAX_PATH_LEN, "%s/spindle_daemon_pid", location); + snprintf(unique_file, MAX_PATH_LEN, "%s/spindle_daemon_pid", commpath); unique_file[MAX_PATH_LEN] = '\0'; fd = open(unique_file, O_CREAT | O_EXCL | O_WRONLY, 0600); if (fd == -1) { @@ -344,14 +349,14 @@ int main(int argc, char *argv[]) } } - orig_location = parse_location(symbolic_location, number); - if (!orig_location) { + char *orig_commpath = parse_location(symbolic_commpath, number); + if (!orig_commpath) { return -1; } - location = realize(orig_location); + commpath = realize(orig_commpath); if (daemon_args) { - launch_daemon(location); + launch_daemon(commpath); } result = establish_connection(); @@ -375,7 +380,7 @@ int main(int argc, char *argv[]) #else shm_cache_limit = cachesize; #endif - shmcache_init(location, number, cachesize, shm_cache_limit); + shmcache_init(commpath, number, cachesize, shm_cache_limit); use_cache = 1; } diff --git a/src/client/client/client.c b/src/client/client/client.c index 0b899e7e..b30b96f2 100644 --- a/src/client/client/client.c +++ b/src/client/client/client.c @@ -41,6 +41,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include "exec_util.h" #include "intercept.h" #include "fixlocale.h" +#include "should_intercept.h" errno_location_t app_errno_location; @@ -70,11 +71,8 @@ static const ElfW(Phdr) *libc_phdrs, *interp_phdrs; static int num_libc_phdrs, num_interp_phdrs; ElfW(Addr) libc_loadoffset, interp_loadoffset; -/* location has the realize'd path to the local file cache. orig_location is not realized and - * may contain symlinks - */ -char *location; -char *orig_location; +static char *commpath; +char *chosen_realized_cachepath, *chosen_parsed_cachepath; number_t number; static int have_stat_patches; @@ -202,8 +200,7 @@ static int init_server_connection() if (!use_ldcs) return 0; - location = getenv("LDCS_LOCATION"); - orig_location = getenv("LDCS_ORIG_LOCATION"); + commpath = getenv("LDCS_COMMPATH"); number = (number_t) strtoul(getenv("LDCS_NUMBER"), NULL, 0); connection = getenv("LDCS_CONNECTION"); rankinfo_s = getenv("LDCS_RANKINFO"); @@ -212,9 +209,9 @@ static int init_server_connection() opts = strtoul(opts_s, NULL, 10); shm_cachesize = atoi(cachesize_s) * 1024; - if (strchr(location, '$')) { - location = parse_location(location, number); - if (!location) { + if (strchr(commpath, '$')) { + commpath = parse_location(commpath, number); + if (!commpath) { exit(-1); } } @@ -222,8 +219,7 @@ static int init_server_connection() if (!(opts & OPT_FOLLOWFORK)) { debug_printf("Disabling environment variables because we're not following forks\n"); unsetenv("LD_AUDIT"); - unsetenv("LDCS_LOCATION"); - unsetenv("LDCS_ORIG_LOCATION"); + unsetenv("LDCS_COMMPATH"); unsetenv("LDCS_NUMBER"); unsetenv("LDCS_CONNECTION"); unsetenv("LDCS_RANKINFO"); @@ -237,14 +233,14 @@ static int init_server_connection() #else shm_cache_limit = shm_cachesize; #endif - shmcache_init(location, number, shm_cachesize, shm_cache_limit); + shmcache_init(commpath, number, shm_cachesize, shm_cache_limit); } if (connection) { /* boostrapper established the connection for us. Reuse it. */ debug_printf("Recreating existing connection to server\n"); - debug_printf3("location = %s, number = %lu, connection = %s, rankinfo = %s\n", - location, (unsigned long) number, connection, rankinfo_s); + debug_printf3("commpath = %s, number = %lu, connection = %s, rankinfo = %s\n", + commpath, (unsigned long) number, connection, rankinfo_s); ldcsid = client_register_connection(connection); if (ldcsid == -1) return -1; @@ -254,20 +250,22 @@ static int init_server_connection() } else { /* Establish a new connection */ - debug_printf("open connection to ldcs %s %lu\n", location, (unsigned long) number); - ldcsid = client_open_connection(location, number); + debug_printf("open connection to ldcs %s %lu\n", commpath, (unsigned long) number); + ldcsid = client_open_connection(commpath, number); if (ldcsid == -1) return -1; send_pid(ldcsid); - send_location(ldcsid, location); + send_location(ldcsid, commpath); send_rankinfo_query(ldcsid, rankinfo+0, rankinfo+1, rankinfo+2, rankinfo+3); #if defined(LIBNUMA) if (opts & OPT_NUMA) send_cpu(ldcsid, get_cur_cpu()); #endif } - + send_cachepath_query( ldcsid, &chosen_realized_cachepath, &chosen_parsed_cachepath ); + assert( chosen_realized_cachepath ); + assert( chosen_parsed_cachepath ); snprintf(debugging_name, 32, "Client.%d", rankinfo[0]); LOGGING_INIT(debugging_name); @@ -475,7 +473,8 @@ char *client_library_load(const char *name) char *orig_file_name = (char *) name; if (is_in_spindle_cache(name)) { - debug_printf2("Library %s is in spindle cache (%s). Translating request\n", name, location); + assert( chosen_realized_cachepath ); + debug_printf2("Library %s is in spindle cache (%s). Translating request\n", name, chosen_realized_cachepath); memset(fixed_name, 0, MAX_PATH_LEN+1); send_orig_path_request(ldcsid, orig_file_name, fixed_name); orig_file_name = fixed_name; diff --git a/src/client/client/intercept_exec.c b/src/client/client/intercept_exec.c index dbf9bae8..edb0199a 100644 --- a/src/client/client/intercept_exec.c +++ b/src/client/client/intercept_exec.c @@ -141,8 +141,7 @@ static char **removeEnvironmentStrs(char **envp) continue; if (strIsPrefix("LD", envp[i])) { if (strIsPrefix("LD_AUDIT=", envp[i]) || - strIsPrefix("LDCS_LOCATION=", envp[i]) || - strIsPrefix("LDCS_ORIG_LOCATION=", envp[i]) || + strIsPrefix("LDCS_COMMPATH=", envp[i]) || strIsPrefix("LDCS_CONNECTION=", envp[i]) || strIsPrefix("LDCS_RANKINFO=", envp[i]) || strIsPrefix("LDCS_OPTIONS=", envp[i]) || @@ -176,8 +175,7 @@ static char **updateEnvironment(char **envp, int *num_modified, int propogate_sp unsetf = orig_unsetenv ? orig_unsetenv : unsetenv; unsetf("SPINDLE"); unsetf("LD_AUDIT"); - unsetf("LDCS_LOCATION"); - unsetf("LDCS_ORIG_LOCATION"); + unsetf("LDCS_COMMPATH"); unsetf("LDCS_CONNECTION"); unsetf("LDCS_RANKINFO"); unsetf("LDCS_OPTIONS"); @@ -198,13 +196,12 @@ static char **updateEnvironment(char **envp, int *num_modified, int propogate_sp if (envp) { debug_printf2("Propogating spindle environment by copying it to new envp list\n"); for (cur = (char **) envp; *cur; cur++, orig_size++); - new_size = orig_size + 10; + new_size = orig_size + 9; newenv = (char **) malloc(new_size * sizeof(char*)); propogateEnvironmentStr(envp, newenv, &pos, "SPINDLE"); propogateEnvironmentStr(envp, newenv, &pos, "LD_AUDIT"); - propogateEnvironmentStr(envp, newenv, &pos, "LDCS_LOCATION"); - propogateEnvironmentStr(envp, newenv, &pos, "LDCS_ORIG_LOCATION"); + propogateEnvironmentStr(envp, newenv, &pos, "LDCS_COMMPATH"); propogateEnvironmentStr(envp, newenv, &pos, "LDCS_CONNECTION"); propogateEnvironmentStr(envp, newenv, &pos, "LDCS_RANKINFO"); propogateEnvironmentStr(envp, newenv, &pos, "LDCS_OPTIONS"); diff --git a/src/client/client/intercept_readlink.c b/src/client/client/intercept_readlink.c index af4266a1..93fb879e 100644 --- a/src/client/client/intercept_readlink.c +++ b/src/client/client/intercept_readlink.c @@ -31,19 +31,20 @@ Place, Suite 330, Boston, MA 02111-1307 USA ssize_t (*orig_readlink)(const char *path, char *buf, size_t bufsiz); ssize_t (*orig_readlinkat)(int dirfd, const char *pathname, char *buf, size_t bufsiz); -extern char *location; static int fix_local_readlink(char *buf, size_t bufsiz) { char spindle_id[32]; - int location_len, result; + int cachepath_len, result; char tmp[MAX_PATH_LEN+1]; + extern char *chosen_realized_cachepath; - location_len = strlen(location); + assert( chosen_realized_cachepath ); + cachepath_len = strlen(chosen_realized_cachepath); snprintf(spindle_id, sizeof(spindle_id), "spindle.%lx", number); - if (strstr(buf, spindle_id) && strncmp(location, buf, location_len) == 0) { + if (strstr(buf, spindle_id) && strncmp(chosen_realized_cachepath, buf, cachepath_len) == 0) { debug_printf2("readlink received spindle cache path %s. Translating\n", buf); - result = send_orig_path_request(ldcsid, buf+location_len+1, tmp); + result = send_orig_path_request(ldcsid, buf+cachepath_len+1, tmp); if (result == -1) return -1; debug_printf2("readlink translated spindle local path %s to %s\n", buf, tmp); diff --git a/src/client/client/should_intercept.c b/src/client/client/should_intercept.c index bfabbbf8..89e4133d 100644 --- a/src/client/client/should_intercept.c +++ b/src/client/client/should_intercept.c @@ -30,21 +30,20 @@ extern int relocate_spindleapi(); -extern char *location; -extern char *orig_location; - int is_in_spindle_cache(const char *pathname) { - static int location_size = 0; - static int orig_location_size = 0; - if (!location_size) { - location_size = strlen(location); + static int cachepath_size = 0; + static int orig_cachepath_size = 0; + extern char *chosen_realized_cachepath, *chosen_parsed_cachepath; + assert( chosen_realized_cachepath ); + if (!cachepath_size) { + cachepath_size = strlen(chosen_realized_cachepath); } - if (!orig_location_size) { - orig_location_size = strlen(orig_location); + if (!orig_cachepath_size) { + orig_cachepath_size = strlen(chosen_parsed_cachepath); } - return ((strncmp(pathname, location, location_size) == 0) || - (strncmp(pathname, orig_location, orig_location_size) == 0)); + return ((strncmp(pathname, chosen_realized_cachepath, cachepath_size) == 0) || + (strncmp(pathname, chosen_parsed_cachepath, orig_cachepath_size) == 0)); } extern int is_local_prefix(const char *path, char **cached_local_prefixes); diff --git a/src/client/client_comlib/client_api.c b/src/client/client_comlib/client_api.c index 96390dca..6101a494 100644 --- a/src/client/client_comlib/client_api.c +++ b/src/client/client_comlib/client_api.c @@ -27,6 +27,7 @@ #include #include #include +#include #include "ldcs_api.h" #include "client_api.h" @@ -36,7 +37,56 @@ static struct lock_t comm_lock; #define COMM_LOCK do { if (lock(&comm_lock) == -1) return -1; } while (0) #define COMM_UNLOCK unlock(&comm_lock) - + + +int send_cachepath_query( int fd, char **chosen_realized_cachepath, char **chosen_parsed_cachepath){ + int retries = 0, max_retries = 10; + struct timespec one_second = { .tv_sec = 1, .tv_nsec = 0 }; + ldcs_message_t message; + char buffer[MAX_PATH_LEN+1]; + buffer[MAX_PATH_LEN] = '\0'; + + + do{ + message.header.type = LDCS_MSG_CHOSEN_CACHEPATH_REQUEST; + message.header.len = 0; + message.data = buffer; + + COMM_LOCK; + + debug_printf3("sending message of type: CHOSEN_CACHEPATH_REQUEST.\n" ); + client_send_msg(fd, &message); + client_recv_msg_static(fd, &message, LDCS_READ_BLOCK); + + COMM_UNLOCK; + + if( message.header.type == LDCS_MSG_NO_CACHEPATH_CONSENSUS_YET ){ + if( retries++ >= max_retries ){ + break; + } + nanosleep( &one_second, NULL ); + continue; + } + break; + + }while( 1 ); + + if (message.header.type != LDCS_MSG_CHOSEN_CACHEPATH || message.header.len > MAX_PATH_LEN) { + err_printf("Got unexpected message of type %d\n", (int) message.header.type); + return -1; + } + char *local_crc = strdup( buffer ); + char *local_cpc = strdup( &buffer[ strlen(local_crc) + 1 ] ); + if( chosen_realized_cachepath ){ + *chosen_realized_cachepath = local_crc; + } + if( chosen_parsed_cachepath ){ + *chosen_parsed_cachepath = local_cpc; + } + + return 0; +} + int send_file_query(int fd, char* path, int dso, char** newpath, int *errcode) { ldcs_message_t message; char buffer[MAX_PATH_LEN+1+sizeof(int)]; @@ -68,7 +118,7 @@ int send_file_query(int fd, char* path, int dso, char** newpath, int *errcode) { if (message.header.type != LDCS_MSG_FILE_QUERY_ANSWER) { err_printf("Got unexpected message of type %d\n", (int) message.header.type); - assert(0); + return -1; } if (message.header.len > sizeof(int)) { @@ -161,7 +211,7 @@ int send_existance_test(int fd, char *path, int *exists) if (message.header.type != LDCS_MSG_EXISTS_ANSWER || message.header.len != sizeof(uint32_t)) { err_printf("Got unexpected message after existance test: %d\n", (int) message.header.type); - assert(0); + return -1; } memcpy(exists, buffer, sizeof(*exists)); @@ -198,7 +248,7 @@ int send_orig_path_request(int fd, const char *path, char *newpath) if (message.header.type != LDCS_MSG_ORIGPATH_ANSWER || message.header.len > MAX_PATH_LEN) { err_printf("Got unexpected message after existance test: %d\n", (int) message.header.type); - assert(0); + return -1; } strncpy(newpath, buffer, MAX_PATH_LEN+1); @@ -315,7 +365,7 @@ int send_cpu(int fd, int cpu) { int send_location(int fd, char *location) { ldcs_message_t message; - message.header.type = LDCS_MSG_LOCATION; + message.header.type = LDCS_MSG_COMMPATH; message.header.len = strlen(location)+1; message.data = location; @@ -346,7 +396,7 @@ int send_ldso_info_request(int fd, const char *ldso_path, char *result_path) if (message.header.type != LDCS_MSG_LOADER_DATA_RESP) { err_printf("Got unexpected message after ldso req: %d\n", (int) message.header.type); - assert(0); + return -1; } return 0; } @@ -388,7 +438,7 @@ int send_rankinfo_query(int fd, int *mylrank, int *mylsize, int *mymdrank, int * if (message.header.type != LDCS_MSG_MYRANKINFO_QUERY_ANSWER || message.header.len != 4*sizeof(int)) { err_printf("Received incorrect response to rankinfo query %d\n", message.header.type); *mylrank = *mylsize = *mymdrank = *mymdsize = -1; - assert(0); + return -1; } p = (int *) message.data; @@ -423,7 +473,7 @@ int send_procmaps_query(int fd, int pid, char *result) if (message.header.type != LDCS_MSG_PROCMAPS_RESP) { err_printf("Received incorrect response to procmaps query %d\n", message.header.type); - assert(0); + return -1; } memcpy(result, buffer, MAX_PATH_LEN); @@ -454,7 +504,7 @@ int send_pickone_query(int fd, char *key, int *result) if (message.header.type != LDCS_MSG_PICKONE_RESP) { err_printf("Received incorrect response to procmaps query %d\n", message.header.type); - assert(0); + return -1; } *result = *((int *) message.data); diff --git a/src/client/client_comlib/client_api.h b/src/client/client_comlib/client_api.h index 74f82346..3d7c41be 100644 --- a/src/client/client_comlib/client_api.h +++ b/src/client/client_comlib/client_api.h @@ -42,6 +42,7 @@ int send_orig_path_request(int fd, const char *path, char *newpath); int send_dirlists_request(int fd, char **local_result, char **exece_result, char **to_free); int send_procmaps_query(int fd, int pid, char *result); int send_pickone_query(int fd, char *key, int *result); +int send_cachepath_query( int fd, char **chosen_symbolic_cachepath, char **chosen_parsed_cachepath); int get_python_prefix(int fd, char **prefix); diff --git a/src/client/config.h.in b/src/client/config.h.in index 2ddcacba..044ca9e1 100644 --- a/src/client/config.h.in +++ b/src/client/config.h.in @@ -6,6 +6,12 @@ /* Whether we are using a broken srun */ #undef BROKEN_SRUN +/* Colon-separated list of potential back-end cache directories */ +#undef CACHEPATHS + +/* Back-end directory for communication and housekeeping */ +#undef COMMPATH + /* Define if were using biter for client/server communication */ #undef COMM_BITER @@ -124,9 +130,6 @@ /* Default mode for slurm launch */ #undef SLURMLAUNCH_ENABLED -/* The default local directory for Spindle */ -#undef SPINDLE_LOC - /* The default colon-separated list of directories that Spindle will not cache files out of */ #undef SPINDLE_LOCAL_PREFIX diff --git a/src/client/configure b/src/client/configure index 8a015a42..eb84ee07 100755 --- a/src/client/configure +++ b/src/client/configure @@ -811,6 +811,9 @@ enable_maintainer_mode with_default_port with_default_num_ports with_localstorage +with_cachepaths +with_cachepath +with_commpath with_default_local_prefix with_testrm with_rm @@ -1532,7 +1535,13 @@ Optional Packages: --with-default-numports=NUM Number of TCP/IP ports to scan for Spindle server communication - --with-localstorage=DIR Directory on back-ends for storing relocated files + --with-localstorage=DIR (obsolete) + Use --with-cachepaths and --with-commpath instead. + --with-cachepaths=DIR Colon-separated list of potential back-end cache + directories +, + --with-compath=DIR Back-end directory for communication and + housekeeping --with-default-local-prefix=DIRS Colon-seperated list of directories that Spindle will not cache files out of @@ -12589,9 +12598,29 @@ fi # Check whether --with-localstorage was given. if test "${with_localstorage+set}" = set; then : - withval=$with_localstorage; SPINDLE_LOC=${withval} + withval=$with_localstorage; as_fn_error $? "requested obsolete option --with-localstorage. Use --with-cachepaths and --with-commpath instead." "$LINENO" 5 +fi + + +# Check whether --with-cachepaths was given. +if test "${with_cachepaths+set}" = set; then : + withval=$with_cachepaths; CACHEPATHS=${withval} +else + CACHEPATHS=$DEFAULT_LOC +fi + + +# Check whether --with-cachepath was given. +if test "${with_cachepath+set}" = set; then : + withval=$with_cachepath; as_fn_error $? "use --with-cachepaths=DIRS (plural) instead of --with-cachepath=DIR to specify one or more cache paths" "$LINENO" 5 +fi + + +# Check whether --with-commpath was given. +if test "${with_commpath+set}" = set; then : + withval=$with_commpath; COMMPATH=${withval} else - SPINDLE_LOC=$DEFAULT_LOC + COMMPATH=$DEFAULT_LOC fi @@ -12599,7 +12628,7 @@ fi if test "${with_default_local_prefix+set}" = set; then : withval=$with_default_local_prefix; SPINDLE_LOCAL_PREFIX=${withval} else - SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$SPINDLE_LOC" + SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$COMMPATH" fi @@ -12619,7 +12648,12 @@ _ACEOF cat >>confdefs.h <<_ACEOF -#define SPINDLE_LOC "$SPINDLE_LOC" +#define COMMPATH "$COMMPATH" +_ACEOF + + +cat >>confdefs.h <<_ACEOF +#define CACHEPATHS "$CACHEPATHS" _ACEOF diff --git a/src/cobo/cobo.c b/src/cobo/cobo.c index 785a5d03..2d1a53a9 100644 --- a/src/cobo/cobo.c +++ b/src/cobo/cobo.c @@ -160,7 +160,7 @@ static char* cobo_getenv(char* envvar, int type) char* str = getenv(envvar); if (str == NULL && type == ENV_REQUIRED) { err_printf("Missing required environment variable: %s\n", envvar); - exit(1); + return NULL; } return str; } @@ -171,7 +171,7 @@ static void* cobo_malloc(size_t n, char* msg) void* p = malloc(n); if (!p) { err_printf("Call to malloc(%lu) failed: %s (%m errno %d)\n", n, msg, errno); - exit(1); + return NULL; } return p; } @@ -513,7 +513,7 @@ static int cobo_connect_hostname(char* hostname, int rank) break; case HSHAKE_INTERNAL_ERROR: err_printf("Internal error doing handshake: %s", spindle_handshake_last_error_str()); - exit(-1); + return -1; break; case HSHAKE_DROP_CONNECTION: debug_printf3("Handshake said to drop connection\n"); @@ -768,7 +768,7 @@ static int cobo_open_tree() if (sockfd < 0) { err_printf("Creating parent socket (socket() %m errno=%d)\n", errno); - exit(1); + return -1; } setCloseOnExec(sockfd); @@ -817,7 +817,7 @@ static int cobo_open_tree() if (!port_is_bound) { /* TODO: would like to send an abort back to server */ err_printf("Failed to open socket on any port\n"); - exit(1); + return -1; } /* accept a connection from parent and receive socket table */ @@ -837,7 +837,7 @@ static int cobo_open_tree() break; case HSHAKE_INTERNAL_ERROR: err_printf("Internal error doing handshake: %s", spindle_handshake_last_error_str()); - exit(-1); + return -1; break; case HSHAKE_DROP_CONNECTION: debug_printf3("Handshake said to drop connection\n"); @@ -907,26 +907,26 @@ static int cobo_open_tree() /* read our rank number */ if (cobo_read_fd(cobo_parent_fd, &cobo_me, sizeof(int)) < 0) { err_printf("Receiving my rank from parent failed\n"); - exit(1); + return -1; } /* discover how many ranks are in our world */ if (cobo_read_fd(cobo_parent_fd, &cobo_nprocs, sizeof(int)) < 0) { err_printf("Receiving number of tasks from parent failed\n"); - exit(1); + return -1; } /* read the size of the hostlist (in bytes) */ if (cobo_read_fd(cobo_parent_fd, &cobo_hostlist_size, sizeof(int)) < 0) { err_printf("Receiving size of hostname table from parent failed\n"); - exit(1); + return -1; } /* allocate space for the hostlist and read it in */ cobo_hostlist = (void*) cobo_malloc(cobo_hostlist_size, "Hostlist data buffer"); if (cobo_read_fd(cobo_parent_fd, cobo_hostlist, cobo_hostlist_size) < 0) { err_printf("Receiving hostname table from parent failed\n"); - exit(1); + return -1; } /* @@ -969,7 +969,7 @@ static int cobo_open_tree() if (cobo_child_fd[i] == -1) { err_printf("Failed to connect to child (rank %d) on %s failed\n", c, child_hostname); - exit(1); + return -1; } /* tell child what rank he is and forward the hostname table to him */ @@ -978,7 +978,7 @@ static int cobo_open_tree() if (forward != COBO_SUCCESS) { err_printf("Failed to forward hostname table to child (rank %d) on %s failed\n", c, child_hostname); - exit(1); + return -1; } /* free the child hostname string */ @@ -1033,7 +1033,7 @@ static int cobo_bcast_tree(void* buf, int size) if (cobo_me != 0) { if (cobo_read_fd(cobo_parent_fd, buf, size) < 0) { err_printf("Receiving broadcast data from parent failed\n"); - exit(1); + return -1; } } @@ -1042,7 +1042,7 @@ static int cobo_bcast_tree(void* buf, int size) if (cobo_write_fd(cobo_child_fd[i], buf, size) < 0) { err_printf("Broadcasting data to child (rank %d) failed\n", cobo_child[i]); - exit(1); + return -1; } } @@ -1058,7 +1058,7 @@ int cobo_bcast_down(void* buf, int size) if (cobo_write_fd(cobo_child_fd[i], buf, size) < 0) { err_printf("Broadcasting data to child (rank %d) failed\n", cobo_child[i]); - exit(1); + return -1; } } return rc; @@ -1080,7 +1080,7 @@ static int cobo_allreduce_max_int_tree(int* sendbuf, int* recvbuf) if (cobo_read_fd(cobo_child_fd[i], &child_val, sizeof(child_val)) < 0) { err_printf("Reducing data from child (rank %d) failed\n", cobo_child[i]); - exit(1); + return -1; } /* compare child's max to our current max */ @@ -1094,7 +1094,7 @@ static int cobo_allreduce_max_int_tree(int* sendbuf, int* recvbuf) /* not the root, so forward our reduction result to our parent */ if (cobo_write_fd(cobo_parent_fd, &max_val, sizeof(max_val)) < 0) { err_printf("Sending reduced data to parent failed\n"); - exit(1); + return -1; } } else { /* we're the root, got the result, set the recvbuf */ @@ -1129,7 +1129,7 @@ static int cobo_gather_tree(void* sendbuf, int sendcount, void* recvbuf) if (cobo_read_fd(cobo_child_fd[i], (char*)bigbuf + offset, sendcount * cobo_child_incl[i]) < 0) { err_printf("Gathering data from child (rank %d) failed\n", cobo_child[i]); - exit(1); + return -1; } offset += sendcount * cobo_child_incl[i]; } @@ -1138,7 +1138,7 @@ static int cobo_gather_tree(void* sendbuf, int sendcount, void* recvbuf) if (cobo_me != 0) { if (cobo_write_fd(cobo_parent_fd, bigbuf, bigcount) < 0) { err_printf("Sending gathered data to parent failed\n"); - exit(1); + return -1; } cobo_free(bigbuf); } @@ -1158,7 +1158,7 @@ static int cobo_scatter_tree(void* sendbuf, int sendcount, void* recvbuf) bigbuf = (void*) cobo_malloc(bigcount, "Temporary scatter buffer in cobo_scatter_tree"); if (cobo_read_fd(cobo_parent_fd, bigbuf, bigcount) < 0) { err_printf("Receiving scatter data from parent failed\n"); - exit(1); + return -1; } } @@ -1169,7 +1169,7 @@ static int cobo_scatter_tree(void* sendbuf, int sendcount, void* recvbuf) if (cobo_write_fd(cobo_child_fd[i], (char*)bigbuf + offset, sendcount * cobo_child_incl[i]) < 0) { err_printf("Scattering data to child (rank %d) failed\n", cobo_child[i]); - exit(1); + return -1; } offset += sendcount * cobo_child_incl[i]; } @@ -1257,7 +1257,7 @@ int cobo_bcast(void* buf, int sendcount, int root) rc = cobo_bcast_tree(buf, sendcount); } else { err_printf("Cannot execute bcast from non-zero root\n"); - exit(1); + return -1; } cobo_gettimeofday(&end); @@ -1283,7 +1283,7 @@ int cobo_gather(void* sendbuf, int sendcount, void* recvbuf, int root) rc = cobo_gather_tree(sendbuf, sendcount, recvbuf); } else { err_printf("Cannot execute gather to non-zero root\n"); - exit(1); + return -1; } cobo_gettimeofday(&end); @@ -1309,7 +1309,7 @@ int cobo_scatter(void* sendbuf, int sendcount, void* recvbuf, int root) rc = cobo_scatter_tree(sendbuf, sendcount, recvbuf); } else { err_printf("Cannot execute scatter from non-zero root\n"); - exit(1); + return -1; } cobo_gettimeofday(&end); @@ -1354,7 +1354,7 @@ int cobo_alltoall(void* sendbuf, int sendcount, void* recvbuf) int rc = COBO_SUCCESS; err_printf("Cannot execute alltoall\n"); - exit(1); + return -1; cobo_gettimeofday(&end); debug_printf3("Exiting cobo_alltoall(), took %f seconds for %d procs\n", cobo_getsecs(&end,&start), cobo_nprocs); @@ -1434,6 +1434,48 @@ int cobo_allgather_str(char* sendstr, char*** recvstr, char** recvbuf) return COBO_SUCCESS; } +int cobo_allreduce( uint64_t *pval, cobo_op_t op ){ + + /* if i have any children, receive their data */ + uint64_t child_val; + for(int i=cobo_num_child-1; i>=0; i--) { + /* read int64_t from child */ + if (cobo_read_fd(cobo_child_fd[i], &child_val, sizeof(int64_t)) < 0) { + err_printf("Reducing data from child (rank %d) failed\n", cobo_child[i]); + return -1; + } + + /* compare child's val to our current val */ + switch( op ){ + case COBO_OP_MIN: if( child_val < *pval ) *pval = child_val; break; + case COBO_OP_MAX: if( child_val > *pval ) *pval = child_val; break; + case COBO_OP_BITWISE_AND: *pval &= child_val; break; + case COBO_OP_BITWISE_OR: *pval |= child_val; break; + case COBO_OP_LOGICAL_AND: *pval = *pval && child_val; break; + case COBO_OP_LOGICAL_OR: *pval = *pval || child_val; break; + case COBO_OP_SUM: *pval += child_val; break; + case COBO_OP_NOOP: break; + default: + err_printf("Illegal op (%d). Ignoring.\n", op); + break; + } + } + + /* forward data to parent if we're not rank 0, otherwise set the recvbuf */ + if (cobo_me != 0) { + /* not the root, so forward our reduction result to our parent */ + if (cobo_write_fd(cobo_parent_fd, pval, sizeof(*pval)) < 0) { + err_printf("Sending reduced data to parent failed\n"); + return -1; + } + } + + /* broadcast result of reduction from rank 0 to all tasks */ + cobo_bcast_tree(pval, sizeof(int64_t)); + + return COBO_SUCCESS; +} + /* provide list of ports and number of ports as input, get number of tasks and my rank as output */ int cobo_open(uint64_t sessionid, int* portlist, int num_ports, int* rank, int* num_ranks) { @@ -1482,7 +1524,7 @@ int cobo_open(uint64_t sessionid, int* portlist, int num_ports, int* rank, int* cobo_ports = cobo_int_dup(portlist, num_ports); if (cobo_ports == NULL) { err_printf("Failed to copy port list\n"); - exit(1); + return -1; } /* open the tree */ @@ -1491,7 +1533,7 @@ int cobo_open(uint64_t sessionid, int* portlist, int num_ports, int* rank, int* /* need to check that tree opened successfully before returning, so do a barrier */ if (cobo_barrier() != COBO_SUCCESS) { err_printf("Failed to open tree\n"); - exit(1); + return -1; } if (cobo_me == 0) { diff --git a/src/cobo/ldcs_cobo.h b/src/cobo/ldcs_cobo.h index edacd4b1..fafbda6a 100644 --- a/src/cobo/ldcs_cobo.h +++ b/src/cobo/ldcs_cobo.h @@ -55,6 +55,7 @@ extern "C" { #define cobo_allgather COMBINE(COBO_NAMESPACE, cobo_allgather) #define cobo_alltoall COMBINE(COBO_NAMESPACE, cobo_alltoall ) #define cobo_allgather_str COMBINE(COBO_NAMESPACE, cobo_allgather_str) +#define cobo_allreduce COMBINE(COBO_NAMESPACE, cobo_allreduce) #define cobo_server_open COMBINE(COBO_NAMESPACE, cobo_server_open) #define cobo_server_close COMBINE(COBO_NAMESPACE, cobo_server_close) #define cobo_server_get_root_socket COMBINE(COBO_NAMESPACE, cobo_server_get_root_socket) @@ -67,6 +68,19 @@ extern "C" { #define cobo_register_preconnect_cb COMBINE(COBO_NAMESPACE, cobo_register_preconnect_cb) #endif +// Used for cobo_allreduce(). +typedef enum{ + COBO_OP_MIN, + COBO_OP_MAX, + COBO_OP_BITWISE_AND, + COBO_OP_BITWISE_OR, + COBO_OP_LOGICAL_AND, + COBO_OP_LOGICAL_OR, + COBO_OP_SUM, + COBO_OP_NOOP, + NUM_COBO_OP +} cobo_op_t; + /* * ========================================================================== * ========================================================================== @@ -128,6 +142,8 @@ int cobo_alltoall (void* sendbuf, int sendcount, void* recvbuf); */ int cobo_allgather_str(char* sendstr, char*** recvstr, char** recvbuf); +int cobo_allreduce(uint64_t *pval, cobo_op_t op); + /* * ========================================================================== * ========================================================================== diff --git a/src/fe/comlib/Makefile.am b/src/fe/comlib/Makefile.am index 56a25ea2..2bb98236 100644 --- a/src/fe/comlib/Makefile.am +++ b/src/fe/comlib/Makefile.am @@ -1,5 +1,5 @@ noinst_LTLIBRARIES = libfe_cobo.la #noinst_LTLIBRARIES = libfe_msocket.la AM_CPPFLAGS = -I$(top_srcdir)/../logging -I$(top_srcdir)/../include -I$(top_srcdir)/../cobo -libfe_cobo_la_SOURCES = cobo_fe_comm.c +libfe_cobo_la_SOURCES = cobo_fe_comm.c ../../server/comlib/ldcs_api_util.c #libfe_msocket_la_SOURCES = msocket_fe_comm.c diff --git a/src/fe/comlib/Makefile.in b/src/fe/comlib/Makefile.in index ba02a246..b9f4599b 100644 --- a/src/fe/comlib/Makefile.in +++ b/src/fe/comlib/Makefile.in @@ -110,7 +110,9 @@ CONFIG_CLEAN_FILES = CONFIG_CLEAN_VPATH_FILES = LTLIBRARIES = $(noinst_LTLIBRARIES) libfe_cobo_la_LIBADD = -am_libfe_cobo_la_OBJECTS = cobo_fe_comm.lo +am__dirstamp = $(am__leading_dot)dirstamp +am_libfe_cobo_la_OBJECTS = cobo_fe_comm.lo \ + ../../server/comlib/ldcs_api_util.lo libfe_cobo_la_OBJECTS = $(am_libfe_cobo_la_OBJECTS) AM_V_lt = $(am__v_lt_@AM_V@) am__v_lt_ = $(am__v_lt_@AM_DEFAULT_V@) @@ -131,7 +133,8 @@ am__v_at_1 = DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir) depcomp = $(SHELL) $(top_srcdir)/../../scripts/depcomp am__maybe_remake_depfiles = depfiles -am__depfiles_remade = ./$(DEPDIR)/cobo_fe_comm.Plo +am__depfiles_remade = ../../server/comlib/$(DEPDIR)/ldcs_api_util.Plo \ + ./$(DEPDIR)/cobo_fe_comm.Plo am__mv = mv -f COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \ $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) @@ -332,7 +335,7 @@ top_srcdir = @top_srcdir@ noinst_LTLIBRARIES = libfe_cobo.la #noinst_LTLIBRARIES = libfe_msocket.la AM_CPPFLAGS = -I$(top_srcdir)/../logging -I$(top_srcdir)/../include -I$(top_srcdir)/../cobo -libfe_cobo_la_SOURCES = cobo_fe_comm.c +libfe_cobo_la_SOURCES = cobo_fe_comm.c ../../server/comlib/ldcs_api_util.c all: all-am .SUFFIXES: @@ -377,16 +380,28 @@ clean-noinstLTLIBRARIES: echo rm -f $${locs}; \ rm -f $${locs}; \ } +../../server/comlib/$(am__dirstamp): + @$(MKDIR_P) ../../server/comlib + @: > ../../server/comlib/$(am__dirstamp) +../../server/comlib/$(DEPDIR)/$(am__dirstamp): + @$(MKDIR_P) ../../server/comlib/$(DEPDIR) + @: > ../../server/comlib/$(DEPDIR)/$(am__dirstamp) +../../server/comlib/ldcs_api_util.lo: \ + ../../server/comlib/$(am__dirstamp) \ + ../../server/comlib/$(DEPDIR)/$(am__dirstamp) libfe_cobo.la: $(libfe_cobo_la_OBJECTS) $(libfe_cobo_la_DEPENDENCIES) $(EXTRA_libfe_cobo_la_DEPENDENCIES) $(AM_V_CCLD)$(LINK) $(libfe_cobo_la_OBJECTS) $(libfe_cobo_la_LIBADD) $(LIBS) mostlyclean-compile: -rm -f *.$(OBJEXT) + -rm -f ../../server/comlib/*.$(OBJEXT) + -rm -f ../../server/comlib/*.lo distclean-compile: -rm -f *.tab.c +@AMDEP_TRUE@@am__include@ @am__quote@../../server/comlib/$(DEPDIR)/ldcs_api_util.Plo@am__quote@ # am--include-marker @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/cobo_fe_comm.Plo@am__quote@ # am--include-marker $(am__depfiles_remade): @@ -424,6 +439,7 @@ mostlyclean-libtool: clean-libtool: -rm -rf .libs _libs + -rm -rf ../../server/comlib/.libs ../../server/comlib/_libs ID: $(am__tagged_files) $(am__define_uniq_tagged_files); mkid -fID $$unique @@ -540,6 +556,8 @@ clean-generic: distclean-generic: -test -z "$(CONFIG_CLEAN_FILES)" || rm -f $(CONFIG_CLEAN_FILES) -test . = "$(srcdir)" || test -z "$(CONFIG_CLEAN_VPATH_FILES)" || rm -f $(CONFIG_CLEAN_VPATH_FILES) + -rm -f ../../server/comlib/$(DEPDIR)/$(am__dirstamp) + -rm -f ../../server/comlib/$(am__dirstamp) maintainer-clean-generic: @echo "This command is intended for maintainers to use" @@ -550,7 +568,8 @@ clean-am: clean-generic clean-libtool clean-noinstLTLIBRARIES \ mostlyclean-am distclean: distclean-am - -rm -f ./$(DEPDIR)/cobo_fe_comm.Plo + -rm -f ../../server/comlib/$(DEPDIR)/ldcs_api_util.Plo + -rm -f ./$(DEPDIR)/cobo_fe_comm.Plo -rm -f Makefile distclean-am: clean-am distclean-compile distclean-generic \ distclean-tags @@ -596,7 +615,8 @@ install-ps-am: installcheck-am: maintainer-clean: maintainer-clean-am - -rm -f ./$(DEPDIR)/cobo_fe_comm.Plo + -rm -f ../../server/comlib/$(DEPDIR)/ldcs_api_util.Plo + -rm -f ./$(DEPDIR)/cobo_fe_comm.Plo -rm -f Makefile maintainer-clean-am: distclean-am maintainer-clean-generic diff --git a/src/fe/comlib/cobo_fe_comm.c b/src/fe/comlib/cobo_fe_comm.c index cef2f225..297fc1d3 100644 --- a/src/fe/comlib/cobo_fe_comm.c +++ b/src/fe/comlib/cobo_fe_comm.c @@ -187,7 +187,7 @@ int ldcs_audit_server_fe_broadcast(ldcs_message_t *msg, void *data) (void)data; int root_fd; - debug_printf("Broadcasting message to daemons\n"); + debug_printf("Broadcasting message %s to daemons\n",_message_type_to_str( msg->header.type ) ); cobo_server_get_root_socket(&root_fd); return write_msg(root_fd, msg); diff --git a/src/fe/config.h.in b/src/fe/config.h.in index 33ba031a..615997be 100644 --- a/src/fe/config.h.in +++ b/src/fe/config.h.in @@ -6,6 +6,12 @@ /* Whether we are using a broken srun */ #undef BROKEN_SRUN +/* Colon-separated list of potential back-end cache directories */ +#undef CACHEPATHS + +/* Back-end directory for communication and housekeeping */ +#undef COMMPATH + /* Define if were using biter for client/server communication */ #undef COMM_BITER @@ -169,9 +175,6 @@ /* Default mode for slurm launch */ #undef SLURMLAUNCH_ENABLED -/* The default local directory for Spindle */ -#undef SPINDLE_LOC - /* The default colon-separated list of directories that Spindle will not cache files out of */ #undef SPINDLE_LOCAL_PREFIX diff --git a/src/fe/configure b/src/fe/configure index 1018b37f..4c11bdc7 100755 --- a/src/fe/configure +++ b/src/fe/configure @@ -832,6 +832,9 @@ enable_maintainer_mode with_default_port with_default_num_ports with_localstorage +with_cachepaths +with_cachepath +with_commpath with_default_local_prefix with_testrm with_rm @@ -1570,7 +1573,13 @@ Optional Packages: --with-default-numports=NUM Number of TCP/IP ports to scan for Spindle server communication - --with-localstorage=DIR Directory on back-ends for storing relocated files + --with-localstorage=DIR (obsolete) + Use --with-cachepaths and --with-commpath instead. + --with-cachepaths=DIR Colon-separated list of potential back-end cache + directories +, + --with-compath=DIR Back-end directory for communication and + housekeeping --with-default-local-prefix=DIRS Colon-seperated list of directories that Spindle will not cache files out of @@ -16439,9 +16448,29 @@ fi # Check whether --with-localstorage was given. if test "${with_localstorage+set}" = set; then : - withval=$with_localstorage; SPINDLE_LOC=${withval} + withval=$with_localstorage; as_fn_error $? "requested obsolete option --with-localstorage. Use --with-cachepaths and --with-commpath instead." "$LINENO" 5 +fi + + +# Check whether --with-cachepaths was given. +if test "${with_cachepaths+set}" = set; then : + withval=$with_cachepaths; CACHEPATHS=${withval} +else + CACHEPATHS=$DEFAULT_LOC +fi + + +# Check whether --with-cachepath was given. +if test "${with_cachepath+set}" = set; then : + withval=$with_cachepath; as_fn_error $? "use --with-cachepaths=DIRS (plural) instead of --with-cachepath=DIR to specify one or more cache paths" "$LINENO" 5 +fi + + +# Check whether --with-commpath was given. +if test "${with_commpath+set}" = set; then : + withval=$with_commpath; COMMPATH=${withval} else - SPINDLE_LOC=$DEFAULT_LOC + COMMPATH=$DEFAULT_LOC fi @@ -16449,7 +16478,7 @@ fi if test "${with_default_local_prefix+set}" = set; then : withval=$with_default_local_prefix; SPINDLE_LOCAL_PREFIX=${withval} else - SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$SPINDLE_LOC" + SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$COMMPATH" fi @@ -16469,7 +16498,12 @@ _ACEOF cat >>confdefs.h <<_ACEOF -#define SPINDLE_LOC "$SPINDLE_LOC" +#define COMMPATH "$COMMPATH" +_ACEOF + + +cat >>confdefs.h <<_ACEOF +#define CACHEPATHS "$CACHEPATHS" _ACEOF diff --git a/src/fe/startup/config_mgr.cc b/src/fe/startup/config_mgr.cc index 36a30b20..1b9d9ade 100644 --- a/src/fe/startup/config_mgr.cc +++ b/src/fe/startup/config_mgr.cc @@ -50,10 +50,16 @@ using namespace std; #define SPINDLE_NUM_PORTS_STR "250" #endif -#if defined(SPINDLE_LOC) -#define SPINDLE_LOC_STR SPINDLE_LOC +#if defined(COMMPATH) +#define SPINDLE_COMMPATH_STR COMMPATH #else -#define SPINDLE_LOC_STR "$TMPDIR" +#define SPINDLE_COMMPATH_STR "$TMPDIR" +#endif + +#if defined(CACHEPATHS) +#define SPINDLE_CACHEPATHS_STR CACHEPATHS +#else +#define SPINDLE_CACHEPATHS_STR "$TMPDIR" #endif #if defined(SPINDLE_LOCAL_PREFIX) @@ -267,8 +273,10 @@ void initOptionsList() "Provides a text file containing a white-space separated list of files that should be relocated to each node before execution begins" }, { confStrip, "strip", shortStrip, groupMisc, cvBool, {}, "true", "Strip debug and symbol information from binaries before distributing them." }, - { confLocation, "location", shortLocation, groupMisc, cvString, {}, SPINDLE_LOC_STR, - "Back-end directory for storing relocated files. Should be a non-shared location such as a ramdisk." }, + { confCommPath, "commpath", shortCommPath, groupMisc, cvString, {}, SPINDLE_COMMPATH_STR, + "Back-end directory communication and housekeeping. Should be a non-shared location such as a ramdisk." }, + { confCachePaths, "cachepaths", shortCachePaths, groupMisc, cvString, {}, SPINDLE_CACHEPATHS_STR, + "Colon-separated list of candidate paths for cached libraries."}, { confNoclean, "noclean", shortNoClean, groupMisc, cvBool, {}, "false", "Don't remove local file cache after execution." }, { confDisableLogging, "disable-logging", shortDisableLogging, groupMisc, cvBool, {}, DISABLE_LOGGING_STR, @@ -735,9 +743,24 @@ bool ConfigMap::toSpindleArgs(spindle_args_t &args, bool alloc_strs) const case confNumPorts: args.num_ports = numresult; break; - case confLocation: { - string loc = strresult + "/spindle.$NUMBER"; - args.location = strdup(loc.c_str()); + case confCommPath: { + string path = strresult + "/spindle.$NUMBER"; + args.commpath = strdup(path.c_str()); + break; + } + case confCachePaths:{ + // Paramemter values are colon-separated lists of paths. + // Append "/spindle.$NUMBER" to each path in the list. + string paths = strresult; + size_t idx = paths.find(":"); + string number_var_with_colon("/spindle.$NUMBER:"); + string number_var_without_colon("/spindle.$NUMBER"); + while( idx != string::npos ){ + paths.replace(idx, 1, number_var_with_colon); + idx = paths.find(":", idx + number_var_with_colon.size()); + }; + paths += number_var_without_colon; + args.candidate_cachepaths = strdup(paths.c_str()); break; } case confCachePrefix: diff --git a/src/fe/startup/config_mgr.h b/src/fe/startup/config_mgr.h index 8e70daa6..ec3c8135 100644 --- a/src/fe/startup/config_mgr.h +++ b/src/fe/startup/config_mgr.h @@ -29,7 +29,8 @@ enum SpindleConfigID { confCmdlineNewgroup, confPort, confNumPorts, - confLocation, + confCommPath, + confCachePaths, confCachePrefix, confPythonPrefix, confLocalPrefix, @@ -82,7 +83,7 @@ enum CmdlineShortOptions { shortAuditType = 'k', shortRelocSO = 'l', shortNoClean = 'n', - shortLocation = 'o', + shortCommPath = 'o', shortPush = 'p', shortPull = 'q', shortPythonPrefix = 'r', @@ -125,7 +126,8 @@ enum CmdlineShortOptions { shortSpindleLevel = 296, shortLocalPrefix = 297, shortExecExcludes = 298, - shortPatchLdso + shortPatchLdso = 299, + shortCachePaths = 300, }; enum CmdlineGroups { diff --git a/src/fe/startup/parse_launcher.cc b/src/fe/startup/parse_launcher.cc index f05ba18a..4484e998 100644 --- a/src/fe/startup/parse_launcher.cc +++ b/src/fe/startup/parse_launcher.cc @@ -292,7 +292,7 @@ void ModifyArgv::modifyCmdLine() snprintf(options_str, 32, "%lu", (unsigned long) params->opts); string options(options_str); - string location(params->location); + string commpath(params->commpath); char number_str[32]; snprintf(number_str, 32, "%lu", (unsigned long) params->number); @@ -319,7 +319,7 @@ void ModifyArgv::modifyCmdLine() if (p == parser->appExecutableAt()) { #if defined(os_bluegene) string bg_env_str = parser->getParser()->getBGString(); - parser->getParser()->addBGEnvStr(n, new_argv, bg_env_str, default_libstr, intercept_libstr, location, number, options, shmcache_size); + parser->getParser()->addBGEnvStr(n, new_argv, bg_env_str, default_libstr, intercept_libstr, commpath, number, options, shmcache_size); #else char **a_argv; int a_argc; diff --git a/src/fe/startup/parse_launcher_args.cc b/src/fe/startup/parse_launcher_args.cc index 84cf7f23..710b009d 100644 --- a/src/fe/startup/parse_launcher_args.cc +++ b/src/fe/startup/parse_launcher_args.cc @@ -127,7 +127,7 @@ static cmdoption_t openmpi_options[] = { }; -static const char *openmpi_bg_env_str = "-x LD_AUDIT=%s -x LDCS_LOCATION=%s -x LDCS_NUMBER=%s -x LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *openmpi_bg_env_str = "-x LD_AUDIT=%s -x LDCS_COMMPATH=%s -x LDCS_NUMBER=%s -x LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; static const unsigned int openmpi_size = (sizeof(openmpi_options) / sizeof(cmdoption_t)); static cmdoption_t srun_options[] = { @@ -218,11 +218,11 @@ static cmdoption_t srun_options[] = { { NULL, "--usage", 0 }, { "-V", "--version", 0 } }; -static const char *srun_bg_env_str = "--runjob-opts=--envs LD_AUDIT=%s LD_PRELOAD=%s LDCS_LOCATION=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; -static const char *srun_bg_env_str_nopreload = "--runjob-opts=--envs LD_AUDIT=%s%s LDCS_LOCATION=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; -static const char *srun_bg_env_bare_str = "%s --envs LD_AUDIT=%s LD_PRELOAD=%s LDCS_LOCATION=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; -static const char *srun_bg_env_bare_str_preload = "%sLD_PRELOAD=%s:%s LD_AUDIT=%s LDCS_LOCATION=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; -static const char *srun_bg_env_bare_str_nopreload = "%s LD_AUDIT=%s LDCS_LOCATION=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *srun_bg_env_str = "--runjob-opts=--envs LD_AUDIT=%s LD_PRELOAD=%s LDCS_COMMPATH=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *srun_bg_env_str_nopreload = "--runjob-opts=--envs LD_AUDIT=%s%s LDCS_COMMPATH=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *srun_bg_env_bare_str = "%s --envs LD_AUDIT=%s LD_PRELOAD=%s LDCS_COMMPATH=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *srun_bg_env_bare_str_preload = "%sLD_PRELOAD=%s:%s LD_AUDIT=%s LDCS_COMMPATH=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; +static const char *srun_bg_env_bare_str_nopreload = "%s LD_AUDIT=%s LDCS_COMMPATH=%s LDCS_NUMBER=%s LDCS_OPTIONS=%s LDCS_CACHESIZE=%s"; static const unsigned int srun_size (sizeof(srun_options) / sizeof(cmdoption_t)); diff --git a/src/fe/startup/spindle_fe.cc b/src/fe/startup/spindle_fe.cc index cb53023b..a038e201 100644 --- a/src/fe/startup/spindle_fe.cc +++ b/src/fe/startup/spindle_fe.cc @@ -41,6 +41,7 @@ static const char *logging_file = NULL; #endif static const char spindle_bootstrap[] = LIBEXECDIR "/spindle_bootstrap"; static bool sendAndWaitForAlive(); +static void determineCachepathConsensus(); #define STARTUP_TIMEOUT 60 @@ -70,7 +71,8 @@ static int pack_data(spindle_args_t *args, void* &buffer, unsigned &buffer_size) buffer_size += sizeof(number_t); buffer_size += sizeof(opt_t); buffer_size += sizeof(unique_id_t); - buffer_size += args->location ? strlen(args->location) + 1 : 1; + buffer_size += args->commpath ? strlen(args->commpath) + 1 : 1; + buffer_size += args->candidate_cachepaths ? strlen(args->candidate_cachepaths) + 1 : 1; buffer_size += args->pythonprefix ? strlen(args->pythonprefix) + 1 : 1; buffer_size += args->preloadfile ? strlen(args->preloadfile) + 1 : 1; buffer_size += args->numa_files ? strlen(args->numa_files) + 1 : 1; @@ -90,7 +92,8 @@ static int pack_data(spindle_args_t *args, void* &buffer, unsigned &buffer_size) pack_param(args->use_launcher, buf, pos); pack_param(args->startup_type, buf, pos); pack_param(args->shm_cache_size, buf, pos); - pack_param(args->location, buf, pos); + pack_param(args->commpath, buf, pos); + pack_param(args->candidate_cachepaths, buf, pos); pack_param(args->pythonprefix, buf, pos); pack_param(args->preloadfile, buf, pos); pack_param(args->bundle_timeout_ms, buf, pos); @@ -229,7 +232,8 @@ int getApplicationArgsFE(spindle_args_t *params, int *spindle_argc, char ***spin (*spindle_argv)[n++] = strdup(numports_s); (*spindle_argv)[n++] = strdup(uniqueid_s); } - (*spindle_argv)[n++] = strdup(params->location); + (*spindle_argv)[n++] = strdup(params->commpath); + (*spindle_argv)[n++] = strdup(params->candidate_cachepaths); (*spindle_argv)[n++] = strdup(number_s); (*spindle_argv)[n++] = strdup(opt_s); (*spindle_argv)[n++] = strdup(cachesize_s); @@ -394,10 +398,12 @@ int spindleInitFE(const char **hosts, spindle_args_t *params) /* Start FE server */ debug_printf("spindle_args_t { number = %lu; port = %u; num_ports = %u; opts = %lu; unique_id = %lu; " - "use_launcher = %u; startup_type = %u; shm_cache_size = %u; location = %s; " + "use_launcher = %u; startup_type = %u; shm_cache_size = %u; commpath = %s; " + "cachepaths = %s; " "pythonprefix = %s; preloadfile = %s; bundle_timeout_ms = %u; bundle_cachesize_kb = %u }\n", (unsigned long) params->number, params->port, params->num_ports, params->opts, params->unique_id, - params->use_launcher, params->startup_type, params->shm_cache_size, params->location, + params->use_launcher, params->startup_type, params->shm_cache_size, params->commpath, + params->candidate_cachepaths, params->pythonprefix, params->preloadfile, params->bundle_timeout_ms, params->bundle_cachesize_kb); printSpindleFlags(params->opts); @@ -427,6 +433,7 @@ int spindleInitFE(const char **hosts, spindle_args_t *params) /* Wait for servers to indicate startup */ sendAndWaitForAlive(); + determineCachepathConsensus(); return 0; } @@ -483,6 +490,17 @@ void markRSHPidReapedFE() clear_fe_rsh_pid(); } +static void determineCachepathConsensus( void ){ + ldcs_message_t consensus_req_msg; + consensus_req_msg.header.type = LDCS_MSG_REQUEST_CACHEPATH_CONSENSUS; + consensus_req_msg.header.len = 0; + consensus_req_msg.data = NULL; + int result = ldcs_audit_server_fe_broadcast(&consensus_req_msg, NULL); + if (result == -1) { + debug_printf("Failure sending cachepath consensus message\n"); + } +} + static bool sendAndWaitForAlive() { int result; diff --git a/src/flux/flux-spindle.c b/src/flux/flux-spindle.c index bd2213f2..dd00aa32 100644 --- a/src/flux/flux-spindle.c +++ b/src/flux/flux-spindle.c @@ -381,8 +381,8 @@ static int sp_getopts (flux_shell_t *shell, struct spindle_ctx *ctx) int numa = 0; const char *relocaout = NULL, *reloclibs = NULL, *relocexec = NULL, *relocpython = NULL; const char *followfork = NULL, *preload = NULL, *level = NULL; - const char *pyprefix = NULL, *location = NULL; - char *numafiles = NULL; + const char *pyprefix = NULL, *commpath = NULL; + char *numafiles = NULL, *cachepaths = NULL; if (flux_shell_getopt_unpack (shell, "spindle", "o", &opts) < 0) return -1; @@ -404,7 +404,7 @@ static int sp_getopts (flux_shell_t *shell, struct spindle_ctx *ctx) * supplied by the user, but not unpacked (This handles typos, etc). */ if (json_unpack_ex (opts, &error, JSON_STRICT, - "{s?i s?i s?i s?i s?s s?s s?s s?s s?s s?s s?s s?i s?s s?s s?s}", + "{s?i s?i s?i s?i s?s s?s s?s s?s s?s s?s s?s s?i s?s s?s s?s s?s}", "noclean", &noclean, "nostrip", &nostrip, "push", &push, @@ -415,11 +415,12 @@ static int sp_getopts (flux_shell_t *shell, struct spindle_ctx *ctx) "reloc-exec", &relocexec, "reloc-python", &relocpython, "python-prefix", &pyprefix, - "location", &location, + "commpath", &commpath, "numa", &numa, "numa-files", &numafiles, "preload", &preload, - "level", &level) < 0) + "level", &level, + "cachepaths", &cachepaths) < 0) logerrno_printf_and_return(1, "Error in spindle option: %s\n", error.text); if (noclean) @@ -462,8 +463,11 @@ static int sp_getopts (flux_shell_t *shell, struct spindle_ctx *ctx) free (ctx->params.pythonprefix); ctx->params.pythonprefix = tmp; } - if (location) { - ctx->params.location = (char *) location; + if( cachepaths ){ + ctx->params.candidate_cachepaths = cachepaths; + } + if (commpath) { + ctx->params.commpath = (char *) commpath; } if (level) { if (strcmp(level, "high") == 0) { diff --git a/src/flux/sessionmgr.c b/src/flux/sessionmgr.c index 49324a2f..17027163 100644 --- a/src/flux/sessionmgr.c +++ b/src/flux/sessionmgr.c @@ -109,16 +109,16 @@ char **strip_start_from_argv(int argc, char **argv) extern char *parse_location(char *loc, int number); extern int spindle_mkdir(char *orig_path); -#if !defined(SPINDLE_LOC) -#error SPINDLE_LOC must be defined in config.h +#if !defined(COMMPATH) +#error COMMPATH must be defined in config.h #endif const char *get_session_dir() { int result; char *dir; - dir = parse_location((char *) (SPINDLE_LOC "/spindle_session"), 0); + dir = parse_location((char *) (COMMPATH "/spindle_session"), 0); if (!dir) { - spindle_debug_printf(1, "ERROR: Could not parse directory for spindle session location from %s/spindle_session\n", SPINDLE_LOC); + spindle_debug_printf(1, "ERROR: Could not parse directory for spindle session location from %s/spindle_session\n", COMMPATH); return NULL; } diff --git a/src/include/ldcs_api.h b/src/include/ldcs_api.h index e8ffa43d..0bcd8f40 100644 --- a/src/include/ldcs_api.h +++ b/src/include/ldcs_api.h @@ -45,7 +45,7 @@ typedef enum { LDCS_MSG_END, LDCS_MSG_CWD, LDCS_MSG_PID, - LDCS_MSG_LOCATION, + LDCS_MSG_COMMPATH, LDCS_MSG_CPU, LDCS_MSG_MYRANKINFO_QUERY, LDCS_MSG_MYRANKINFO_QUERY_ANSWER, @@ -85,6 +85,10 @@ typedef enum { LDCS_MSG_PICKONE_RESP, LDCS_MSG_ALIVE_REQ, LDCS_MSG_ALIVE_RESP, + LDCS_MSG_REQUEST_CACHEPATH_CONSENSUS, + LDCS_MSG_CHOSEN_CACHEPATH_REQUEST, + LDCS_MSG_CHOSEN_CACHEPATH, + LDCS_MSG_NO_CACHEPATH_CONSENSUS_YET, LDCS_MSG_UNKNOWN } ldcs_message_ids_t; diff --git a/src/include/spindle_launch.h b/src/include/spindle_launch.h index 47a4d92e..ca7b8d3d 100644 --- a/src/include/spindle_launch.h +++ b/src/include/spindle_launch.h @@ -123,8 +123,11 @@ typedef struct { /* Size of client shared memory cache */ unsigned int shm_cache_size; - /* The local-disk location where Spindle will store its cache */ - char *location; + /* The local-disk location for communication and housekeeping. */ + char *commpath; + + /* Path[s] for cached libraries. */ + char *candidate_cachepaths; /* Colon-separated list of candidate paths (max 64) */ /* Colon-seperated list of directories where Python is installed */ char *pythonprefix; diff --git a/src/logging/spindle_logd.cc b/src/logging/spindle_logd.cc index 0048170b..372b55de 100644 --- a/src/logging/spindle_logd.cc +++ b/src/logging/spindle_logd.cc @@ -202,7 +202,7 @@ class TestVerifier std::vector err_strings; std::set > target_libs; std::set > libs_loaded; - char *location; + char *cachepath; void logerror(std::string s) { @@ -249,7 +249,8 @@ class TestVerifier tmp_s = getenv("TEMPDIR"); if (!tmp_s) tmp_s = "/tmp"; - location = strdup(tmp_s); + // These are reasonable fallbacks that should be replaced via messages, below. + cachepath = strdup(tmp_s); } ~TestVerifier() @@ -269,7 +270,7 @@ class TestVerifier strstr(filename, "bin") == NULL && strstr(filename, ".py") == NULL) return true; - bool is_from_temp = (strstr(filename, location) != NULL) && (strncmp(filename, "/__not_exist", 12) != 0); + bool is_from_temp = (strstr(filename, cachepath) != NULL) && (strncmp(filename, "/__not_exist", 12) != 0); bool is_local_test = strstr(filename, "liblocal") != NULL; if (is_from_temp && !is_local_test && ret_code == -1) { @@ -295,12 +296,12 @@ class TestVerifier char buffer[4096]; int ret; - if (strstr(s, " location=" ) == s ){ - free( location ); - const char *loc_start = strstr( s, "=") + 1; - size_t loc_len = strlen( loc_start ); - location = strdup( loc_start ); - location[ loc_len - 1 ] = '\0'; // Remove trailing '\n'. + if (strstr(s, " cachepath=" ) == s ){ + free( cachepath ); + const char *cachepath_start = strstr( s, "=") + 1; + size_t cachepath_len = strlen( cachepath_start ); + cachepath = strdup( cachepath_start ); + cachepath[ cachepath_len - 1 ] = '\0'; // Remove trailing '\n'. } if (strstr(s, "open(") == s) { const char *first_quote, *last_quote, *equals; diff --git a/src/server/auditserver/cleanup_proc.cc b/src/server/auditserver/cleanup_proc.cc index c23a9f7d..8f0e7f6a 100644 --- a/src/server/auditserver/cleanup_proc.cc +++ b/src/server/auditserver/cleanup_proc.cc @@ -51,11 +51,17 @@ static bool longest_str_first(const string &a, const string &b) return a.size() > b.size(); } -static void rmDirSet(const set &dirs, const char *prefix_dir) +static void rmDirSet(const set &dirs, const char *cachepath, const char *commpath) { - string path_sep("/"); - size_t prefix_size = prefix_dir ? strlen(prefix_dir) : 0; - + string path_sep("/"); + if( !cachepath || !commpath ){ + // Should never happen. + err_printf( "cachepath (%s) and/or commpath (%s) is NULL. Unable to cleanup files.\n" ); + return; + } + size_t cachepath_len = strlen(cachepath); + size_t commpath_len = strlen(commpath); + for (set::const_iterator i = dirs.begin(); i != dirs.end(); i++) { DIR *dir = opendir(i->c_str()); if (!dir) @@ -71,8 +77,10 @@ static void rmDirSet(const set &dirs, const char *prefix_dir) if (dirs.find(componentpath) != dirs.end()) continue; - if (strncmp(prefix_dir, componentpath.c_str(), prefix_size) != 0) { - err_printf("Tried to clean a file %s that wasn't in our prefix %s\n", componentpath.c_str(), prefix_dir); + if ( (strncmp(cachepath, componentpath.c_str(), cachepath_len) != 0) && + (strncmp(commpath, componentpath.c_str(), commpath_len ) != 0) ){ + err_printf( "File for deletion (%s) is outside of cachepath (%s) and commpath (%s).\n", + componentpath.c_str(), cachepath, commpath ); continue; } unlink(componentpath.c_str()); @@ -82,26 +90,28 @@ static void rmDirSet(const set &dirs, const char *prefix_dir) vector ordered_dirs(dirs.begin(), dirs.end()); sort(ordered_dirs.begin(), ordered_dirs.end(), longest_str_first); for (vector::iterator i = ordered_dirs.begin(); i != ordered_dirs.end(); i++) { - if (strncmp(prefix_dir, i->c_str(), prefix_size) != 0) { - err_printf("Tried to rmdir directory %s that wasn't in our prefix %s\n", i->c_str(), prefix_dir); - continue; - } + if ( (strncmp(cachepath, i->c_str(), cachepath_len) != 0) && + (strncmp(commpath, i->c_str(), commpath_len) != 0) ){ + err_printf( "Directory for deletion (%s) is outside of cachepath (%s) and commpath (%s).\n", + i->c_str(), cachepath, commpath ); + continue; + } rmdir(i->c_str()); - } + } } class CleanupProc { - friend void init_cleanup_proc(const char *); + friend void init_cleanup_proc(const char *, const char *); private: set dirs; int write_dir_fd; int read_dir_fd; bool has_error; pid_t child_pid; - const char *prefix_dir; + const char *cachepath, *commpath; - CleanupProc(const char *pd); + CleanupProc(const char *cachepath, const char *commpath); void rmDirs(); void cleanupMain(); public: @@ -110,11 +120,12 @@ class CleanupProc bool hadError(); }; -CleanupProc::CleanupProc(const char *pd) : +CleanupProc::CleanupProc(const char *cachepath, const char *commpath) : write_dir_fd(-1), read_dir_fd(-1), has_error(false), - prefix_dir(pd) + cachepath(cachepath), + commpath(commpath) { int fds[2]; int result; @@ -185,7 +196,7 @@ bool CleanupProc::hadError() void CleanupProc::rmDirs() { - rmDirSet(dirs, prefix_dir); + rmDirSet(dirs, cachepath, commpath); } void CleanupProc::cleanupMain() @@ -240,10 +251,10 @@ void CleanupProc::addDir(const char *dir) static CleanupProc *proc = NULL; static set local_dircache; -void init_cleanup_proc(const char *location_dir) +void init_cleanup_proc(const char *cachepath, const char *commpath) { assert(!proc); - proc = new CleanupProc(location_dir); + proc = new CleanupProc(cachepath, commpath); if (proc->hadError()) { delete proc; proc = NULL; @@ -269,13 +280,13 @@ int lookup_prev_mkdir(const char *dir) return (i != local_dircache.end()) ? 1 : 0; } -void cleanup_created_dirs(const char *prefix_dir) +void cleanup_created_dirs(const char *cachepath, const char *commpath) { if (proc) { proc->triggerCleanup(); } else { debug_printf("Cleaning files with local unlink/rmdirs.\n"); - rmDirSet(local_dircache, prefix_dir); - } + rmDirSet(local_dircache, cachepath, commpath); + } } diff --git a/src/server/auditserver/cleanup_proc.h b/src/server/auditserver/cleanup_proc.h index f5f71123..ab8bb939 100644 --- a/src/server/auditserver/cleanup_proc.h +++ b/src/server/auditserver/cleanup_proc.h @@ -18,11 +18,11 @@ Place, Suite 330, Boston, MA 02111-1307 USA extern "C" { #endif -void init_cleanup_proc(const char *location_dir); +void init_cleanup_proc(const char *cachepath, const char *commpath); void track_mkdir(const char *dir); -int lookup_prev_mkdir(const char *dir); -void cleanup_created_dirs(const char *location_dir); - +int lookup_prev_mkdir(const char *dir); +void cleanup_created_dirs(const char *cachepath, const char *commpath); + #if defined(__cplusplus) } #endif diff --git a/src/server/auditserver/ldcs_audit_server_filemngt.c b/src/server/auditserver/ldcs_audit_server_filemngt.c index bb2e9294..1d0f410b 100644 --- a/src/server/auditserver/ldcs_audit_server_filemngt.c +++ b/src/server/auditserver/ldcs_audit_server_filemngt.c @@ -45,9 +45,9 @@ Place, Suite 330, Boston, MA 02111-1307 USA #error LIBEXECDIR must be defined #endif -char *_ldcs_audit_server_tmpdir; +char *_ldcs_audit_server_cachepath; static char *normalized_tmpdir; - +static char *_ldcs_audit_server_commpath; extern int spindle_mkdir(char *path); static char *filemngt_normalize_dir(char *dir) { @@ -55,25 +55,26 @@ static char *filemngt_normalize_dir(char *dir) { return newpath ? newpath : dir; } -int ldcs_audit_server_filemngt_init (char* location) { +int ldcs_audit_server_filemngt_init (char *cachepath, char *commpath) { int rc=0; - _ldcs_audit_server_tmpdir = location; - if (-1 == spindle_mkdir(_ldcs_audit_server_tmpdir)) { - err_printf("mkdir: ERROR during mkdir %s\n", _ldcs_audit_server_tmpdir); + _ldcs_audit_server_cachepath = cachepath; + _ldcs_audit_server_commpath = commpath; + if (-1 == spindle_mkdir(_ldcs_audit_server_cachepath)) { + err_printf("mkdir: ERROR during mkdir %s\n", _ldcs_audit_server_cachepath); _error("mkdir failed"); } - normalized_tmpdir = filemngt_normalize_dir(location); + normalized_tmpdir = filemngt_normalize_dir(cachepath); return(rc); } /* Returns NULL if not a cached file. Otherwise, returns pointer to global portion of string */ char* ldcs_is_a_cachedfile (char* filename) { - int len = strlen(_ldcs_audit_server_tmpdir); + int len = strlen(_ldcs_audit_server_cachepath); int norm_len = strlen(normalized_tmpdir); - if ( strncmp(_ldcs_audit_server_tmpdir, filename, len) == 0 ) + if ( strncmp(_ldcs_audit_server_cachepath, filename, len) == 0 ) return filename + len + 1; if ( strncmp(normalized_tmpdir, filename, norm_len) == 0 ) return filename + norm_len + 1; @@ -105,7 +106,7 @@ char *filemngt_calc_localname(char *global_name, calc_local_t reqtype) size_t dirpart_size, filepart_size; int cut_dirpart_slash; - lastslash = strrchr(_ldcs_audit_server_tmpdir, '/'); + lastslash = strrchr(_ldcs_audit_server_cachepath, '/'); if (lastslash && lastslash[1] == '\0') endslash = ""; else @@ -153,10 +154,10 @@ char *filemngt_calc_localname(char *global_name, calc_local_t reqtype) cut_dirpart_slash = (dirpart[0] == '/') ? 1 : 0; - snprintf(target, sizeof(target), "%s%s%s", _ldcs_audit_server_tmpdir, endslash, dirpart+cut_dirpart_slash); + snprintf(target, sizeof(target), "%s%s%s", _ldcs_audit_server_cachepath, endslash, dirpart+cut_dirpart_slash); spindle_mkdir(target); - snprintf(target, sizeof(target), "%s%s%s/%s", _ldcs_audit_server_tmpdir, endslash, dirpart+cut_dirpart_slash, filepart); + snprintf(target, sizeof(target), "%s%s%s/%s", _ldcs_audit_server_cachepath, endslash, dirpart+cut_dirpart_slash, filepart); GCC7_ENABLE_WARNING; @@ -287,7 +288,7 @@ int filemngt_decode_packet(node_peer_t peer, ldcs_message_t *msg, char *filename **/ int ldcs_audit_server_filemngt_clean() { - cleanup_created_dirs(_ldcs_audit_server_tmpdir); + cleanup_created_dirs(_ldcs_audit_server_cachepath, _ldcs_audit_server_commpath); return 0; } @@ -829,7 +830,7 @@ int filemngt_convert_proc_maps(int pid, char *new_maps_filename, int new_maps_fi int result; debug_printf2("Asked to convert /proc/%d/maps to remove spindle paths\n", pid); - result = translate_proc_pid_maps(_ldcs_audit_server_tmpdir, pid, new_maps_filename, new_maps_filename_size); + result = translate_proc_pid_maps(_ldcs_audit_server_cachepath, pid, new_maps_filename, new_maps_filename_size); if (result == -1) { new_maps_filename[0] = '\0'; return -1; diff --git a/src/server/auditserver/ldcs_audit_server_filemngt.h b/src/server/auditserver/ldcs_audit_server_filemngt.h index 6f45c485..592b9c82 100644 --- a/src/server/auditserver/ldcs_audit_server_filemngt.h +++ b/src/server/auditserver/ldcs_audit_server_filemngt.h @@ -23,7 +23,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include "ldcs_audit_server_md.h" -int ldcs_audit_server_filemngt_init (char* location); +int ldcs_audit_server_filemngt_init (char *cachepath, char *commpath); int filemngt_read_file(char *filename, void *buffer, size_t *size, int strip, int *err, int *was_stripped); int filemngt_encode_packet(char *filename, void *filecontents, size_t filesize, diff --git a/src/server/auditserver/ldcs_audit_server_handlers.c b/src/server/auditserver/ldcs_audit_server_handlers.c index 2791057c..67d2a04b 100644 --- a/src/server/auditserver/ldcs_audit_server_handlers.c +++ b/src/server/auditserver/ldcs_audit_server_handlers.c @@ -21,6 +21,9 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include #include #include +#include +#include +#include #include "ldcs_api.h" #include "ldcs_api_listen.h" @@ -178,7 +181,10 @@ static int handle_setup_alias(ldcs_process_data_t *procdata, char *pathname, cha static int handle_client_dirlists_req(ldcs_process_data_t *procdata, int nc); static int handle_close_client_query(ldcs_process_data_t *procdata, int nc); static int handle_alive_msg(ldcs_process_data_t *procdata, ldcs_message_t *msg); +static int handle_cachepath_consensus(ldcs_process_data_t *procdata, ldcs_message_t *msg); +static int handle_chosen_cachepath_request(ldcs_process_data_t *procdata, int nc); +extern void getValidCachePathByIndex( uint64_t validBitIdx, char **realizedCachePath, char **parsedCachePath, char **symbolicCachePath ); /** * Query from client to server. Returns info about client's rank in server data structures. **/ @@ -196,10 +202,10 @@ static int handle_client_info_msg(ldcs_process_data_t *procdata, int nc, ldcs_me client->remote_pid=mypid; debug_printf2("Server recvd pid %d from %d\n", mypid, nc); } - else if(msg->header.type == LDCS_MSG_LOCATION) { + else if(msg->header.type == LDCS_MSG_COMMPATH) { strncpy(client->remote_location, msg->data, sizeof(client->remote_location)-1); client->remote_location[sizeof(client->remote_location)-1] = '\0'; - debug_printf2("Server recvd location %s from %d\n", msg->data, nc); + debug_printf2("Server recvd remote_location %s from %d\n", msg->data, nc); } else if (msg->header.type == LDCS_MSG_CPU) { int clientcpu; @@ -1865,7 +1871,7 @@ int handle_client_message(ldcs_process_data_t *procdata, int nc, ldcs_message_t switch (msg->header.type) { case LDCS_MSG_CWD: case LDCS_MSG_PID: - case LDCS_MSG_LOCATION: + case LDCS_MSG_COMMPATH: case LDCS_MSG_CPU: return handle_client_info_msg(procdata, nc, msg); case LDCS_MSG_PYTHONPREFIX_REQ: @@ -1892,6 +1898,8 @@ int handle_client_message(ldcs_process_data_t *procdata, int nc, ldcs_message_t return handle_client_pickone_msg(procdata, nc, msg); case LDCS_MSG_END: return handle_client_end(procdata, nc); + case LDCS_MSG_CHOSEN_CACHEPATH_REQUEST: + return handle_chosen_cachepath_request(procdata, nc); default: err_printf("Received unexpected message from client %d: %d\n", nc, (int) msg->header.type); assert(0); @@ -1989,6 +1997,8 @@ int handle_server_message(ldcs_process_data_t *procdata, node_peer_t peer, ldcs_ case LDCS_MSG_ALIVE_REQ: case LDCS_MSG_ALIVE_RESP: return handle_alive_msg(procdata, msg); + case LDCS_MSG_REQUEST_CACHEPATH_CONSENSUS: + return handle_cachepath_consensus(procdata, msg); default: err_printf("Received unexpected message from node: %d\n", (int) msg->header.type); assert(0); @@ -2288,7 +2298,7 @@ static int handle_client_fileexist_msg(ldcs_process_data_t *procdata, int nc, ld return handle_client_progress(procdata, nc); } -extern char *_ldcs_audit_server_tmpdir; +extern char *_ldcs_audit_server_cachepath; static int handle_client_origpath_msg(ldcs_process_data_t *procdata, int nc, ldcs_message_t *msg) { ldcs_client_t *client; @@ -2303,7 +2313,7 @@ static int handle_client_origpath_msg(ldcs_process_data_t *procdata, int nc, ldc lookuppath[MAX_PATH_LEN] = '\0'; if (*origpath != '/' && *origpath != '.') - snprintf(lookuppath, MAX_PATH_LEN, "%s/%s", _ldcs_audit_server_tmpdir, origpath); + snprintf(lookuppath, MAX_PATH_LEN, "%s/%s", _ldcs_audit_server_cachepath, origpath); else strncpy(lookuppath, origpath, MAX_PATH_LEN); @@ -2950,6 +2960,93 @@ static int handle_client_pickone_msg(ldcs_process_data_t *procdata, int nc, ldcs } } +/** + * Handle LDCS_MSG_REQUEST_CACHEPATH_CONSENSUS to determine which cachepaths are + * available across all of the servers. + */ +static int cachepath_consensus_reached; +static int handle_cachepath_consensus(ldcs_process_data_t *procdata, ldcs_message_t *msg){ + + int num_children = ldcs_audit_server_md_get_num_children(procdata); + + debug_printf( "Processing REQUEST_CACHEPATH_CONSENSUS.\n" ); + debug_printf( " procdata->cachepath_bitidx = %#"PRIx64"\n", procdata->cachepath_bitidx ); + debug_printf( " procdata->cachepaths = %s\n", procdata->cachepaths ); + debug_printf( " procdata->cachepath = %s [should be null]\n", procdata->cachepath ); + debug_printf( " procdata->commpath = %s\n", procdata->commpath ); + debug_printf( " num_children = %d\n", num_children ); + + if (num_children) { + spindle_broadcast(procdata, msg); + debug_printf( "Successfully broadcast REQUEST_CACHEPATH_CONSENSUS\n" ); + msgbundle_force_flush(procdata); + debug_printf( "Successfully flushed the broadcast of REQUEST_CACHEPATH_CONSENSUS\n" ); + } + + ldcs_audit_server_md_allreduce_AND( &procdata->cachepath_bitidx ); + debug_printf( "The consensus value for procdata->cachepath_bitidx is: %#"PRIx64"\n", procdata->cachepath_bitidx ); + + if( procdata->cachepath_bitidx == 0 ){ + err_printf("No valid cachepath path available. Falling back to \"commpath\" path (%s).\n", procdata->commpath); + procdata->cachepath = procdata->commpath; + }else{ + getValidCachePathByIndex( procdata->cachepath_bitidx, + &procdata->cachepath, + &procdata->parsed_cachepath, + &procdata->symbolic_cachepath); + debug_printf( "The consensus cachepath is: %s\n", procdata->cachepath ); + debug_printf( "The consensus parsed_cachepath is: %s\n", procdata->parsed_cachepath ); + debug_printf( "The consensus symbolic_cachepath is: %s\n", procdata->symbolic_cachepath ); + debug_printf( "The commpath is: %s\n", procdata->commpath ); + } + + debug_printf( "Arrived at cachepath consensus: %s. Now delaying to flush race condition.\n", procdata->cachepath ); + debug_printf( "Delay completed.\n"); + + debug_printf3("Initializing file cache cachepath %s\n", procdata->cachepath); + ldcs_audit_server_filemngt_init(procdata->cachepath, procdata->commpath); + + test_printf(" cachepath=%s\n", procdata->cachepath); + cachepath_consensus_reached = 1; + return 0; +} + +/** + * Handle LDCS_MSG_CHOSEN_CACHEPATH_REQUEST + */ +static int handle_chosen_cachepath_request(ldcs_process_data_t *procdata, int nc){ + ldcs_message_t msg; + int connid; + ldcs_client_t *client; + + assert(nc != -1); + client = procdata->client_table + nc; + connid = client->connid; + if (client->state != LDCS_CLIENT_STATUS_ACTIVE || connid < 0) + return 0; + + + if( cachepath_consensus_reached ){ + msg.header.type = LDCS_MSG_CHOSEN_CACHEPATH; + msg.header.len = strlen(procdata->cachepath) + 1 + strlen(procdata->parsed_cachepath) + 1; + msg.data = calloc( 1, msg.header.len ); + strcpy( msg.data, procdata->cachepath ); + strcpy( &msg.data[ strlen(procdata->cachepath)+1 ], procdata->parsed_cachepath ); + }else{ + msg.header.type = LDCS_MSG_NO_CACHEPATH_CONSENSUS_YET; + msg.header.len = 0; + msg.data = NULL; + } + + ldcs_send_msg(connid, &msg); + free( msg.data ); + procdata->server_stat.clientmsg.cnt++; + procdata->server_stat.clientmsg.time += ldcs_get_time() - client->query_arrival_time; + + return 0; +} + + /** * Handle alive message, which is a broadcast/response ping through all servers */ @@ -3158,7 +3255,7 @@ int exit_note_cb(int fd, int serverid, void *data) eresult = -1; } - result = handleExitNote(fd, procdata->location); + result = handleExitNote(fd, procdata->commpath); if (result == -1) { debug_printf("handleExitNote failed\n"); eresult = -1; diff --git a/src/server/auditserver/ldcs_audit_server_md.h b/src/server/auditserver/ldcs_audit_server_md.h index eb5bf9f6..a4640370 100644 --- a/src/server/auditserver/ldcs_audit_server_md.h +++ b/src/server/auditserver/ldcs_audit_server_md.h @@ -107,6 +107,9 @@ int ldcs_audit_server_md_broadcast_noncontig(ldcs_process_data_t *ldcs_process_d int ldcs_audit_server_md_get_num_children(ldcs_process_data_t *procdata); int ldcs_audit_server_md_is_parent(node_peer_t peer); + +void ldcs_audit_server_md_allreduce_AND( uint64_t *val ); + #if defined(__cplusplus) } diff --git a/src/server/auditserver/ldcs_audit_server_md_cobo.c b/src/server/auditserver/ldcs_audit_server_md_cobo.c index 08c9b952..27393a55 100644 --- a/src/server/auditserver/ldcs_audit_server_md_cobo.c +++ b/src/server/auditserver/ldcs_audit_server_md_cobo.c @@ -401,3 +401,7 @@ int ldcs_audit_server_md_get_num_children(ldcs_process_data_t *procdata) cobo_get_num_childs(&num_childs); return num_childs; } + +void ldcs_audit_server_md_allreduce_AND( uint64_t *val ){ + cobo_allreduce( val, COBO_OP_BITWISE_AND ); +} diff --git a/src/server/auditserver/ldcs_audit_server_md_msocket.c b/src/server/auditserver/ldcs_audit_server_md_msocket.c index c9d616c2..5cfe9f2d 100644 --- a/src/server/auditserver/ldcs_audit_server_md_msocket.c +++ b/src/server/auditserver/ldcs_audit_server_md_msocket.c @@ -62,7 +62,6 @@ int ldcs_audit_server_md_init ( ldcs_process_data_t *ldcs_process_data ) { int rc=0; char* ldcs_nportsstr=getenv("LDCS_NPORTS"); - char* ldcs_locmodstr=getenv("LDCS_LOCATION_MOD"); int usedport; int serverfd, serverid, i; @@ -128,21 +127,6 @@ int ldcs_audit_server_md_init ( ldcs_process_data_t *ldcs_process_data ) { ldcs_listen_unregister_fd(serverfd); - - if(ldcs_locmodstr) { - int ldcs_locmod=atoi(ldcs_locmodstr); - if(ldcs_locmod>0) { - char buffer[MAX_PATH_LEN]; - debug_printf3("multiple server per node add modifier to location mod=%d\n",ldcs_locmod); - if(strlen(ldcs_process_data->location)+10location,ldcs_process_data->md_rank%ldcs_locmod); - debug_printf3("change location to %s (locmod=%d)\n",buffer,ldcs_locmod); - free(ldcs_process_data->location); - ldcs_process_data->location=strdup(buffer); - } else _error("location path too long"); - } - } - return(rc); } diff --git a/src/server/auditserver/ldcs_audit_server_process.c b/src/server/auditserver/ldcs_audit_server_process.c index 312095be..b6ecdbff 100644 --- a/src/server/auditserver/ldcs_audit_server_process.c +++ b/src/server/auditserver/ldcs_audit_server_process.c @@ -37,6 +37,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include "msgbundle.h" #include "exitnote.h" #include "cleanup_proc.h" +#include "parseloc.h" //#define GPERFTOOLS #if defined(GPERFTOOLS) @@ -112,7 +113,7 @@ void startprofile(spindle_args_t *args) char hostname[257]; char *home = getenv("HOME"); if (!home || !*home) - home = ldcs_process_data.location; + home = ldcs_process_data.commpath; gethostname(hostname, sizeof(hostname)); snprintf(filename, 4096, "%s/spindled.%lu.%s.%d.prof", home, (unsigned long) args->number, hostname, getpid()); ProfilerStart(filename); @@ -141,7 +142,10 @@ int ldcs_audit_server_process(spindle_args_t *args) startprofile(args); debug_printf3("Initializing server data structures\n"); - ldcs_process_data.location = args->location; + ldcs_process_data.commpath = args->commpath; + ldcs_process_data.cachepaths = args->candidate_cachepaths; + ldcs_process_data.cachepath = NULL; + ldcs_process_data.cachepath_bitidx = 0; ldcs_process_data.number = args->number; ldcs_process_data.pythonprefix = args->pythonprefix; ldcs_process_data.localprefix = args->local_prefixes; @@ -191,14 +195,12 @@ int ldcs_audit_server_process(spindle_args_t *args) } ldcs_process_data.server_stat.hostname=ldcs_process_data.hostname; - debug_printf3("Initializing file cache location %s\n", ldcs_process_data.location); - ldcs_audit_server_filemngt_init(ldcs_process_data.location); if (ldcs_process_data.opts & OPT_PROCCLEAN) - init_cleanup_proc(ldcs_process_data.location); + init_cleanup_proc(ldcs_process_data.cachepath, ldcs_process_data.commpath); debug_printf3("Initializing connections for clients at %s and %lu\n", - ldcs_process_data.location, (unsigned long) ldcs_process_data.number); - serverid = ldcs_create_server(ldcs_process_data.location, ldcs_process_data.number); + ldcs_process_data.commpath, (unsigned long) ldcs_process_data.number); + serverid = ldcs_create_server(ldcs_process_data.commpath, ldcs_process_data.number); if (serverid == -1) { err_printf("Unable to setup area for client connections\n"); return -1; @@ -214,7 +216,7 @@ int ldcs_audit_server_process(spindle_args_t *args) ldcs_listen_register_fd(fd, serverid, &_ldcs_server_CB, (void *) &ldcs_process_data); if (args->opts & OPT_BEEXIT) { - fd = createExitNote(args->location); + fd = createExitNote(args->commpath); if (fd != -1) { ldcs_listen_register_fd(fd, serverid, exit_note_cb, (void *) &ldcs_process_data); } @@ -229,6 +231,10 @@ int ldcs_audit_server_process(spindle_args_t *args) if (fd != -1) { ldcs_listen_register_fd(fd, serverid, forceExitCB, (void *) &ldcs_process_data); } + determineValidCachePaths( + &ldcs_process_data.cachepath_bitidx, + ldcs_process_data.cachepaths, + ldcs_process_data.number ); return 0; } @@ -248,7 +254,7 @@ int ldcs_audit_server_run() _ldcs_server_stat_print(&ldcs_process_data.server_stat); - debug_printf("destroy server (%s,%lu)\n", ldcs_process_data.location, (unsigned long) ldcs_process_data.number); + debug_printf("destroy server (%s,%lu)\n", ldcs_process_data.commpath, (unsigned long) ldcs_process_data.number); ldcs_destroy_server(ldcs_process_data.serverid); /* destroy md support (multi-daemon) */ diff --git a/src/server/auditserver/ldcs_audit_server_process.h b/src/server/auditserver/ldcs_audit_server_process.h index 18b3320a..82b60023 100644 --- a/src/server/auditserver/ldcs_audit_server_process.h +++ b/src/server/auditserver/ldcs_audit_server_process.h @@ -125,7 +125,15 @@ struct ldcs_process_data_struct int exit_readys_recvd; ldcs_dist_model_t dist_model; ldcs_client_t* client_table; - char *location; + char *commpath; /* Single user-specified path for fifo, daemons, etc. */ + /* (Everything except the cachepath.) */ + char *cachepaths; /* Up to 64 colon-separated list of candidate cachepaths. */ + char *cachepath; /* The earliest path in the list available to all servers. */ + /* (Environment variables replaced, symbolic links realized.) */ + char *symbolic_cachepath; /* The original representation of the cachepath. */ + char *parsed_cachepath; /* The cachepath with environment variables replaced. */ + /* (Symbolic links, if any, remain.) */ + uint64_t cachepath_bitidx; /* Bit index of valid cachepaths on a given server. */ char *hostname; char *pythonprefix; char *localprefix; diff --git a/src/server/comlib/ldcs_api_util.c b/src/server/comlib/ldcs_api_util.c index 2bc2455d..af101815 100644 --- a/src/server/comlib/ldcs_api_util.c +++ b/src/server/comlib/ldcs_api_util.c @@ -51,7 +51,7 @@ char* _message_type_to_str (ldcs_message_ids_t type) { STR_CASE(LDCS_MSG_END); STR_CASE(LDCS_MSG_CWD); STR_CASE(LDCS_MSG_PID); - STR_CASE(LDCS_MSG_LOCATION); + STR_CASE(LDCS_MSG_COMMPATH); STR_CASE(LDCS_MSG_CPU); STR_CASE(LDCS_MSG_MYRANKINFO_QUERY); STR_CASE(LDCS_MSG_MYRANKINFO_QUERY_ANSWER); @@ -91,6 +91,10 @@ char* _message_type_to_str (ldcs_message_ids_t type) { STR_CASE(LDCS_MSG_PICKONE_RESP); STR_CASE(LDCS_MSG_ALIVE_REQ); STR_CASE(LDCS_MSG_ALIVE_RESP); + STR_CASE(LDCS_MSG_REQUEST_CACHEPATH_CONSENSUS); + STR_CASE(LDCS_MSG_CHOSEN_CACHEPATH_REQUEST); + STR_CASE(LDCS_MSG_CHOSEN_CACHEPATH); + STR_CASE(LDCS_MSG_NO_CACHEPATH_CONSENSUS_YET); STR_CASE(LDCS_MSG_UNKNOWN); } return "unknown"; diff --git a/src/server/config.h.in b/src/server/config.h.in index 8d1842cf..24040627 100644 --- a/src/server/config.h.in +++ b/src/server/config.h.in @@ -6,6 +6,12 @@ /* Whether we are using a broken srun */ #undef BROKEN_SRUN +/* Colon-separated list of potential back-end cache directories */ +#undef CACHEPATHS + +/* Back-end directory for communication and housekeeping */ +#undef COMMPATH + /* Define if were using biter for client/server communication */ #undef COMM_BITER @@ -151,9 +157,6 @@ /* Default mode for slurm launch */ #undef SLURMLAUNCH_ENABLED -/* The default local directory for Spindle */ -#undef SPINDLE_LOC - /* The default colon-separated list of directories that Spindle will not cache files out of */ #undef SPINDLE_LOCAL_PREFIX diff --git a/src/server/configure b/src/server/configure index 81b43b17..83bff908 100755 --- a/src/server/configure +++ b/src/server/configure @@ -838,6 +838,9 @@ enable_maintainer_mode with_default_port with_default_num_ports with_localstorage +with_cachepaths +with_cachepath +with_commpath with_default_local_prefix with_testrm with_rm @@ -1567,7 +1570,13 @@ Optional Packages: --with-default-numports=NUM Number of TCP/IP ports to scan for Spindle server communication - --with-localstorage=DIR Directory on back-ends for storing relocated files + --with-localstorage=DIR (obsolete) + Use --with-cachepaths and --with-commpath instead. + --with-cachepaths=DIR Colon-separated list of potential back-end cache + directories +, + --with-compath=DIR Back-end directory for communication and + housekeeping --with-default-local-prefix=DIRS Colon-seperated list of directories that Spindle will not cache files out of @@ -16436,9 +16445,29 @@ fi # Check whether --with-localstorage was given. if test "${with_localstorage+set}" = set; then : - withval=$with_localstorage; SPINDLE_LOC=${withval} + withval=$with_localstorage; as_fn_error $? "requested obsolete option --with-localstorage. Use --with-cachepaths and --with-commpath instead." "$LINENO" 5 +fi + + +# Check whether --with-cachepaths was given. +if test "${with_cachepaths+set}" = set; then : + withval=$with_cachepaths; CACHEPATHS=${withval} +else + CACHEPATHS=$DEFAULT_LOC +fi + + +# Check whether --with-cachepath was given. +if test "${with_cachepath+set}" = set; then : + withval=$with_cachepath; as_fn_error $? "use --with-cachepaths=DIRS (plural) instead of --with-cachepath=DIR to specify one or more cache paths" "$LINENO" 5 +fi + + +# Check whether --with-commpath was given. +if test "${with_commpath+set}" = set; then : + withval=$with_commpath; COMMPATH=${withval} else - SPINDLE_LOC=$DEFAULT_LOC + COMMPATH=$DEFAULT_LOC fi @@ -16446,7 +16475,7 @@ fi if test "${with_default_local_prefix+set}" = set; then : withval=$with_default_local_prefix; SPINDLE_LOCAL_PREFIX=${withval} else - SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$SPINDLE_LOC" + SPINDLE_LOCAL_PREFIX="$DEFAULT_LOCAL_PREFIX:$COMMPATH" fi @@ -16466,7 +16495,12 @@ _ACEOF cat >>confdefs.h <<_ACEOF -#define SPINDLE_LOC "$SPINDLE_LOC" +#define COMMPATH "$COMMPATH" +_ACEOF + + +cat >>confdefs.h <<_ACEOF +#define CACHEPATHS "$CACHEPATHS" _ACEOF diff --git a/src/server/startup/spindle_be.cc b/src/server/startup/spindle_be.cc index 7493c020..d5a234a1 100644 --- a/src/server/startup/spindle_be.cc +++ b/src/server/startup/spindle_be.cc @@ -58,7 +58,8 @@ static int unpack_data(spindle_args_t *args, void *buffer, int buffer_size) unpack_param(args->use_launcher, buf, pos); unpack_param(args->startup_type, buf, pos); unpack_param(args->shm_cache_size, buf, pos); - unpack_param(args->location, buf, pos); + unpack_param(args->commpath, buf, pos); + unpack_param(args->candidate_cachepaths, buf, pos); unpack_param(args->pythonprefix, buf, pos); unpack_param(args->preloadfile, buf, pos); unpack_param(args->bundle_timeout_ms, buf, pos); @@ -70,7 +71,7 @@ static int unpack_data(spindle_args_t *args, void *buffer, int buffer_size) unpack_param(args->session_key, buf, pos); unpack_param(args->exec_excludes, buf, pos); assert(pos == buffer_size); - + assert( args->candidate_cachepaths ); return 0; } @@ -135,6 +136,8 @@ int spindleRunBE(unsigned int port, unsigned int num_ports, unique_id_t unique_i result = ldcs_audit_server_network_setup(port, num_ports, unique_id, &setup_data, &setup_data_size); if (result == -1) { err_printf("Error setting up network in spindleRunBE\n"); + if (args.startup_type == startup_external) + LOGGING_FINI; return -1; } unpack_data(&args, setup_data, setup_data_size); @@ -143,20 +146,23 @@ int spindleRunBE(unsigned int port, unsigned int num_ports, unique_id_t unique_i assert(args.port == port); - /* Expand environment variables in location. */ - char *new_location = parse_location(args.location, args.number); - if (!new_location) { - err_printf("Failed to convert location %s\n", args.location); + /* Expand environment variables in commpath. */ + char *new_commpath = parse_location(args.commpath, args.number); + if (!new_commpath) { + err_printf("Failed to convert commpath %s\n", args.commpath); + if (args.startup_type == startup_external) + LOGGING_FINI; return -1; } - debug_printf("Translated location from %s to %s\n", args.location, new_location); - free(args.location); - args.location = new_location; - test_printf(" location=%s\n", args.location); + debug_printf("Translated commpath from %s to %s\n", args.commpath, new_commpath); + free(args.commpath); + args.commpath = strdup(new_commpath); result = ldcs_audit_server_process(&args); if (result == -1) { err_printf("Error in ldcs_audit_server_process\n"); + if (args.startup_type == startup_external) + LOGGING_FINI; return -1; } @@ -164,6 +170,8 @@ int spindleRunBE(unsigned int port, unsigned int num_ports, unique_id_t unique_i result = post_setup(&args); if (result == -1) { err_printf("post_setup callback errored. Returning\n"); + if (args.startup_type == startup_external) + LOGGING_FINI; return -1; } } @@ -172,11 +180,12 @@ int spindleRunBE(unsigned int port, unsigned int num_ports, unique_id_t unique_i ldcs_audit_server_run(); if (result == -1) { err_printf("Error in ldcs_audit_server_process\n"); + if (args.startup_type == startup_external) + LOGGING_FINI; return -1; } - - if (args.startup_type == startup_external) + if (args.startup_type == startup_external) LOGGING_FINI; return 0; diff --git a/src/slurm_plugin/plugin_utils.c b/src/slurm_plugin/plugin_utils.c index 9a00e111..f77531e4 100644 --- a/src/slurm_plugin/plugin_utils.c +++ b/src/slurm_plugin/plugin_utils.c @@ -301,10 +301,8 @@ int isFEHost(char **hostlist, unsigned int num_hosts) static char* locSpecificDir(spindle_args_t *params) { char *dir = NULL, *expanded_dir = NULL, *realized_dir = NULL; - char hostname[256], session_id_str[32]; - size_t unique_file_len; - - dir = params->location; + + dir = params->commpath; if (!dir) { sdprintf(1, "ERROR: Location not filled in\n"); goto done; @@ -858,7 +856,7 @@ int registerFEPid(pid_t pid, spindle_args_t *args) int fd; int result; - snprintf(pid_file, sizeof(pid_file), "%s/fepid", args->location); + snprintf(pid_file, sizeof(pid_file), "%s/fepid", args->commpath); pid_file[sizeof(pid_file)-1] = '\0'; snprintf(pid_s, sizeof(pid_s), "%d\n", (int) pid); @@ -889,7 +887,7 @@ int readFEPid(pid_t *pid, spindle_args_t *args) pid_t pid_result; int fd, result; - snprintf(pid_file, sizeof(pid_file), "%s/fepid", args->location); + snprintf(pid_file, sizeof(pid_file), "%s/fepid", args->commpath); pid_file[sizeof(pid_file)-1] = '\0'; sdprintf(2, "Reading FE pid from %s\n", pid_file); diff --git a/src/slurm_plugin/slurm_plugin.c b/src/slurm_plugin/slurm_plugin.c index f2fedfeb..0989299c 100644 --- a/src/slurm_plugin/slurm_plugin.c +++ b/src/slurm_plugin/slurm_plugin.c @@ -104,6 +104,7 @@ static int enable_spindle = 0; static int start_session = 0; extern char *parse_location(char *loc, number_t number); +extern char *realize(char *path); // CLI options for srun struct spank_option spank_options[] = @@ -703,7 +704,7 @@ static unique_id_t getUniqueID(spank_t spank, int session_enabled) static int fillInArgs(spank_t spank, spindle_args_t *args, int argc, char **argv, unique_id_t unique_id, int session_enabled) { int result; - char *oldlocation; + char *symbolic_commpath, *orig_commpath; char *err_string; args->unique_id = unique_id; @@ -730,10 +731,15 @@ static int fillInArgs(spank_t spank, spindle_args_t *args, int argc, char **argv args->opts |= OPT_BEEXIT; } - oldlocation = args->location; + + symbolic_commpath = args->commpath; + orig_commpath = parse_location(symbolic_commpath, args->number); + if( !orig_commpath ){ + return -1; + } + args->commpath = realize(orig_commpath); + current_spank = spank; - args->location = parse_location(oldlocation, args->number); - free(oldlocation); return 0; } @@ -1261,8 +1267,8 @@ static int handleExit(void *params, char **output_str) return 0; } - if (!args.location) { - sdprintf(2, "WARNING: spindleExitBE not called since location is NULL\n"); + if (!args.commpath) { + sdprintf(2, "WARNING: spindleExitBE not called since commpath is NULL\n"); } else { // The task_exit callback is run for _each proc_, so we use // isBEProc to pick only one proc per node to call spindleExitBE. @@ -1279,9 +1285,9 @@ static int handleExit(void *params, char **output_str) return 0; } } else { - result = spindleExitBE(args.location); + result = spindleExitBE(args.commpath); if (result == -1) { - sdprintf(1, "ERROR: spindleExitBE returned an error on location %s\n", args.location); + sdprintf(1, "ERROR: spindleExitBE returned an error on commpath %s\n", args.commpath); return -1; } } diff --git a/src/utils/parseloc.c b/src/utils/parseloc.c index f0ebae30..b14cab00 100644 --- a/src/utils/parseloc.c +++ b/src/utils/parseloc.c @@ -22,6 +22,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include #include #include +#include #if !defined(USE_PLUGIN_DEBUG) #include "spindle_debug.h" @@ -34,13 +35,7 @@ Place, Suite 330, Boston, MA 02111-1307 USA #include "ccwarns.h" #include "spindle_launch.h" -#if defined(__cplusplus) -extern "C" { -#endif - char *parse_location(char *loc, number_t number); -#if defined(__cplusplus) -} -#endif +extern int spindle_mkdir(char *orig_path); #if defined(CUSTOM_GETENV) extern char *custom_getenv(char*); @@ -168,38 +163,64 @@ char *parse_location_noerr(char *loc, number_t number) **/ char *realize(char *path) { + int local_errno; char *result; - char *origpath, *cur_slash = NULL, *trailing; - struct stat buf; + char *origpath, *cur_slash = NULL, *prev_slash = NULL; + struct stat *buf = calloc( 1, sizeof( struct stat ) ); char newpath[MAX_PATH_LEN+1]; int lastpos; newpath[MAX_PATH_LEN] = '\0'; origpath = strdup(path); - for (;;) { - if (stat(origpath, &buf) != -1) - break; - if (cur_slash) - *cur_slash = '/'; + errno=0; + while( stat( origpath, buf ) == -1 ){ + local_errno = errno; + debug_printf("Failed to stat '%s' (%s).\n", origpath, strerror(local_errno)); + prev_slash = cur_slash; cur_slash = strrchr(origpath, '/'); - if (!cur_slash) - break; - *cur_slash = '\0'; + if( prev_slash ) + *prev_slash = '/'; + if( cur_slash ) + *cur_slash = '\0'; + else{ + debug_printf("Nothing in the original path can be stat'ed. (%s)\n", path); + return NULL; + } } - if (cur_slash) - trailing = cur_slash + 1; - else - trailing = ""; + errno = 0; result = realpath(origpath, newpath); if (!result) { + local_errno = errno; + err_printf( + "Error: realpath(3) failed to create canonical version of '%s' (%s). Returning '%s'.\n", + origpath, strerror(local_errno), path ); + errno = 0; + int rc = stat( origpath, buf ); + local_errno = errno; + err_printf( + " Statting that path results in rc=%d, errno=%d, error='%s'.\n", + rc, local_errno, strerror(local_errno)); free(origpath); - return path; + return NULL; } + free(buf); - strncat(newpath, "/", MAX_PATH_LEN); - strncat(newpath, trailing, MAX_PATH_LEN); - newpath[MAX_PATH_LEN] = '\0'; + if( cur_slash ){ + if( strlen( newpath ) + strlen( cur_slash+1 ) > MAX_PATH_LEN ){ + err_printf( + "Error: The realized path exceeds MAX_PATH_LEN (%d).\n" + " Original path: '%s'\n" + " Statable part: '%s'\n" + " Canonical version: '%s'\n" + " Returning original path.\n", + MAX_PATH_LEN, path, origpath, newpath); + free(origpath); + return path; + } + strncat(newpath, "/", 2); + strncat(newpath, cur_slash+1, MAX_PATH_LEN - strlen( newpath )); + } free(origpath); lastpos = strlen(newpath)-1; @@ -280,3 +301,84 @@ int is_local_prefix(const char *path, char **local_prefixes) { return 0; } +/* validateCandidatePath determines if candidatePath passes parse_location(), realize(), and spindle_mkdir(), which is to say, can + * spindle create a directory from this path? + * + * If not NULL, then realizedPath, parsedPath, and/or symbolicPath will hold the respective intermediate/final results. + * + * Return 1 if the candidatePath is valid, otherwise 0. + */ +static int validateCandidatePath( char *candidatePath, char **realizedPath, char **parsedPath, char **symbolicPath, number_t number ){ + int rc; + char *parsedCandidatePath, *realizedCandidatePath; + parsedCandidatePath = parse_location( candidatePath, number ); + if( parsedCandidatePath ){ + realizedCandidatePath = realize( parsedCandidatePath ); + if( realizedCandidatePath ){ + rc = spindle_mkdir( parsedCandidatePath ); + if( 0 == rc ){ + // candidatePath is going to be freed in the calling function. + // symbolicPath needs a strdup(). parsedPath() and realizedPath() + // allocate their own memory for strings right now; the extra + // strdup()s are just in case a future implementation decides + // to modify the string passed instead of returning a new one. + if( symbolicPath) *symbolicPath = strdup(candidatePath); + if( parsedPath ) *parsedPath = strdup(parsedCandidatePath); + if( realizedPath) *realizedPath = strdup(realizedCandidatePath); + return 1; + }else{ + debug_printf2("Unable to create directory %s, moving on to the next candidate.\n", realizedCandidatePath ); + } + }else{ + debug_printf2( "Unable to realize candidate %s, moving on to the next candidate.\n", parsedCandidatePath ); + } + }else{ + debug_printf2("Unable to parse candidate %s, moving on to the next candidate.\n", candidatePath ); + } + return 0; +} + +/** + * determineValidCachePaths() works exclusively with the cachepaths parameter. Because not all paths may be valid on all + * compute nodes, and because we want to have all nodes reach a consensus on which cache path to use, we + * determine the validity of all paths in the origPathList, save the intermediate results, and return a bit + * index to the user. Via allReduce() all nodes reach a consensus on the set of valid paths, and retrieves + * that informatino via getValidPathByIndex(). + */ +static char *realizedCachePaths[64], *parsedCachePaths[64], *symbolicCachePaths[64]; + +void determineValidCachePaths( uint64_t *validBitIdx, char *origPathList, number_t number ){ + + char *saveptr, *candidatePath, *pathList = strdup( origPathList ); + size_t pathList_len = strlen( pathList ); + uint64_t bitoffset = 0; + + *validBitIdx = 0; + debug_printf2("origPathList='%s', number='%lu'.\n", origPathList, number ); + + candidatePath = strtok_r( pathList, ":", &saveptr ); + while( NULL != candidatePath && bitoffset < 64 ){ + *validBitIdx |= validateCandidatePath( + candidatePath, + &realizedCachePaths[bitoffset], + &parsedCachePaths[bitoffset], + &symbolicCachePaths[bitoffset], number ) << bitoffset; + bitoffset++; + candidatePath = strtok_r( NULL, ":", &saveptr ); + } + memset( pathList, 'Q', pathList_len ); + free( pathList ); +} + +void getValidCachePathByIndex( uint64_t validBitIdx, char **realizedCachePath, char **parsedCachePath, char **symbolicCachePath ){ + uint64_t bitoffset = 0; + if (!validBitIdx){ + return; + } + while( (bitoffset < 64) && (((1 << bitoffset) & validBitIdx) == 0) ){ + bitoffset++; + } + if( realizedCachePath ) *realizedCachePath = realizedCachePaths[bitoffset]; + if( parsedCachePath ) *parsedCachePath = parsedCachePaths[bitoffset]; + if( symbolicCachePath ) *symbolicCachePath = symbolicCachePaths[bitoffset]; +} diff --git a/src/client/beboot/parseloc.h b/src/utils/parseloc.h similarity index 65% rename from src/client/beboot/parseloc.h rename to src/utils/parseloc.h index c5362e2e..a99409c3 100644 --- a/src/client/beboot/parseloc.h +++ b/src/utils/parseloc.h @@ -24,6 +24,13 @@ extern "C" { #include "spindle_launch.h" char *parse_location(char *loc, number_t number); +char *parse_location_noerr(char *loc, number_t number); +char *realize(char *path); +char **parse_colonsep_prefixes(char *colonsep_list, number_t number); +int is_local_prefix(const char *path, char **local_prefixes); +int validateCandidatePath( char *candidatePath, char **realizedPath, char **parsedPath, char **symbolicPath, number_t number ); +void determineValidCachePaths( uint64_t *validBitIdx, char *origPathList, number_t number ); +void getValidCachePathByIndex( uint64_t validBitIdx, char **realizedCachePath, char **parsedCachePath, char **symbolicCachePath ); #if defined(__cplusplus) } diff --git a/testsuite/test_driver.c b/testsuite/test_driver.c index acdc54e5..50ed9102 100644 --- a/testsuite/test_driver.c +++ b/testsuite/test_driver.c @@ -1258,7 +1258,7 @@ static char* getCacheLocation(char *env_var) return strdup(last_slash); } -static int checkLinkForLeak(const char *path, const char *spindle_loc) +static int checkLinkForLeak(const char *path, const char *cachepath) { char link_target[4096]; int result, error; @@ -1271,18 +1271,19 @@ static int checkLinkForLeak(const char *path, const char *spindle_loc) return -1; } - if (strstr(link_target, spindle_loc)) { - err_printf("Link at '%s' has path '%s', which leaks spindle path with '%s'\n", path, link_target, spindle_loc); + if (strstr(link_target, cachepath)) { + err_printf("Link at '%s' has path '%s', which leaks spindle path with '%s'\n", path, link_target, cachepath); return -1; } return 0; } -static int checkPathForLeak(const char *what, const char *path, const char *spindle_loc) + +static int checkPathForLeak(const char *what, const char *path, const char *cachepath) { - if (strstr(path, spindle_loc)) { - err_printf("%s: Path '%s' leaks spindle path with '%s'\n", what, path, spindle_loc); + if (strstr(path, cachepath)) { + err_printf("%s: Path '%s' leaks spindle path with '%s'\n", what, path, cachepath); return -1; } return 0; @@ -1290,14 +1291,15 @@ static int checkPathForLeak(const char *what, const char *path, const char *spin static int leak_check_cb(struct dl_phdr_info *p, size_t psize, void *opaque) { - char *spindle_loc = (char *) opaque; + char *cachepath = (char *) opaque; + assert( cachepath ); if (!p->dlpi_name || p->dlpi_name[0] == '\0') return 0; - checkPathForLeak("dl_iterate_phdr", p->dlpi_name, spindle_loc); + checkPathForLeak("dl_iterate_phdr", p->dlpi_name, cachepath); return 0; } -static int check_proc_maps(char *path, char *spindle_loc) +static int check_proc_maps(char *path, char *cachepath) { int fd, error, result; struct stat statbuf; @@ -1336,8 +1338,8 @@ static int check_proc_maps(char *path, char *spindle_loc) maps[filesize] = '\0'; close(fd); - if (strstr(maps, spindle_loc)) { - err_printf("Found leaked spindle path '%s' in maps '%s'\n", spindle_loc, path); + if (strstr(maps, cachepath)) { + err_printf("Found leaked spindle path '%s' in maps '%s'\n", cachepath, path); return -1; } @@ -1347,17 +1349,15 @@ static int check_proc_maps(char *path, char *spindle_loc) void check_for_path_leaks() { - char *spindle_loc = NULL; + char *cachepath = NULL; DIR *proc_fds = NULL; struct dirent *d; char path[4096]; struct link_map *lm; char *dlerr_msg = NULL; - spindle_loc = getCacheLocation("LDCS_LOCATION"); - if (!spindle_loc) - spindle_loc = getCacheLocation("LDCS_ORIG_LOCATION"); - if (!spindle_loc) { + cachepath = getCacheLocation("LDCS_CHOSEN_PARSED_CACHEPATH"); + if (!cachepath) { err_printf("Failed to calculate cache location"); goto done; } @@ -1373,11 +1373,16 @@ void check_for_path_leaks() for (d = readdir(proc_fds); d != NULL; d = readdir(proc_fds)) { if (d->d_name[0] == '.') continue; + // Ignore Spindle fifo files for now. + if ( strncmp( "315", d->d_name, 3 ) == 0 ) + continue; + if ( strncmp( "316", d->d_name, 3 ) == 0 ) + continue; strncpy(path, "/proc/self/fd/", sizeof(path)); strncat(path, d->d_name, sizeof(path)-1); - checkLinkForLeak(path, spindle_loc); + checkLinkForLeak(path, cachepath); } - checkLinkForLeak("/proc/self/exe", spindle_loc); + checkLinkForLeak("/proc/self/exe", cachepath); /** * Check link_maps for leaked spindle paths @@ -1385,22 +1390,22 @@ void check_for_path_leaks() for (lm = _r_debug.r_map; lm != NULL; lm = lm->l_next) { if (!lm->l_name || lm->l_name[0] == '\0') continue; - checkPathForLeak("link_map", lm->l_name, spindle_loc); + checkPathForLeak("link_map", lm->l_name, cachepath); } /** * Check libraries in dl_iterate_phdr for leaked paths **/ - dl_iterate_phdr(leak_check_cb, spindle_loc); + dl_iterate_phdr(leak_check_cb, cachepath); /** * Check /proc/pid/maps under various aliases for leaked names **/ - check_proc_maps("/proc/self/maps", spindle_loc); + check_proc_maps("/proc/self/maps", cachepath); snprintf(path, sizeof(path), "/proc/self/task/%d/maps", getpid()); - check_proc_maps(path, spindle_loc); + check_proc_maps(path, cachepath); snprintf(path, sizeof(path), "/proc/%d/maps", getpid()); - check_proc_maps(path, spindle_loc); + check_proc_maps(path, cachepath); /** * Check that dlerror doesn't leak the /__not_exists/ prefix @@ -1412,8 +1417,8 @@ void check_for_path_leaks() } done: - if (spindle_loc) - free(spindle_loc); + if (cachepath) + free(cachepath); if (proc_fds) closedir(proc_fds); }