#define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
cfs_num_physpages >> (25 - CFS_PAGE_SHIFT)), \
2UL)
-#define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
- cfs_num_physpages >> (25 - CFS_PAGE_SHIFT)), \
- 2UL)
-#define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
- cfs_num_physpages >> (25 - CFS_PAGE_SHIFT)), \
- 2UL)
/* Absolute limits */
#define MDS_THREADS_MIN 2
/*
* Initialized in mdt_mod_init().
*/
-unsigned long mdt_num_threads;
+static unsigned long mdt_num_threads;
+static unsigned long mdt_min_threads;
+static unsigned long mdt_max_threads;
/* ptlrpc request handler for MDT. All handlers are
* grouped into several slices - struct mdt_opc_slice,
* We'd like to have a mechanism to set this on a per-device
* basis, but alas...
*/
- .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
- MDT_MAX_THREADS),
- .psc_max_threads = MDT_MAX_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD
};
.psc_req_portal = MDS_READPAGE_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
- MDT_MAX_THREADS),
- .psc_max_threads = MDT_MAX_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD
};
m->mdt_readpage_service =
.psc_req_portal = MDS_SETATTR_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
- MDT_MAX_THREADS),
- .psc_max_threads = MDT_MAX_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD
};
.psc_req_portal = SEQ_CONTROLLER_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = SEQ_NUM_THREADS,
- .psc_max_threads = SEQ_NUM_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
};
.psc_req_portal = SEQ_METADATA_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = SEQ_NUM_THREADS,
- .psc_max_threads = SEQ_NUM_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
};
.psc_req_portal = SEQ_DATA_PORTAL,
.psc_rep_portal = OSC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = SEQ_NUM_THREADS,
- .psc_max_threads = SEQ_NUM_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
};
.psc_req_portal = FLD_REQUEST_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = FLD_NUM_THREADS,
- .psc_max_threads = FLD_NUM_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
};
.psc_req_portal = MDS_MDS_PORTAL,
.psc_rep_portal = MDC_REPLY_PORTAL,
.psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
- .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
- MDT_MAX_THREADS),
- .psc_max_threads = MDT_MAX_THREADS,
+ .psc_min_threads = mdt_min_threads,
+ .psc_max_threads = mdt_max_threads,
.psc_ctx_tags = LCT_MD_THREAD
};
m->mdt_xmds_service =
llo_local_obj_register(&mdt_last_recv);
- mdt_num_threads = MDT_NUM_THREADS;
+ if (mdt_num_threads > 0) {
+ if (mdt_num_threads > MDT_MAX_THREADS)
+ mdt_num_threads = MDT_MAX_THREADS;
+ if (mdt_num_threads < MDT_MIN_THREADS)
+ mdt_num_threads = MDT_MIN_THREADS;
+ mdt_max_threads = mdt_min_threads = mdt_num_threads;
+ } else {
+ mdt_max_threads = MDT_MAX_THREADS;
+ mdt_min_threads = MDT_MIN_THREADS;
+ if (mdt_min_threads < MDT_NUM_THREADS)
+ mdt_min_threads = MDT_NUM_THREADS;
+ }
+
lprocfs_mdt_init_vars(&lvars);
rc = class_register_type(&mdt_obd_device_ops, NULL,
lvars.module_vars, LUSTRE_MDT_NAME,
}
run_test 52 "check recovering objects from lost+found"
+# Checks threads_min/max/started for some service
+#
+# Arguments: service name (OST or MDT), facet (e.g., ost1, $SINGLEMDS), and a
+# parameter pattern prefix like 'ost.*.ost'.
+thread_sanity() {
+ local modname=$1
+ local facet=$2
+ local parampat=$3
+ local opts=$4
+ local tmin
+ local tmin2
+ local tmax
+ local tmax2
+ local tstarted
+ local paramp
+ local msg="Insane $modname thread counts"
+ shift 4
+
+ setup
+ check_mount || return 41
+
+ # We need to expand $parampat, but it may match multiple parameters, so
+ # we'll pick the first one
+ if ! paramp=$(do_facet $facet "lctl get_param -N ${parampat}.threads_min"|head -1); then
+ error "Couldn't expand ${parampat}.threads_min parameter name"
+ return 22
+ fi
+
+ # Remove the .threads_min part
+ paramp=${paramp%.threads_min}
+
+ # Check for sanity in defaults
+ tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0)
+ tmax=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0)
+ tstarted=$(do_facet $facet "lctl get_param -n ${paramp}.threads_started" || echo 0)
+ lassert 23 "$msg (PDSH problems?)" '(($tstarted && $tmin && $tmax))' || return $?
+ lassert 24 "$msg" '(($tstarted >= $tmin && $tstarted <= tmax ))' || return $?
+
+ # Check that we can lower min/max
+ do_facet $facet "lctl set_param ${paramp}.threads_min=$((tmin - 1))"
+ do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmax - 10))"
+ tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0)
+ tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0)
+ lassert 25 "$msg" '(($tmin2 == ($tmin - 1) && $tmax2 == ($tmax -10)))' || return $?
+
+ # Check that we can set min/max to the same value
+ do_facet $facet "lctl set_param ${paramp}.threads_min=$tmin"
+ do_facet $facet "lctl set_param ${paramp}.threads_max=$tmin"
+ tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0)
+ tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0)
+ lassert 26 "$msg" '(($tmin2 == $tmin && $tmax2 == $tmin))' || return $?
+
+ # Check that we can't set max < min
+ do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmin - 1))"
+ tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0)
+ tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0)
+ lassert 27 "$msg" '(($tmin <= $tmax2))' || return $?
+
+ # We need to ensure that we get the module options desired; to do this
+ # we set LOAD_MODULES_REMOTE=true and we call setmodopts below.
+ LOAD_MODULES_REMOTE=true
+ cleanup
+ local oldvalue
+ setmodopts -a $modname "$opts" oldvalue
+
+ load_modules
+ setup
+ check_mount || return 41
+
+ # Restore previous setting of MODOPTS_*
+ setmodopts $modname "$oldvalue"
+
+ # Check that $opts took
+ tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min")
+ tmax=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max")
+ tstarted=$(do_facet $facet "lctl get_param -n ${paramp}.threads_started")
+ lassert 28 "$msg" '(($tstarted == $tmin && $tstarted == $tmax ))' || return $?
+ cleanup
+
+ # Workaround a YALA bug where YALA expects that modules will remain
+ # loaded on the servers
+ LOAD_MODULES_REMOTE=false
+ setup
+ cleanup
+}
+
+test_53a() {
+ thread_sanity OST ost1 'ost.*.ost' 'oss_num_threads=64'
+}
+run_test 53a "check OSS thread count params"
+
+test_53b() {
+ thread_sanity MDT $SINGLEMDS 'mdt.*.*.' 'mdt_num_threads=64'
+}
+run_test 53b "check MDT thread count params"
+
cleanup_gss
equals_msg `basename $0`: test complete
[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
[ $failed ] && exit 1 || true
}
+# lrepl - Lustre test Read-Eval-Print Loop.
+#
+# This function implements a REPL for the Lustre test framework. It
+# doesn't exec an actual shell because the user may want to inspect
+# variables and use functions from the test framework.
+lrepl() {
+ local line
+ local rawline
+ local prompt
+
+ cat <<EOF
+ This is an interactive read-eval-print loop interactive shell
+ simulation that you can use to debug failing tests. You can
+ enter most bash command lines (see notes below).
+
+ Use this REPL to inspect variables, set them, call test
+ framework shell functions, etcetera.
+
+ 'exit' or EOF to exit this shell.
+
+ set \$retcode to 0 to cause the assertion failure that
+ triggered this REPL to be ignored.
+
+ Examples:
+ do_facet ost1 lctl get_param ost.*.ost.threads_*
+ do_rpc_nodes \$OSTNODES unload_modules
+
+ NOTES:
+ All but the last line of multi-line statements or blocks
+ must end in a backslash.
+
+ "Here documents" are not supported.
+
+ History is not supported, but command-line editing is.
+
+EOF
+
+ # Prompt escapes don't work in read -p, sadly.
+ prompt=":test_${testnum:-UNKNOWN}:$(uname -n):$(basename $PWD)% "
+
+ # We use read -r to get close to a shell experience
+ while read -e -r -p "$prompt" rawline; do
+ line=
+ case "$rawline" in
+ # Don't want to exit-exit, just exit the REPL
+ exit) break;;
+ # We need to handle continuations, and read -r doesn't do
+ # that for us. Yet we need read -r.
+ #
+ # We also use case/esac to compare lines read to "*\\"
+ # because [ "$line" = *\\ ] and variants of that don't work.
+ *\\) line="$rawline"
+ while read -e -r -p '> ' rawline
+ do
+ line="$line"$'\n'"$rawline"
+ case "$rawline" in
+ # We could check for here documents by matching
+ # against *<<*, but who cares.
+ *\\) continue;;
+ *) break;;
+ esac
+ done
+ ;;
+ *) line=$rawline
+ esac
+
+ case "$line" in
+ *\\) break;;
+ esac
+
+ # Finally! Time to eval.
+ eval "$line"
+ done
+
+ echo $'\n\tExiting interactive shell...\n'
+ return 0
+}
+
+# lassert - Lustre test framework assert
+#
+# Arguments: failure code, failure message, expression/statement
+#
+# lassert evaluates the expression given, and, if false, calls
+# error() to trigger test failure. If REPL_ON_LASSERT is true then
+# lassert will call lrepl() to give the user an interactive shell.
+# If the REPL sets retcode=0 then the assertion failure will be
+# ignored.
+lassert() {
+ local retcode=$1
+ local msg=$2
+ shift 2
+
+ echo "checking $* ($(eval echo \""$*"\"))..."
+ eval "$@" && return 0;
+
+ if ${REPL_ON_LASSERT:-false}; then
+ echo "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
+$msg"
+ lrepl
+ fi
+
+ error "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
+$msg"
+ return $retcode
+}
+
+# setmodopts- set module options for subsequent calls to load_modules
+#
+# Usage: setmodopts module_name new_value [var_in_which_to_save_old_value]
+# setmodopts -a module_name new_value [var_in_which_to_save_old_value]
+#
+# In the second usage the new value is appended to the old.
+setmodopts() {
+ local _append=false
+
+ if [ "$1" = -a ]; then
+ _append=true
+ shift
+ fi
+
+ local _var=MODOPTS_$1
+ local _newvalue=$2
+ local _savevar=$3
+ local _oldvalue
+
+ # Dynamic naming of variables is a pain in bash. In ksh93 we could
+ # write "nameref opts_var=${modname}_MODOPTS" then assign directly
+ # to opts_var. Associative arrays would also help, alternatively.
+ # Alas, we're stuck with eval until all distros move to a more recent
+ # version of bash. Fortunately we don't need to eval unset and export.
+
+ if [ -z "$_newvalue" ]; then
+ unset $_var
+ return 0
+ fi
+
+ _oldvalue=${!var}
+ $_append && _newvalue="$_oldvalue $_newvalue"
+ export $_var="$_newvalue"
+ echo setmodopts: ${_var}=${_newvalue}
+
+ [ -n "$_savevar" ] && eval $_savevar=\""$_oldvalue"\"
+}
+
echoerr () { echo "$@" 1>&2 ; }
signaled() {
fi
. $LUSTRE/tests/test-framework.sh
-init_test_env $@
+init_test_env
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-cmd=$1
-shift
-$cmd $@
+# Reset the trap on ERR set by the framework. Noticing this failure is the
+# framework's job.
+trap ERR
-exit $?
+# Execute the command
+"$@"
. $EXCEPT_LIST_FILE
fi
+[ -z "$MODPROBECONF" -a -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
+[ -z "$MODPROBECONF" -a -f /etc/modprobe.d/Lustre ] && MODPROBECONF=/etc/modprobe.d/Lustre
+
assert_DIR () {
local failed=""
[[ $DIR/ = $MOUNT/* ]] || \
/sbin/lsmod | grep -q $1
}
+# Load a module on the system where this is running.
+#
+# Synopsis: load_module module_name [module arguments for insmod/modprobe]
+#
+# If module arguments are not given but MODOPTS_<MODULE> is set, then its value
+# will be used as the arguments. Otherwise arguments will be obtained from
+# /etc/modprobe.conf, from /etc/modprobe.d/Lustre, or else none will be used.
+#
load_module() {
+ local optvar
EXT=".ko"
module=$1
shift
BASE=`basename $module $EXT`
+ # If no module arguments were passed, get them from $MODOPTS_<MODULE>, else from
+ # modprobe.conf
+ if [ $# -eq 0 ]; then
+ # $MODOPTS_<MODULE>; we could use associative arrays, but that's not in
+ # Bash until 4.x, so we resort to eval.
+ optvar="MODOPTS_$(basename $module | tr a-z A-Z)"
+ eval set -- \$$optvar
+ if [ $# -eq 0 -a -n "$MODPROBECONF" ]; then
+ # Nothing in $MODOPTS_<MODULE>; try modprobe.conf
+ set -- $(grep "^options\\s*\<${module}\>" $MODPROBECONF)
+ # Get rid of "options $module"
+ (($# > 0)) && shift 2
+
+ # Ensure we have accept=all for lnet
+ if [ $module = lnet ]; then
+ # OK, this is a bit wordy...
+ local arg accept_all_present=false
+ for arg in "$@"; do
+ [ "$arg" = accept=all ] && accept_all_present=true
+ done
+ $accept_all_present || set -- "$@" accept=all
+ fi
+ fi
+ fi
+
+ [ $# -gt 0 ] && echo "${module} options: '$*'"
+
module_loaded ${BASE} && return
+ # Note that insmod will ignore anything in modprobe.conf, which is why we're
+ # passing options on the command-line.
if [ "$BASE" == "lnet_selftest" ] && \
[ -f ${LUSTRE}/../lnet/selftest/${module}${EXT} ]; then
insmod ${LUSTRE}/../lnet/selftest/${module}${EXT}
-
elif [ -f ${LUSTRE}/${module}${EXT} ]; then
- insmod ${LUSTRE}/${module}${EXT} $@
+ insmod ${LUSTRE}/${module}${EXT} "$@"
else
# must be testing a "make install" or "rpm" installation
# note failed to load ptlrpc_gss is considered not fatal
if [ "$BASE" == "ptlrpc_gss" ]; then
- modprobe $BASE $@ 2>/dev/null || echo "gss/krb5 is not supported"
+ modprobe $BASE "$@" 2>/dev/null || echo "gss/krb5 is not supported"
else
- modprobe $BASE $@
+ modprobe $BASE "$@"
fi
fi
}
load_modules_local() {
if [ -n "$MODPROBE" ]; then
# use modprobe
- return 0
+ echo "Using modprobe to load modules"
+ return 0
fi
if [ "$HAVE_MODULES" = true ]; then
- # we already loaded
+ # we already loaded
+ echo "Modules already loaded"
return 0
fi
HAVE_MODULES=true
load_module ../libcfs/libcfs/libcfs
[ "$PTLDEBUG" ] && lctl set_param debug="$PTLDEBUG"
[ "$SUBSYSTEM" ] && lctl set_param subsystem_debug="${SUBSYSTEM# }"
- local MODPROBECONF=
- [ -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
- [ ! "$MODPROBECONF" -a -d /etc/modprobe.d ] && MODPROBECONF=/etc/modprobe.d/Lustre
- [ -z "$LNETOPTS" -a "$MODPROBECONF" ] && \
- LNETOPTS=$(awk '/^options lnet/ { print $0}' $MODPROBECONF | sed 's/^options lnet //g')
- echo $LNETOPTS | grep -q "accept=all" || LNETOPTS="$LNETOPTS accept=all";
- echo "lnet options: '$LNETOPTS'"
- # note that insmod will ignore anything in modprobe.conf
- load_module ../lnet/lnet/lnet $LNETOPTS
+ load_module ../lnet/lnet/lnet
LNETLND=${LNETLND:-"socklnd/ksocklnd"}
load_module ../lnet/klnds/$LNETLND
load_module lvfs/lvfs
[ "$1" = "$HOSTNAME" ]
}
+# Outputs environment variable assignments that should be passed to remote nodes
+get_env_vars() {
+ local var
+ local value
+
+ for var in ${!MODOPTS_*}; do
+ value=${!var}
+ echo "${var}=\"$value\""
+ done
+}
+
do_nodes() {
local verbose=false
# do not stripe off hostname if verbose, bug 19215
fi
if $verbose ; then
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\")"
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")"
else
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")" | sed -re "s/\w+:\s//g"
fi
return ${PIPESTATUS[0]}
}
error() {
error_noexit "$@"
- if $FAIL_ON_ERROR; then
- reset_fail_loc
- exit 1
- fi
+ exit 1
}
error_exit() {
- error_noexit "$@"
- exit 1
+ error "$@"
}
# use only if we are ignoring failures for this test, bugno required.
git://git.whamcloud.com / fs / lustre-release.git / commitdiff