summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorRichard Wareing <rwareing@fb.com>2014-02-12 18:37:55 -0800
committerJeff Darcy <jdarcy@redhat.com>2015-10-08 05:44:27 -0700
commit722ed512220395af8a707756b49df67afacda795 (patch)
treeae2e4cf59a5b7fb33ecb506eb823bb049b520a05
parentebe40c5047e7501a5bd1747cb4d62277ae0db6e9 (diff)
xlators: add JSON FOP statistics dumps every N seconds
Summary: - Adds a thread to the io-stats translator which dumps out statistics every N seconds where N is configurable by an option called "diagnostics.stats-dump-interval" - Thread cleanly starts/stops when translator is unloaded - Updates macros to use "Atomic Builtins" (e.g. intel CPU extentions) to use memory barries to update counters vs using locks. This should reduce overhead and prevent any deadlock bugs due to lock contention. Test Plan: - Test on development machine - Run prove -v tests/basic/stats-dump.t Change-Id: If071239d8fdc185e4e8fd527363cc042447a245d BUG: 1266476 Signed-off-by: Jeff Darcy <jdarcy@redhat.com> Reviewed-on: http://review.gluster.org/12209 Tested-by: NetBSD Build System <jenkins@build.gluster.org> Tested-by: Gluster Build System <jenkins@build.gluster.com> Reviewed-by: Avra Sengupta <asengupt@redhat.com>
-rw-r--r--configure.ac13
-rw-r--r--libglusterfs/src/xlator.c1
-rw-r--r--libglusterfs/src/xlator.h1
-rw-r--r--tests/basic/afr/arbiter.t8
-rw-r--r--tests/basic/stats-dump.t43
-rw-r--r--tests/bugs/cli/bug-1030580.t5
-rw-r--r--tests/bugs/cli/bug-1047416.t5
-rw-r--r--tests/bugs/snapshot/bug-1155042-dont-display-deactivated-snapshots.t6
-rw-r--r--xlators/debug/io-stats/src/Makefile.am3
-rw-r--r--xlators/debug/io-stats/src/io-stats-mem-types.h2
-rw-r--r--xlators/debug/io-stats/src/io-stats.c578
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-volume-set.c5
-rw-r--r--xlators/nfs/server/src/nfs.c20
13 files changed, 581 insertions, 109 deletions
diff --git a/configure.ac b/configure.ac
index ae7b61c7eff..e93bae92666 100644
--- a/configure.ac
+++ b/configure.ac
@@ -891,6 +891,19 @@ AC_SUBST(ARGP_STANDALONE_CPPFLAGS)
AC_SUBST(ARGP_STANDALONE_LDADD)
AC_SUBST(ARGP_STANDALONE_DIR)
+# Check for atomic operation support
+echo -n "checking for atomic operation support... "
+AC_LANG_CONFTEST([int main() { long int a = 4; __sync_fetch_and_add_8 (&a, 1); }])
+$CC conftest.c $CFLAGS -o conftest > /dev/null 2> /dev/null
+ret=$?
+rm -f conftest.o conftest
+if test $ret -eq 0 ; then
+ echo "yes"
+ AC_DEFINE(HAVE_ATOMIC_BUILTINS, 1, [have atomic builtins])
+else
+ echo "no"
+fi
+
AC_CHECK_HEADER([malloc.h], AC_DEFINE(HAVE_MALLOC_H, 1, [have malloc.h]))
AC_CHECK_FUNC([llistxattr], [have_llistxattr=yes])
diff --git a/libglusterfs/src/xlator.c b/libglusterfs/src/xlator.c
index 30775354253..ed120d23a8f 100644
--- a/libglusterfs/src/xlator.c
+++ b/libglusterfs/src/xlator.c
@@ -409,6 +409,7 @@ xlator_init (xlator_t *xl)
if (xl->mem_acct_init)
xl->mem_acct_init (xl);
+ xl->instance_name = NULL;
if (!xl->init) {
gf_msg (xl->name, GF_LOG_WARNING, 0, LG_MSG_INIT_FAILED,
"No init() found");
diff --git a/libglusterfs/src/xlator.h b/libglusterfs/src/xlator.h
index d7dbfb413be..27b598d7e60 100644
--- a/libglusterfs/src/xlator.h
+++ b/libglusterfs/src/xlator.h
@@ -853,6 +853,7 @@ struct _xlator {
/* Built during parsing */
char *name;
char *type;
+ char *instance_name; /* Used for multi NFSd */
xlator_t *next;
xlator_t *prev;
xlator_list_t *parents;
diff --git a/tests/basic/afr/arbiter.t b/tests/basic/afr/arbiter.t
index 84d2ccece51..cecbc605541 100644
--- a/tests/basic/afr/arbiter.t
+++ b/tests/basic/afr/arbiter.t
@@ -9,6 +9,7 @@ TEST glusterd;
TEST pidof glusterd
# Non arbiter replica 3 volumes should not have arbiter-count option enabled.
+TEST mkdir -p $B0/${V0}{0,1,2}
TEST $CLI volume create $V0 replica 3 $H0:$B0/${V0}{0,1,2}
TEST $CLI volume start $V0
TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 --entry-timeout=0 $M0;
@@ -17,7 +18,14 @@ EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0
TEST $CLI volume stop $V0
TEST $CLI volume delete $V0
+# Make sure we clean up *all the way* so we don't get "brick X is already part
+# of a volume" errors.
+cleanup;
+TEST glusterd;
+TEST pidof glusterd
+
# Create and mount a replica 3 arbiter volume.
+TEST mkdir -p $B0/${V0}{0,1,2}
TEST $CLI volume create $V0 replica 3 arbiter 1 $H0:$B0/${V0}{0,1,2}
TEST $CLI volume set $V0 performance.write-behind off
TEST $CLI volume set $V0 cluster.self-heal-daemon off
diff --git a/tests/basic/stats-dump.t b/tests/basic/stats-dump.t
new file mode 100644
index 00000000000..0a680e44e55
--- /dev/null
+++ b/tests/basic/stats-dump.t
@@ -0,0 +1,43 @@
+#!/bin/bash
+
+. $(dirname $0)/../include.rc
+. $(dirname $0)/../volume.rc
+. $(dirname $0)/../nfs.rc
+
+cleanup;
+
+TEST glusterd
+TEST pidof glusterd
+TEST $CLI volume create $V0 replica 3 $H0:$B0/${V0}{0,1,2}
+TEST $CLI volume set $V0 diagnostics.latency-measurement on
+TEST $CLI volume set $V0 diagnostics.count-fop-hits on
+TEST $CLI volume set $V0 diagnostics.stats-dump-interval 1
+TEST $CLI volume set $V0 nfs.disable off
+TEST $CLI volume start $V0
+sleep 1
+TEST glusterfs --volfile-id=/$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+TEST mount_nfs $H0:/$V0 $N0 nolock,soft,intr
+
+for i in {1..10};do
+ dd if=/dev/zero of=$M0/fuse_testfile$i bs=4k count=100
+done
+
+for i in {1..10};do
+ dd if=/dev/zero of=$N0/nfs_testfile$i bs=4k count=100
+done
+sleep 2
+
+# Verify we have non-zero write counts from the bricks, gNFSd
+# and the FUSE mount
+BRICK_OUTPUT="$(grep 'aggr.fop.write.count": "0"' ${GLUSTERD_WORKDIR}/stats/glusterfsd__d_backends_patchy?.dump)"
+BRICK_RET="$?"
+NFSD_OUTPUT="$(grep 'aggr.fop.write.count": "0"' ${GLUSTERD_WORKDIR}/stats/glusterfs_nfsd.dump)"
+NFSD_RET="$?"
+FUSE_OUTPUT="$(grep 'aggr.fop.write.count": "0"' ${GLUSTERD_WORKDIR}/stats/glusterfs_patchy.dump)"
+FUSE_RET="$?"
+
+TEST [ 0 -ne "$BRICK_RET" ]
+TEST [ 0 -ne "$NFSD_RET" ]
+TEST [ 0 -ne "$FUSE_RET" ]
+
+cleanup;
diff --git a/tests/bugs/cli/bug-1030580.t b/tests/bugs/cli/bug-1030580.t
index a907950e73f..ac8b1d8f6db 100644
--- a/tests/bugs/cli/bug-1030580.t
+++ b/tests/bugs/cli/bug-1030580.t
@@ -12,10 +12,15 @@ function write_to_file {
TEST glusterd
TEST pidof glusterd
TEST $CLI volume create $V0 replica 2 $H0:$B0/${V0}0 $H0:$B0/${V0}1
+# Increasing the json stats dump time interval, so that it doesn't mess with the test.
+TEST $CLI volume set $V0 diagnostics.stats-dump-interval 3600
TEST $CLI volume start $V0
TEST $CLI volume profile $V0 start
TEST glusterfs --volfile-id=/$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+# Clear the profile info uptill now.
+TEST $CLI volume profile $V0 info clear
+
# Verify 'volume profile info' prints both cumulative and incremental stats
write_to_file &
wait
diff --git a/tests/bugs/cli/bug-1047416.t b/tests/bugs/cli/bug-1047416.t
index 6e1b0a48467..864301034c9 100644
--- a/tests/bugs/cli/bug-1047416.t
+++ b/tests/bugs/cli/bug-1047416.t
@@ -12,10 +12,15 @@ function write_to_file {
TEST glusterd
TEST pidof glusterd
TEST $CLI volume create $V0 replica 2 $H0:$B0/${V0}0 $H0:$B0/${V0}1
+# Increasing the json stats dump time interval, so that it doesn't mess with the test.
+TEST $CLI volume set $V0 diagnostics.stats-dump-interval 3600
TEST $CLI volume start $V0
TEST $CLI volume profile $V0 start
TEST glusterfs --volfile-id=/$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+# Clear the profile info uptill now.
+TEST $CLI volume profile $V0 info clear
+
# Verify 'volume profile info' prints both cumulative and incremental stats
write_to_file &
wait
diff --git a/tests/bugs/snapshot/bug-1155042-dont-display-deactivated-snapshots.t b/tests/bugs/snapshot/bug-1155042-dont-display-deactivated-snapshots.t
index 6697c263ac1..c5a285eb775 100644
--- a/tests/bugs/snapshot/bug-1155042-dont-display-deactivated-snapshots.t
+++ b/tests/bugs/snapshot/bug-1155042-dont-display-deactivated-snapshots.t
@@ -21,16 +21,16 @@ TEST $GFS --volfile-server=$H0 --volfile-id=$V0 $M0
# in the USS world
gluster snapshot config activate-on-create enable
for i in {1..10}; do $CLI snapshot create snap$i $V0 no-timestamp; done
-EXPECT 10 uss_count_snap_displayed $M0
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 10 uss_count_snap_displayed $M0
# snapshots should not be displayed after deactivation
for i in {1..10}; do $CLI snapshot deactivate snap$i --mode=script; done
-EXPECT 0 uss_count_snap_displayed $M0
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 0 uss_count_snap_displayed $M0
# activate all the snapshots and check if all the activated snapshots
# are displayed again
for i in {1..10}; do $CLI snapshot activate snap$i --mode=script; done
-EXPECT 10 uss_count_snap_displayed $M0
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 10 uss_count_snap_displayed $M0
cleanup;
diff --git a/xlators/debug/io-stats/src/Makefile.am b/xlators/debug/io-stats/src/Makefile.am
index a55933b6d19..ecbc82061a7 100644
--- a/xlators/debug/io-stats/src/Makefile.am
+++ b/xlators/debug/io-stats/src/Makefile.am
@@ -11,7 +11,8 @@ noinst_HEADERS = io-stats-mem-types.h
AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
-I$(top_srcdir)/rpc/xdr/src \
- -I$(top_srcdir)/rpc/rpc-lib/src
+ -I$(top_srcdir)/rpc/rpc-lib/src \
+ -DDATADIR=\"$(localstatedir)\"
AM_CFLAGS = -Wall $(GF_CFLAGS)
diff --git a/xlators/debug/io-stats/src/io-stats-mem-types.h b/xlators/debug/io-stats/src/io-stats-mem-types.h
index c30dfb17e7c..d7a1055a571 100644
--- a/xlators/debug/io-stats/src/io-stats-mem-types.h
+++ b/xlators/debug/io-stats/src/io-stats-mem-types.h
@@ -13,6 +13,8 @@
#include "mem-types.h"
+extern const char *__progname;
+
enum gf_io_stats_mem_types_ {
gf_io_stats_mt_ios_conf = gf_common_mt_end + 1,
gf_io_stats_mt_ios_fd,
diff --git a/xlators/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c
index 68a2dfc6360..3157a65dae1 100644
--- a/xlators/debug/io-stats/src/io-stats.c
+++ b/xlators/debug/io-stats/src/io-stats.c
@@ -113,6 +113,9 @@ struct ios_conf {
gf_boolean_t measure_latency;
struct ios_stat_head list[IOS_STATS_TYPE_MAX];
struct ios_stat_head thru_list[IOS_STATS_THRU_MAX];
+ int32_t ios_dump_interval;
+ pthread_t dump_thread;
+ gf_boolean_t dump_thread_should_die;
};
@@ -126,10 +129,11 @@ struct ios_fd {
};
typedef enum {
- IOS_DUMP_TYPE_NONE = 0,
- IOS_DUMP_TYPE_FILE = 1,
- IOS_DUMP_TYPE_DICT = 2,
- IOS_DUMP_TYPE_MAX = 3
+ IOS_DUMP_TYPE_NONE = 0,
+ IOS_DUMP_TYPE_FILE = 1,
+ IOS_DUMP_TYPE_DICT = 2,
+ IOS_DUMP_TYPE_JSON_FILE = 3,
+ IOS_DUMP_TYPE_MAX = 4
} ios_dump_type_t;
struct ios_dump_args {
@@ -158,6 +162,12 @@ is_fop_latency_started (call_frame_t *frame)
return memcmp (&frame->begin, &epoch, sizeof (epoch));
}
+#ifdef GF_LINUX_HOST_OS
+#define _IOS_DUMP_DIR DATADIR "/lib/glusterd/stats"
+#else
+#define _IOS_DUMP_DIR DATADIR "/db/glusterd/stats"
+#endif
+
#define END_FOP_LATENCY(frame, op) \
do { \
struct ios_conf *conf = NULL; \
@@ -193,6 +203,16 @@ is_fop_latency_started (call_frame_t *frame)
conf->incremental.fop_hits[GF_FOP_##op]++; \
} while (0)
+#if defined(HAVE_ATOMIC_BUILTINS)
+#define STATS_LOCK(x)
+#define STATS_UNLOCK(x)
+#define STATS_ADD(x,i) __sync_add_and_fetch (&x, i)
+#else
+#define STATS_LOCK(x) LOCK (x)
+#define STATS_UNLOCK(x) UNLOCK (x)
+#define STATS_ADD(x,i) (x) += (i)
+#endif
+
#define UPDATE_PROFILE_STATS(frame, op) \
do { \
struct ios_conf *conf = NULL; \
@@ -200,7 +220,7 @@ is_fop_latency_started (call_frame_t *frame)
if (!is_fop_latency_started (frame)) \
break; \
conf = this->private; \
- LOCK (&conf->lock); \
+ STATS_LOCK (&conf->lock); \
{ \
if (conf && conf->measure_latency && \
conf->count_fop_hits) { \
@@ -209,113 +229,107 @@ is_fop_latency_started (call_frame_t *frame)
update_ios_latency (conf, frame, GF_FOP_##op);\
} \
} \
- UNLOCK (&conf->lock); \
+ STATS_UNLOCK (&conf->lock); \
} while (0)
-#define BUMP_READ(fd, len) \
- do { \
- struct ios_conf *conf = NULL; \
- struct ios_fd *iosfd = NULL; \
- int lb2 = 0; \
- \
- conf = this->private; \
- lb2 = log_base2 (len); \
- ios_fd_ctx_get (fd, this, &iosfd); \
- if (!conf) \
- break; \
- \
- LOCK (&conf->lock); \
- { \
- conf->cumulative.data_read += len; \
- conf->incremental.data_read += len; \
- conf->cumulative.block_count_read[lb2]++; \
- conf->incremental.block_count_read[lb2]++; \
- \
- if (iosfd) { \
- iosfd->data_read += len; \
- iosfd->block_count_read[lb2]++; \
- } \
- } \
- UNLOCK (&conf->lock); \
+#define BUMP_READ(fd, len) \
+ do { \
+ struct ios_conf *conf = NULL; \
+ struct ios_fd *iosfd = NULL; \
+ int lb2 = 0; \
+ \
+ conf = this->private; \
+ lb2 = log_base2 (len); \
+ ios_fd_ctx_get (fd, this, &iosfd); \
+ if (!conf) \
+ break; \
+ \
+ STATS_LOCK (&conf->lock); \
+ { \
+ STATS_ADD (conf->cumulative.data_read, len); \
+ STATS_ADD (conf->incremental.data_read, len); \
+ STATS_ADD (conf->cumulative.block_count_read[lb2], 1); \
+ STATS_ADD (conf->incremental.block_count_read[lb2], 1);\
+ \
+ if (iosfd) { \
+ STATS_ADD (iosfd->data_read, len); \
+ STATS_ADD (iosfd->block_count_read[lb2], 1); \
+ } \
+ } \
+ STATS_UNLOCK (&conf->lock); \
} while (0)
+#define BUMP_WRITE(fd, len) \
+ do { \
+ struct ios_conf *conf = NULL; \
+ struct ios_fd *iosfd = NULL; \
+ int lb2 = 0; \
+ \
+ conf = this->private; \
+ lb2 = log_base2 (len); \
+ ios_fd_ctx_get (fd, this, &iosfd); \
+ if (!conf) \
+ break; \
+ STATS_LOCK (&conf->lock); \
+ { \
+ STATS_ADD (conf->cumulative.data_written, len); \
+ STATS_ADD (conf->incremental.data_written, len); \
+ STATS_ADD (conf->cumulative.block_count_write[lb2], 1);\
+ STATS_ADD (conf->incremental.block_count_write[lb2], 1);\
+ \
+ if (iosfd) { \
+ STATS_ADD (iosfd->data_written, len); \
+ STATS_ADD (iosfd->block_count_write[lb2], 1); \
+ } \
+ } \
+ STATS_UNLOCK (&conf->lock); \
+ } while (0)
-#define BUMP_WRITE(fd, len) \
+#define BUMP_STATS(iosstat, type) \
do { \
- struct ios_conf *conf = NULL; \
- struct ios_fd *iosfd = NULL; \
- int lb2 = 0; \
+ struct ios_conf *conf = NULL; \
+ uint64_t value = 0; \
\
conf = this->private; \
- lb2 = log_base2 (len); \
- ios_fd_ctx_get (fd, this, &iosfd); \
- if (!conf) \
- break; \
\
- LOCK (&conf->lock); \
+ LOCK(&iosstat->lock); \
{ \
- conf->cumulative.data_written += len; \
- conf->incremental.data_written += len; \
- conf->cumulative.block_count_write[lb2]++; \
- conf->incremental.block_count_write[lb2]++; \
- \
- if (iosfd) { \
- iosfd->data_written += len; \
- iosfd->block_count_write[lb2]++; \
- } \
+ value = STATS_ADD (iosstat->counters[type], 1); \
} \
- UNLOCK (&conf->lock); \
- } while (0)
-
-
-#define BUMP_STATS(iosstat, type) \
- do { \
- struct ios_conf *conf = NULL; \
- uint64_t value = 0; \
- \
- conf = this->private; \
- \
- LOCK(&iosstat->lock); \
- { \
- iosstat->counters[type]++; \
- value = iosstat->counters[type]; \
- } \
- UNLOCK (&iosstat->lock); \
- ios_stat_add_to_list (&conf->list[type], \
- value, iosstat); \
- \
+ UNLOCK (&iosstat->lock); \
+ ios_stat_add_to_list (&conf->list[type], \
+ value, iosstat); \
} while (0)
-
-#define BUMP_THROUGHPUT(iosstat, type) \
- do { \
- struct ios_conf *conf = NULL; \
- double elapsed; \
- struct timeval *begin, *end; \
- double throughput; \
- int flag = 0; \
- \
- begin = &frame->begin; \
- end = &frame->end; \
- \
- elapsed = (end->tv_sec - begin->tv_sec) * 1e6 \
- + (end->tv_usec - begin->tv_usec); \
- throughput = op_ret / elapsed; \
- \
- conf = this->private; \
- LOCK(&iosstat->lock); \
- { \
- if (iosstat->thru_counters[type].throughput \
- <= throughput) { \
- iosstat->thru_counters[type].throughput = \
- throughput; \
- gettimeofday (&iosstat-> \
- thru_counters[type].time, NULL); \
+#define BUMP_THROUGHPUT(iosstat, type) \
+ do { \
+ struct ios_conf *conf = NULL; \
+ double elapsed; \
+ struct timeval *begin, *end; \
+ double throughput; \
+ int flag = 0; \
+ \
+ begin = &frame->begin; \
+ end = &frame->end; \
+ \
+ elapsed = (end->tv_sec - begin->tv_sec) * 1e6 \
+ + (end->tv_usec - begin->tv_usec); \
+ throughput = op_ret / elapsed; \
+ \
+ conf = this->private; \
+ STATS_LOCK (&iosstat->lock); \
+ { \
+ if (iosstat->thru_counters[type].throughput \
+ <= throughput) { \
+ iosstat->thru_counters[type].throughput = \
+ throughput; \
+ gettimeofday (&iosstat-> \
+ thru_counters[type].time, NULL); \
flag = 1; \
- } \
- } \
- UNLOCK (&iosstat->lock); \
- if (flag) \
+ } \
+ } \
+ STATS_UNLOCK (&iosstat->lock); \
+ if (flag) \
ios_stat_add_to_list (&conf->thru_list[type], \
throughput, iosstat); \
} while (0)
@@ -536,7 +550,7 @@ ios_stats_cleanup (xlator_t *this, inode_t *inode)
fprintf (logfp, fmt); \
fprintf (logfp, "\n"); \
} \
- gf_log (this->name, GF_LOG_INFO, fmt); \
+ gf_log (this->name, GF_LOG_DEBUG, fmt); \
} while (0)
int
@@ -581,6 +595,227 @@ ios_dump_throughput_stats (struct ios_stat_head *list_head, xlator_t *this,
}
int
+_io_stats_get_key_prefix (xlator_t *this, char **key_prefix) {
+ char *key_root = "gluster";
+ char *xlator_name = NULL;
+ char *instance_name = NULL;
+ size_t key_len = 0;
+ int bytes_written = 0;
+ int i = 0;
+ int ret = 0;
+
+ xlator_name = strdupa (this->name);
+ for (i = 0; i < strlen (xlator_name); i++) {
+ if (xlator_name[i] == '/')
+ xlator_name[i] = '_';
+ }
+
+ instance_name = this->instance_name;
+ if (this->name && strcmp (this->name, "glustershd") == 0) {
+ xlator_name = "shd";
+ } else if (this->prev &&
+ strcmp (this->prev->name, "nfs-server") == 0) {
+ xlator_name = "nfsd";
+ if (this->prev->instance_name)
+ instance_name = strdupa (this->prev->instance_name);
+ }
+
+ if (strcmp (__progname, "glusterfsd") == 0)
+ key_root = "gluster.brick";
+
+ if (instance_name) {
+ /* +3 for 2 x "." + NULL */
+ key_len = strlen (key_root) + strlen (xlator_name) +
+ strlen (instance_name) + 3;
+ *key_prefix = GF_CALLOC (key_len, sizeof (char),
+ gf_common_mt_char);
+ if (!key_prefix) {
+ ret = -ENOMEM;
+ goto err;
+ }
+ bytes_written = snprintf (*key_prefix, key_len, "%s.%s.%s",
+ key_root, xlator_name, instance_name);
+ if (bytes_written != key_len - 1) {
+ ret = -EINVAL;
+ goto err;
+ }
+ } else {
+ /* +2 for 1 x "." + NULL */
+ key_len = strlen (key_root) + strlen (xlator_name) + 2;
+ *key_prefix = GF_CALLOC (key_len, sizeof (char),
+ gf_common_mt_char);
+ if (!key_prefix) {
+ ret = -ENOMEM;
+ goto err;
+ }
+ bytes_written = snprintf (*key_prefix, key_len, "%s.%s",
+ key_root, xlator_name);
+ if (bytes_written != key_len - 1) {
+ ret = -EINVAL;
+ goto err;
+ }
+ }
+ return 0;
+err:
+ GF_FREE (*key_prefix);
+ *key_prefix = NULL;
+ return ret;
+}
+
+int
+io_stats_dump_global_to_json_logfp (xlator_t *this,
+ struct ios_global_stats *stats, struct timeval *now, int interval,
+ FILE *logfp)
+{
+ int i = 0;
+ int j = 0;
+ struct ios_conf *conf = NULL;
+ char *key_prefix = NULL;
+ char *str_prefix = NULL;
+ char *lc_fop_name = NULL;
+ int ret = 1; /* Default to error */
+ int rw_size;
+ char *rw_unit = NULL;
+ long fop_hits;
+ float fop_lat_ave;
+ float fop_lat_min;
+ float fop_lat_max;
+ double interval_sec;
+
+ interval_sec = ((now->tv_sec * 1000000.0 + now->tv_usec) -
+ (stats->started_at.tv_sec * 1000000.0 +
+ stats->started_at.tv_usec)) / 1000000.0;
+
+ conf = this->private;
+
+ ret = _io_stats_get_key_prefix (this, &key_prefix);
+ if (ret) {
+ goto out;
+ }
+
+ if (interval == -1) {
+ str_prefix = "aggr";
+
+ } else {
+ str_prefix = "inter";
+ }
+ ios_log (this, logfp, "{");
+
+ for (i = 0; i < 31; i++) {
+ rw_size = (1 << i);
+ if (rw_size >= 1024 * 1024) {
+ rw_size = rw_size / (1024 * 1024);
+ rw_unit = "mb";
+ } else if (rw_size >= 1024) {
+ rw_size = rw_size / 1024;
+ rw_unit = "kb";
+ } else {
+ rw_unit = "b";
+ }
+
+ if (interval == -1) {
+ ios_log (this, logfp,
+ "\"%s.%s.read_%d%s\": \"%"PRId64"\",",
+ key_prefix, str_prefix, rw_size, rw_unit,
+ stats->block_count_read[i]);
+ ios_log (this, logfp,
+ "\"%s.%s.write_%d%s\": \"%"PRId64"\",",
+ key_prefix, str_prefix, rw_size, rw_unit,
+ stats->block_count_write[i]);
+ } else {
+ ios_log (this, logfp,
+ "\"%s.%s.read_%d%s_per_sec\": \"%0.2lf\",",
+ key_prefix, str_prefix, rw_size, rw_unit,
+ (double)(stats->block_count_read[i] /
+ interval_sec));
+ ios_log (this, logfp,
+ "\"%s.%s.write_%d%s_per_sec\": \"%0.2lf\",",
+ key_prefix, str_prefix, rw_size, rw_unit,
+ (double)(stats->block_count_write[i] /
+ interval_sec));
+ }
+ }
+
+ if (interval == -1) {
+ ios_log (this, logfp, "\"%s.%s.fds.open_count\": \"%"PRId64
+ "\",", key_prefix, str_prefix,
+ conf->cumulative.nr_opens);
+ ios_log (this, logfp,
+ "\"%s.%s.fds.max_open_count\": \"%"PRId64"\",",
+ key_prefix, str_prefix, conf->cumulative.max_nr_opens);
+ }
+
+ for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+ lc_fop_name = strdupa (gf_fop_list[i]);
+ for (j = 0; lc_fop_name[j]; j++) {
+ lc_fop_name[j] = tolower (lc_fop_name[j]);
+ }
+
+ fop_hits = 0;
+ fop_lat_ave = 0.0;
+ fop_lat_min = 0.0;
+ fop_lat_max = 0.0;
+ if (stats->fop_hits[i]) {
+ fop_hits = stats->fop_hits[i];
+ if (stats->latency[i].avg) {
+ fop_lat_ave = stats->latency[i].avg;
+ fop_lat_min = stats->latency[i].min;
+ fop_lat_max = stats->latency[i].max;
+ }
+ }
+ if (interval == -1) {
+ ios_log (this, logfp,
+ "\"%s.%s.fop.%s.count\": \"%"PRId64"\",",
+ key_prefix, str_prefix, lc_fop_name,
+ fop_hits);
+ } else {
+ ios_log (this, logfp,
+ "\"%s.%s.fop.%s.per_sec\": \"%0.2lf\",",
+ key_prefix, str_prefix, lc_fop_name,
+ (double)(fop_hits / interval_sec));
+ }
+
+ ios_log (this, logfp,
+ "\"%s.%s.fop.%s.latency_ave_usec\": \"%0.2lf\",",
+ key_prefix, str_prefix, lc_fop_name, fop_lat_ave);
+ ios_log (this, logfp,
+ "\"%s.%s.fop.%s.latency_min_usec\": \"%0.2lf\",",
+ key_prefix, str_prefix, lc_fop_name, fop_lat_min);
+ ios_log (this, logfp,
+ "\"%s.%s.fop.%s.latency_max_usec\": \"%0.2lf\",",
+ key_prefix, str_prefix, lc_fop_name, fop_lat_max);
+ }
+ if (interval == -1) {
+ ios_log (this, logfp, "\"%s.%s.uptime\": \"%"PRId64"\",",
+ key_prefix, str_prefix,
+ (uint64_t) (now->tv_sec - stats->started_at.tv_sec));
+ ios_log (this, logfp, "\"%s.%s.bytes_read\": \"%"PRId64"\",",
+ key_prefix, str_prefix, stats->data_read);
+ ios_log (this, logfp, "\"%s.%s.bytes_written\": \"%"PRId64"\"",
+ key_prefix, str_prefix, stats->data_written);
+ } else {
+ ios_log (this, logfp,
+ "\"%s.%s.sample_interval_sec\": \"%0.2lf\",",
+ key_prefix, str_prefix,
+ interval_sec);
+ ios_log (this, logfp,
+ "\"%s.%s.bytes_read_per_sec\": \"%0.2lf\",",
+ key_prefix, str_prefix,
+ (double)(stats->data_read / interval_sec));
+ ios_log (this, logfp,
+ "\"%s.%s.bytes_written_per_sec\": \"%0.2lf\"",
+ key_prefix, str_prefix,
+ (double)(stats->data_written / interval_sec));
+ }
+
+ ios_log (this, logfp, "}");
+ ret = 0;
+out:
+ GF_FREE (key_prefix);
+ return ret;
+}
+
+int
io_stats_dump_global_to_logfp (xlator_t *this, struct ios_global_stats *stats,
struct timeval *now, int interval, FILE* logfp)
{
@@ -876,6 +1111,10 @@ io_stats_dump_global (xlator_t *this, struct ios_global_stats *stats,
switch (args->type) {
+ case IOS_DUMP_TYPE_JSON_FILE:
+ ret = io_stats_dump_global_to_json_logfp (
+ this, stats, now, interval, args->u.logfp);
+ break;
case IOS_DUMP_TYPE_FILE:
ret = io_stats_dump_global_to_logfp (this, stats, now,
interval, args->u.logfp);
@@ -904,6 +1143,7 @@ ios_dump_args_init (struct ios_dump_args *args, ios_dump_type_t type,
args->type = type;
switch (args->type) {
+ case IOS_DUMP_TYPE_JSON_FILE:
case IOS_DUMP_TYPE_FILE:
args->u.logfp = output;
break;
@@ -2220,6 +2460,8 @@ conditional_dump (dict_t *dict, char *key, data_t *value, void *data)
char *filename = NULL;
FILE *logfp = NULL;
struct ios_dump_args args = {0};
+ int pid;
+ char dump_key[100];
stub = data;
this = stub->this;
@@ -2229,6 +2471,8 @@ conditional_dump (dict_t *dict, char *key, data_t *value, void *data)
memcpy (filename, data_to_str (value), value->len);
if (fnmatch ("*io*stat*dump", key, 0) == 0) {
+ pid = getpid ();
+
if (!strncmp (filename, "", 1)) {
gf_log (this->name, GF_LOG_ERROR, "No filename given");
@@ -2239,15 +2483,116 @@ conditional_dump (dict_t *dict, char *key, data_t *value, void *data)
gf_log (this->name, GF_LOG_ERROR, "failed to open %s "
"for writing", filename);
return -1;
- }
- (void) ios_dump_args_init (&args, IOS_DUMP_TYPE_FILE,
- logfp);
+ }
+ sprintf (dump_key, "*io*stat*%d_json_dump", pid);
+ if (fnmatch (dump_key, key, 0) == 0) {
+ (void) ios_dump_args_init (
+ &args, IOS_DUMP_TYPE_JSON_FILE,
+ logfp);
+ } else {
+ (void) ios_dump_args_init (&args, IOS_DUMP_TYPE_FILE,
+ logfp);
+ }
io_stats_dump (this, &args, GF_CLI_INFO_ALL, _gf_false);
fclose (logfp);
}
return 0;
}
+int
+_ios_destroy_dump_thread (struct ios_conf *conf) {
+ conf->dump_thread_should_die = _gf_true;
+ if (conf->ios_dump_interval > 0) {
+ (void) pthread_cancel (conf->dump_thread);
+ (void) pthread_join (conf->dump_thread, NULL);
+ }
+ return 0;
+}
+
+void *
+_ios_dump_thread (xlator_t *this) {
+ struct ios_conf *conf = NULL;
+ FILE *logfp = NULL;
+ struct ios_dump_args args = {0};
+ int i;
+ int bytes_written = 0;
+ char filename[PATH_MAX];
+ char *xlator_name;
+ char *instance_name;
+ gf_boolean_t log_fopen_failure = _gf_true;
+ int old_cancel_type;
+
+ conf = this->private;
+ gf_log (this->name, GF_LOG_INFO, "IO stats dump thread started, "
+ "polling IO stats every %d seconds", conf->ios_dump_interval);
+ xlator_name = strdupa (this->name);
+ for (i = 0; i < strlen (xlator_name); i++) {
+ if (xlator_name[i] == '/')
+ xlator_name[i] = '_';
+ }
+ instance_name = this->instance_name;
+ if (this->name && strcmp (this->name, "glustershd") == 0) {
+ xlator_name = "shd";
+ } else if (this->prev &&
+ strcmp (this->prev->name, "nfs-server") == 0) {
+ xlator_name = "nfsd";
+ instance_name = this->prev->instance_name;
+ }
+ if (mkdir (_IOS_DUMP_DIR, S_IRWXU | S_IRWXO | S_IRWXG) == (-1)) {
+ if (errno != EEXIST) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "could not create stats-dump directory %s",
+ _IOS_DUMP_DIR);
+ goto out;
+ }
+ }
+ if (instance_name) {
+ bytes_written = snprintf (filename, PATH_MAX,
+ "%s/%s_%s_%s.dump", _IOS_DUMP_DIR,
+ __progname, xlator_name, instance_name);
+ } else {
+ bytes_written = snprintf (filename, PATH_MAX, "%s/%s_%s.dump",
+ _IOS_DUMP_DIR, __progname, xlator_name);
+ }
+ if (bytes_written >= PATH_MAX) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Invalid path for IO Stats dump: %s", filename);
+ goto out;
+ }
+ while (1) {
+ if (conf->dump_thread_should_die)
+ break;
+ (void) pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS,
+ &old_cancel_type);
+ sleep (conf->ios_dump_interval);
+ (void) pthread_setcanceltype (PTHREAD_CANCEL_DEFERRED,
+ &old_cancel_type);
+ /*
+ * It's not clear whether we should reopen this each time, or
+ * just hold it open and rewind/truncate on each iteration.
+ * Leaving it alone for now.
+ */
+ logfp = fopen (filename, "w+");
+ if (!logfp) {
+ if (log_fopen_failure) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "could not open stats-dump file %s",
+ filename);
+ log_fopen_failure = _gf_false;
+ }
+ continue;
+ }
+ (void) ios_dump_args_init (
+ &args, IOS_DUMP_TYPE_JSON_FILE,
+ logfp);
+ io_stats_dump (this, &args, GF_CLI_INFO_ALL, _gf_false);
+ fclose (logfp);
+ log_fopen_failure = _gf_true;
+ }
+out:
+ gf_log (this->name, GF_LOG_INFO, "IO stats dump thread terminated");
+ return NULL;
+}
int
io_stats_setxattr (call_frame_t *frame, xlator_t *this,
@@ -2747,6 +3092,7 @@ reconfigure (xlator_t *this, dict_t *options)
int logger = -1;
uint32_t log_buf_size = 0;
uint32_t log_flush_timeout = 0;
+ int32_t old_dump_interval;
if (!this || !this->private)
goto out;
@@ -2762,6 +3108,14 @@ reconfigure (xlator_t *this, dict_t *options)
GF_OPTION_RECONF ("latency-measurement", conf->measure_latency,
options, bool, out);
+ old_dump_interval = conf->ios_dump_interval;
+ GF_OPTION_RECONF ("ios-dump-interval", conf->ios_dump_interval, options,
+ int32, out);
+ if ((old_dump_interval <= 0) && (conf->ios_dump_interval > 0)) {
+ pthread_create (&conf->dump_thread, NULL,
+ (void *) &_ios_dump_thread, this);
+ }
+
GF_OPTION_RECONF ("sys-log-level", sys_log_str, options, str, out);
if (sys_log_str) {
sys_log_level = glusterd_check_log_level (sys_log_str);
@@ -2827,6 +3181,7 @@ ios_conf_destroy (struct ios_conf *conf)
return;
ios_destroy_top_stats (conf);
+ _ios_destroy_dump_thread (conf);
LOCK_DESTROY (&conf->lock);
GF_FREE(conf);
}
@@ -2887,7 +3242,10 @@ init (xlator_t *this)
GF_OPTION_INIT ("count-fop-hits", conf->count_fop_hits, bool, out);
GF_OPTION_INIT ("latency-measurement", conf->measure_latency,
- bool, out);
+ bool, out);
+
+ GF_OPTION_INIT ("ios-dump-interval", conf->ios_dump_interval,
+ int32, out);
GF_OPTION_INIT ("sys-log-level", sys_log_str, str, out);
if (sys_log_str) {
@@ -2921,6 +3279,10 @@ init (xlator_t *this)
this->private = conf;
+ if (conf->ios_dump_interval > 0) {
+ pthread_create (&conf->dump_thread, NULL,
+ (void *) &_ios_dump_thread, this);
+ }
ret = 0;
out:
if (!this->private) {
@@ -2940,9 +3302,9 @@ fini (xlator_t *this)
return;
conf = this->private;
- this->private = NULL;
ios_conf_destroy (conf);
+ this->private = NULL;
gf_log (this->name, GF_LOG_INFO,
"io-stats translator unloaded");
return;
@@ -3127,6 +3489,14 @@ struct volume_options options[] = {
.description = "If on stats related to file-operations would be "
"tracked inside GlusterFS data-structures."
},
+ { .key = { "ios-dump-interval" },
+ .type = GF_OPTION_TYPE_INT,
+ .min = 0,
+ .max = 3600,
+ .default_value = "5",
+ .description = "Interval (in seconds) at which to auto-dump "
+ "statistics. Zero disables automatic dumping."
+ },
{ .key = { "latency-measurement" },
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index c18bc5d5a76..e93a22eafdd 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -768,6 +768,11 @@ struct volopt_map_entry glusterd_volopt_map[] = {
.op_version = GD_OP_VERSION_3_6_0,
.flags = OPT_FLAG_CLIENT_OPT
},
+ { .key = "diagnostics.stats-dump-interval",
+ .voltype = "debug/io-stats",
+ .option = "ios-dump-interval",
+ .op_version = 1
+ },
/* IO-cache xlator options */
{ .key = "performance.cache-max-file-size",
diff --git a/xlators/nfs/server/src/nfs.c b/xlators/nfs/server/src/nfs.c
index e057b17c3d6..4dda35c49ef 100644
--- a/xlators/nfs/server/src/nfs.c
+++ b/xlators/nfs/server/src/nfs.c
@@ -729,7 +729,7 @@ struct nfs_state *
nfs_init_state (xlator_t *this)
{
struct nfs_state *nfs = NULL;
- int ret = -1;
+ int i = 0, ret = -1;
unsigned int fopspoolsize = 0;
char *optstr = NULL;
gf_boolean_t boolt = _gf_false;
@@ -856,6 +856,23 @@ nfs_init_state (xlator_t *this)
}
}
+ if (dict_get (this->options, "transport.socket.bind-address")) {
+ ret = dict_get_str (this->options,
+ "transport.socket.bind-address",
+ &optstr);
+ if (ret < 0) {
+ gf_log (GF_NFS, GF_LOG_ERROR, "Failed to parse "
+ "transport.socket.bind-address string");
+ } else {
+ this->instance_name = gf_strdup (optstr);
+ for (i = 0; i < strlen (this->instance_name); i++) {
+ if (this->instance_name[i] == '.' ||
+ this->instance_name[i] == ':')
+ this->instance_name[i] = '_';
+ }
+ }
+ }
+
if (dict_get(this->options, "transport.socket.listen-port") == NULL) {
if (nfs->override_portnum)
ret = gf_asprintf (&optstr, "%d",
@@ -1531,6 +1548,7 @@ fini (xlator_t *this)
nfs = (struct nfs_state *)this->private;
gf_msg_debug (GF_NFS, 0, "NFS service going down");
nfs_deinit_versions (&nfs->versions, this);
+ GF_FREE (this->instance_name);
return 0;
}