diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index ae83eb2a9f8434192e60da1a4bc0cdc45b850a43..040c6e3612324cfb9c836a719cd7e2bbcc2b362e 100644
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -270,7 +270,7 @@ Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
Severity : minor
Bugzilla : 11706
-Description: service threads may hog cpus when ther eis a lot of requests
+Description: service threads may hog cpus when there are a lot of requests
coming
Details : Insert cond_resched to give other threads a chance to use some of
the cpu
@@ -285,6 +285,14 @@ Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
initialization will be done asynchronausly, so there will be race
between add unlink log and unlink log initialization.
+Severity : normal
+Bugzilla : 12597
+Description: brw_stats were being printed incorrectly
+Details : brw_stats were being printed as log2 but all of them were not
+ recorded as log2. Also remove some code duplication arising from
+ filter_tally_{read,write}.
+
+
--------------------------------------------------------------------------------
2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
diff --git a/lustre/obdfilter/filter_internal.h b/lustre/obdfilter/filter_internal.h
index 093976febb7a8373aaf6c5b95a3e07e838e01730..9fb48ad4fa9c62ac8d532d0c82b51fdc25ed48c4 100644
--- a/lustre/obdfilter/filter_internal.h
+++ b/lustre/obdfilter/filter_internal.h
@@ -174,23 +174,13 @@ int filter_recov_log_mds_ost_cb(struct llog_handle *llh,
struct llog_rec_hdr *rec, void *data);
#ifdef LPROCFS
-void filter_tally_write(struct obd_export *exp, struct page **pages,
- int nr_pages, unsigned long *blocks,
- int blocks_per_page);
-void filter_tally_read(struct obd_export *exp, struct page **pages,
- int nr_pages, unsigned long *blocks,
- int blocks_per_page);
+void filter_tally(struct obd_export *exp, struct page **pages, int nr_pages,
+ unsigned long *blocks, int blocks_per_page, int wr);
int lproc_filter_attach_seqstat(struct obd_device *dev);
#else
-static inline void filter_tally_write(struct obd_export *exp,
- struct page **pages, int nr_pages,
- unsigned long *blocks, int blocks_per_page) {}
-static inline void filter_tally_read(struct obd_export *exp,
- struct page **pages, int nr_pages,
- unsigned long *blocks, int blocks_per_page) {}
-static inline void filter_tally_read(struct filter_obd *filter,
- struct page **pages, int nr_pages,
- unsigned long *blocks, int blocks_per_page) {}
+static inline void filter_tally(struct obd_export *exp, struct page **pages,
+ int nr_pages, unsigned long *blocks,
+ int blocks_per_page, int wr) {}
static inline int lproc_filter_attach_seqstat(struct obd_device *dev) {}
#endif
diff --git a/lustre/obdfilter/filter_io_24.c b/lustre/obdfilter/filter_io_24.c
index 19645c83d91db99a250e0c3676544fb9e1f80341..c126aa5a3c28d6bd8a63cb20fa7dc6e2247dc40a 100644
--- a/lustre/obdfilter/filter_io_24.c
+++ b/lustre/obdfilter/filter_io_24.c
@@ -240,9 +240,9 @@ int filter_direct_io(int rw, struct dentry *dchild, struct filter_iobuf *buf,
if (rw == OBD_BRW_WRITE) {
if (rc == 0) {
- filter_tally_write(exp, iobuf->maplist, iobuf->nr_pages,
- KIOBUF_GET_BLOCKS(iobuf),
- blocks_per_page);
+ filter_tally(exp, iobuf->maplist, iobuf->nr_pages,
+ KIOBUF_GET_BLOCKS(iobuf), blocks_per_page,
+ 1);
if (attr->ia_size > inode->i_size)
attr->ia_valid |= ATTR_SIZE;
@@ -264,8 +264,8 @@ int filter_direct_io(int rw, struct dentry *dchild, struct filter_iobuf *buf,
if (rc)
GOTO(cleanup, rc);
} else {
- filter_tally_read(exp, iobuf->maplist, iobuf->nr_pages,
- KIOBUF_GET_BLOCKS(iobuf), blocks_per_page);
+ filter_tally(exp, iobuf->maplist, iobuf->nr_pages,
+ KIOBUF_GET_BLOCKS(iobuf), blocks_per_page, 0);
}
rc = filter_clear_page_cache(inode, iobuf);
diff --git a/lustre/obdfilter/filter_io_26.c b/lustre/obdfilter/filter_io_26.c
index eeae4bdf6f2d64c23ed5ad526b8ed998f4bcf6f0..64881e5bc6dc231cfc30975e583ea3ed9c021652 100644
--- a/lustre/obdfilter/filter_io_26.c
+++ b/lustre/obdfilter/filter_io_26.c
@@ -523,9 +523,9 @@ int filter_direct_io(int rw, struct dentry *dchild, struct filter_iobuf *iobuf,
if (rw == OBD_BRW_WRITE) {
if (rc == 0) {
- filter_tally_write(exp, iobuf->dr_pages,
- iobuf->dr_npages, iobuf->dr_blocks,
- blocks_per_page);
+ filter_tally(exp, iobuf->dr_pages,
+ iobuf->dr_npages, iobuf->dr_blocks,
+ blocks_per_page, 1);
if (attr->ia_size > inode->i_size)
attr->ia_valid |= ATTR_SIZE;
rc = fsfilt_setattr(obd, dchild,
@@ -547,9 +547,8 @@ int filter_direct_io(int rw, struct dentry *dchild, struct filter_iobuf *iobuf,
if (rc != 0)
RETURN(rc);
} else if (rc == 0) {
- filter_tally_read(exp, iobuf->dr_pages,
- iobuf->dr_npages, iobuf->dr_blocks,
- blocks_per_page);
+ filter_tally(exp, iobuf->dr_pages, iobuf->dr_npages,
+ iobuf->dr_blocks, blocks_per_page, 0);
}
rc = filter_clear_page_cache(inode, iobuf);
diff --git a/lustre/obdfilter/lproc_obdfilter.c b/lustre/obdfilter/lproc_obdfilter.c
index 9d3b8e125f0effab564a8a264c6026198e6781ed..c40ba57495b37196da5347581b101cdc9e107eb2 100644
--- a/lustre/obdfilter/lproc_obdfilter.c
+++ b/lustre/obdfilter/lproc_obdfilter.c
@@ -216,8 +216,8 @@ static struct lprocfs_vars lprocfs_module_vars[] = {
{ 0 }
};
-void filter_tally_write(struct obd_export *exp, struct page **pages,
- int nr_pages, unsigned long *blocks,int blocks_per_page)
+void filter_tally(struct obd_export *exp, struct page **pages, int nr_pages,
+ unsigned long *blocks, int blocks_per_page, int wr)
{
struct filter_obd *filter = &exp->exp_obd->u.filter;
struct filter_export_data *fed = &exp->exp_filter_data;
@@ -230,50 +230,10 @@ void filter_tally_write(struct obd_export *exp, struct page **pages,
if (nr_pages == 0)
return;
- lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_W_PAGES],
+ lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_R_PAGES + wr],
nr_pages);
- lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_W_PAGES], nr_pages);
-
- while (nr_pages-- > 0) {
- if (last_page && (*pages)->index != (last_page->index + 1))
- discont_pages++;
- last_page = *pages;
- pages++;
- for (i = 0; i < blocks_per_page; i++) {
- if (last_block && *blocks != (*last_block + 1))
- discont_blocks++;
- last_block = blocks++;
- }
- }
-
- lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_W_DISCONT_PAGES],
- discont_pages);
- lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_W_DISCONT_BLOCKS],
- discont_blocks);
-
- lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_W_DISCONT_PAGES],
- discont_pages);
- lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_W_DISCONT_BLOCKS],
- discont_blocks);
-}
-
-void filter_tally_read(struct obd_export *exp, struct page **pages,
- int nr_pages, unsigned long *blocks, int blocks_per_page)
-{
- struct filter_obd *filter = &exp->exp_obd->u.filter;
- struct filter_export_data *fed = &exp->exp_filter_data;
- struct page *last_page = NULL;
- unsigned long *last_block = NULL;
- unsigned long discont_pages = 0;
- unsigned long discont_blocks = 0;
- int i;
-
- if (nr_pages == 0)
- return;
-
- lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_R_PAGES],
+ lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_R_PAGES + wr],
nr_pages);
- lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_R_PAGES], nr_pages);
while (nr_pages-- > 0) {
if (last_page && (*pages)->index != (last_page->index + 1))
@@ -287,40 +247,51 @@ void filter_tally_read(struct obd_export *exp, struct page **pages,
}
}
- lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_PAGES],
+ lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_PAGES +wr],
discont_pages);
- lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_BLOCKS],
- discont_blocks);
-
- lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_PAGES],
+ lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_PAGES + wr],
discont_pages);
- lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_BLOCKS],
+ lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_BLOCKS+wr],
+ discont_blocks);
+ lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_BLOCKS + wr],
discont_blocks);
}
#define pct(a,b) (b ? a * 100 / b : 0)
-static void display_brw_stats(struct seq_file *seq, struct obd_histogram *read,
- struct obd_histogram *write)
+static void display_brw_stats(struct seq_file *seq, char *name, char *units,
+ struct obd_histogram *read, struct obd_histogram *write, int log2)
{
- unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
+ unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
int i;
+ seq_printf(seq, "\n%26s read | write\n", " ");
+ seq_printf(seq, "%-22s %-5s %% cum %% | %-5s %% cum %%\n",
+ name, units, units);
+
read_tot = lprocfs_oh_sum(read);
write_tot = lprocfs_oh_sum(write);
-
- read_cum = 0;
- write_cum = 0;
for (i = 0; i < OBD_HIST_MAX; i++) {
- unsigned long r = read->oh_buckets[i];
- unsigned long w = write->oh_buckets[i];
+ r = read->oh_buckets[i];
+ w = write->oh_buckets[i];
read_cum += r;
write_cum += w;
- seq_printf(seq, "%u:\t\t%10lu %3lu %3lu | %10lu %3lu %3lu\n",
- 1 << i, r, pct(r, read_tot),
- pct(read_cum, read_tot), w,
- pct(w, write_tot),
- pct(write_cum, write_tot));
+ if (read_cum == 0 && write_cum == 0)
+ continue;
+
+ if (!log2)
+ seq_printf(seq, "%u", i);
+ else if (i < 10)
+ seq_printf(seq, "%u", 1<<i);
+ else if (i < 20)
+ seq_printf(seq, "%uK", 1<<(i-10));
+ else
+ seq_printf(seq, "%uM", 1<<(i-20));
+
+ seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
+ r, pct(r, read_tot), pct(read_cum, read_tot),
+ w, pct(w, write_tot), pct(write_cum, write_tot));
+
if (read_cum == read_tot && write_cum == write_tot)
break;
}
@@ -329,91 +300,44 @@ static void display_brw_stats(struct seq_file *seq, struct obd_histogram *read,
static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
{
struct timeval now;
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
- int i;
-#endif
-
- do_gettimeofday(&now);
/* this sampling races with updates */
-
+ do_gettimeofday(&now);
seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
now.tv_sec, now.tv_usec);
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "pages per brw brws %% cum %% |");
- seq_printf(seq, " rpcs %% cum %%\n");
-
- display_brw_stats(seq, &brw_stats->hist[BRW_R_PAGES], &brw_stats->hist[BRW_W_PAGES]);
-
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "discont pages rpcs %% cum %% |");
- seq_printf(seq, " rpcs %% cum %%\n");
+ display_brw_stats(seq, "pages per bulk r/w", "rpcs",
+ &brw_stats->hist[BRW_R_PAGES],
+ &brw_stats->hist[BRW_W_PAGES], 1);
- display_brw_stats(seq, &brw_stats->hist[BRW_R_DISCONT_PAGES],
- &brw_stats->hist[BRW_W_DISCONT_PAGES]);
+ display_brw_stats(seq, "discontiguous pages", "rpcs",
+ &brw_stats->hist[BRW_R_DISCONT_PAGES],
+ &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "discont blocks rpcs %% cum %% |");
- seq_printf(seq, " rpcs %% cum %%\n");
+ display_brw_stats(seq, "discontiguous blocks", "rpcs",
+ &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
+ &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
- display_brw_stats(seq, &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
- &brw_stats->hist[BRW_W_DISCONT_BLOCKS]);
-
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "dio frags rpcs %% cum %% |");
- seq_printf(seq, " rpcs %% cum %%\n");
-
- display_brw_stats(seq, &brw_stats->hist[BRW_R_DIO_FRAGS],
- &brw_stats->hist[BRW_W_DIO_FRAGS]);
+ display_brw_stats(seq, "disk fragmented I/Os", "ios",
+ &brw_stats->hist[BRW_R_DIO_FRAGS],
+ &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "disk ios in flight ios %% cum %% |");
- seq_printf(seq, " rpcs %% cum %%\n");
-
- display_brw_stats(seq, &brw_stats->hist[BRW_R_RPC_HIST],
- &brw_stats->hist[BRW_W_RPC_HIST]);
-
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "io time (1/%ds) rpcs %% cum %% |", HZ);
- seq_printf(seq, " rpcs %% cum %%\n");
-
- display_brw_stats(seq, &brw_stats->hist[BRW_R_IO_TIME],
- &brw_stats->hist[BRW_W_IO_TIME]);
-
- seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
- seq_printf(seq, "disk I/O size count %% cum %% |");
- seq_printf(seq, " count %% cum %%\n");
-
- read_tot = lprocfs_oh_sum(&brw_stats->hist[BRW_R_DISK_IOSIZE]);
- write_tot = lprocfs_oh_sum(&brw_stats->hist[BRW_W_DISK_IOSIZE]);
-
- read_cum = 0;
- write_cum = 0;
- for (i = 0; i < OBD_HIST_MAX; i++) {
- unsigned long r = brw_stats->hist[BRW_R_DISK_IOSIZE].oh_buckets[i];
- unsigned long w = brw_stats->hist[BRW_W_DISK_IOSIZE].oh_buckets[i];
-
- read_cum += r;
- write_cum += w;
- if (read_cum == 0 && write_cum == 0)
- continue;
-
- if (i < 10)
- seq_printf(seq, "%u", 1<<i);
- else if (i < 20)
- seq_printf(seq, "%uK", 1<<(i-10));
- else
- seq_printf(seq, "%uM", 1<<(i-20));
-
- seq_printf(seq, ":\t\t%10lu %3lu %3lu | %10lu %3lu %3lu\n",
- r, pct(r, read_tot), pct(read_cum, read_tot),
- w, pct(w, write_tot), pct(write_cum, write_tot));
- if (read_cum == read_tot && write_cum == write_tot)
- break;
+ display_brw_stats(seq, "disk I/Os in flight", "ios",
+ &brw_stats->hist[BRW_R_RPC_HIST],
+ &brw_stats->hist[BRW_W_RPC_HIST], 0);
+
+ {
+ char title[24];
+ sprintf(title, "I/O time (1/%ds)", HZ);
+ display_brw_stats(seq, title, "ios",
+ &brw_stats->hist[BRW_R_IO_TIME],
+ &brw_stats->hist[BRW_W_IO_TIME], 1);
}
+
+ display_brw_stats(seq, "disk I/O size", "ios",
+ &brw_stats->hist[BRW_R_DISK_IOSIZE],
+ &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
#endif
}