[PATCH 0/4] LVM/cmirrord support write improvement.

[PATCH 0/4] LVM/cmirrord support write improvement.

[dongmao zhang]

These patches work with dm-log-userspace enhancement.
LVM, device-mapper and cmirrord begin to support
merging mark/flush request and delaying the clear
request. This would increase the write throughput
of cluster raid1 based on my own test.

dongmao zhang (4):
  cmirrord add support_delay_clear option
  cmirrord create new function _clog_flush
  add support_delay_flush to device-mapper
  cluster LVM use flag DM_SUPPORT_DELAY_FLUSH

 daemons/cmirrord/functions.c |   60 ++++++++++++++++++++++++++++++++---------
 lib/mirror/mirrored.c        |    4 +++
 libdm/libdevmapper.h         |    3 ++
 libdm/libdm-deptree.c        |    8 +++++
 4 files changed, 62 insertions(+), 13 deletions(-)

-- 
1.7.3.4

[PATCH 1/4] cmirrord add support_delay_clear option

[dongmao zhang]

---
 daemons/cmirrord/functions.c |   20 +++++++++++++++-----
 1 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/daemons/cmirrord/functions.c b/daemons/cmirrord/functions.c
index f6e0918..17f588e 100644
--- a/daemons/cmirrord/functions.c
+++ b/daemons/cmirrord/functions.c
@@ -75,6 +75,7 @@ struct log_c {
                 FORCESYNC,      /* Force a sync to happen */
         } sync;
 
+	uint32_t support_delay_flush;
 	uint32_t state;         /* current operational state of the log */
 
 	struct dm_list mark_list;
@@ -362,7 +363,7 @@ static int find_disk_path(char *major_minor_str, char *path_rtn, int *unlink_pat
 	// return r ? -errno : 0;
 }
 
-static int _clog_ctr(char *uuid, uint64_t luid,
+static int _clog_ctr(char *uuid, uint64_t luid, uint32_t version,
 		     int argc, char **argv, uint64_t device_size)
 {
 	int i;
@@ -373,6 +374,7 @@ static int _clog_ctr(char *uuid, uint64_t luid,
 	struct log_c *lc = NULL;
 	enum sync log_sync = DEFAULTSYNC;
 	uint32_t block_on_error = 0;
+	uint32_t support_delay_flush = 0;
 
 	int disk_log;
 	char disk_path[128];
@@ -431,6 +433,8 @@ static int _clog_ctr(char *uuid, uint64_t luid,
 			log_sync = NOSYNC;
 		else if (!strcmp(argv[i], "block_on_error"))
 			block_on_error = 1;
+		else if (!strcmp(argv[i], "support_delay_flush") && version > 2)
+			support_delay_flush = 1;
 	}
 
 	lc = dm_zalloc(sizeof(*lc));
@@ -451,6 +455,7 @@ static int _clog_ctr(char *uuid, uint64_t luid,
 	lc->log_dev_failed = 0;
 	strncpy(lc->uuid, uuid, DM_UUID_LEN);
 	lc->luid = luid;
+	lc->support_delay_flush = support_delay_flush;
 
 	if (get_log(lc->uuid, lc->luid) ||
 	    get_pending_log(lc->uuid, lc->luid)) {
@@ -552,6 +557,8 @@ static int clog_ctr(struct dm_ulog_request *rq)
 	char *dev_size_str;
 	uint64_t device_size;
 
+	struct log_c *tmplc = NULL;
+
 	/* Sanity checks */
 	if (!rq->data_size) {
 		LOG_ERROR("Received constructor request with no data");
@@ -594,14 +601,17 @@ static int clog_ctr(struct dm_ulog_request *rq)
 		return -EINVAL;
 	}
 
-	r = _clog_ctr(rq->uuid, rq->luid, argc - 1, argv + 1, device_size);
+	r = _clog_ctr(rq->uuid, rq->luid, rq->version, argc - 1, argv + 1, device_size);
 
 	/* We join the CPG when we resume */
 
 	/* No returning data */
-	if ((rq->version > 1) && !strcmp(argv[0], "clustered-disk"))
-		rq->data_size = sprintf(rq->data, "%s", argv[1]) + 1;
-	else
+	if ((rq->version > 1) && !strcmp(argv[0], "clustered-disk")) {
+		rq->data_size = sprintf(rq->data, "%s", argv[1]);
+		tmplc = get_pending_log(rq->uuid, rq->luid);
+		if (tmplc && tmplc->support_delay_flush)
+			rq->data_size += sprintf(rq->data + rq->data_size, " %s", "support_delay_flush") + 1;
+	} else
 		rq->data_size = 0;
 
 	if (r) {
-- 
1.7.3.4

[PATCH 2/4] cmirrord create new function _clog_flush

[dongmao zhang]

_clog_flush is used by both clog_mark_region and clog_flush
---
 daemons/cmirrord/functions.c |   40 ++++++++++++++++++++++++++++++++--------
 1 files changed, 32 insertions(+), 8 deletions(-)

diff --git a/daemons/cmirrord/functions.c b/daemons/cmirrord/functions.c
index 17f588e..33e909f 100644
--- a/daemons/cmirrord/functions.c
+++ b/daemons/cmirrord/functions.c
@@ -1037,14 +1037,17 @@ static int clog_in_sync(struct dm_ulog_request *rq)
 }
 
 /*
- * clog_flush
- * @rq
+ * _clog_flush
+ * @lc
+ * @server
  *
+ * real flush
+ *
+ * Returns: 0 on success, -EXXX on error
  */
-static int clog_flush(struct dm_ulog_request *rq, int server)
+static int _clog_flush(struct log_c *lc, int server)
 {
 	int r = 0;
-	struct log_c *lc = get_log(rq->uuid, rq->luid);
 
 	if (!lc)
 		return -EINVAL;
@@ -1057,10 +1060,10 @@ static int clog_flush(struct dm_ulog_request *rq, int server)
 	 * if we are the server.
 	 */
 	if (server && (lc->disk_fd >= 0)) {
-		r = rq->error = write_log(lc);
+		r = write_log(lc);
 		if (r)
 			LOG_ERROR("[%s] Error writing to disk log",
-				  SHORT_UUID(lc->uuid));
+				SHORT_UUID(lc->uuid));
 		else 
 			LOG_DBG("[%s] Disk log written", SHORT_UUID(lc->uuid));
 	}
@@ -1068,6 +1071,21 @@ static int clog_flush(struct dm_ulog_request *rq, int server)
 	lc->touched = 0;
 
 	return r;
+}
+
+
+/*
+ * clog_flush
+ * @rq
+ *
+ */
+static int clog_flush(struct dm_ulog_request *rq, int server)
+{
+	int r = 0;
+	struct log_c *lc = get_log(rq->uuid, rq->luid);
+
+	r = rq->error = _clog_flush(lc, server);
+	return r;
 
 }
 
@@ -1123,7 +1141,7 @@ static int mark_region(struct log_c *lc, uint64_t region, uint32_t who)
  *
  * Returns: 0 on success, -EXXX on failure
  */
-static int clog_mark_region(struct dm_ulog_request *rq, uint32_t originator)
+static int clog_mark_region(struct dm_ulog_request *rq, uint32_t originator, int server)
 {
 	int r;
 	int count;
@@ -1149,6 +1167,12 @@ static int clog_mark_region(struct dm_ulog_request *rq, uint32_t originator)
 
 	rq->data_size = 0;
 
+	/*if kernel support delay clear flush, cmirrord could
+	 *run flush immediatly after mark region*/
+
+	if(lc->support_delay_flush)
+		return _clog_flush(lc, server);
+
 	return 0;
 }
 
@@ -1686,7 +1710,7 @@ int do_request(struct clog_request *rq, int server)
 		r = clog_flush(&rq->u_rq, server);
 		break;
 	case DM_ULOG_MARK_REGION:
-		r = clog_mark_region(&rq->u_rq, rq->originator);
+		r = clog_mark_region(&rq->u_rq, rq->originator, server);
 		break;
 	case DM_ULOG_CLEAR_REGION:
 		r = clog_clear_region(&rq->u_rq, rq->originator);
-- 
1.7.3.4

[PATCH 3/4] add support_delay_flush to device-mapper

[dongmao zhang]

---
 libdm/libdevmapper.h  |    3 +++
 libdm/libdm-deptree.c |    8 ++++++++
 2 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/libdm/libdevmapper.h b/libdm/libdevmapper.h
index b287eef..dab07d6 100644
--- a/libdm/libdevmapper.h
+++ b/libdm/libdevmapper.h
@@ -667,6 +667,9 @@ int dm_tree_node_add_mirror_target(struct dm_tree_node *node,
 #define DM_BLOCK_ON_ERROR	0x00000004	/* On error, suspend I/O */
 #define DM_CORELOG		0x00000008	/* In-memory log */
 
+/*for cluster raid1, delay flush to improve performance*/
+#define DM_SUPPORT_DELAY_FLUSH		0x00000010
+
 int dm_tree_node_add_mirror_target_log(struct dm_tree_node *node,
 					  uint32_t region_size,
 					  unsigned clustered,
diff --git a/libdm/libdm-deptree.c b/libdm/libdm-deptree.c
index 752a44b..ff13b38 100644
--- a/libdm/libdm-deptree.c
+++ b/libdm/libdm-deptree.c
@@ -2054,6 +2054,7 @@ static int _mirror_emit_segment_line(struct dm_task *dmt, struct load_segment *s
 				     char *params, size_t paramsize)
 {
 	int block_on_error = 0;
+	int support_delay_flush;
 	int handle_errors = 0;
 	int dm_log_userspace = 0;
 	struct utsname uts;
@@ -2107,6 +2108,9 @@ static int _mirror_emit_segment_line(struct dm_task *dmt, struct load_segment *s
 		 */
 		if (KERNEL_VERSION(kmaj, kmin, krel) >= KERNEL_VERSION(2, 6, 31))
 			dm_log_userspace = 1;
+
+		if(seg->flags & DM_SUPPORT_DELAY_FLUSH)
+			support_delay_flush = 1;
 	}
 
 	/* Region size */
@@ -2164,6 +2168,10 @@ static int _mirror_emit_segment_line(struct dm_task *dmt, struct load_segment *s
 	if (block_on_error)
 		EMIT_PARAMS(pos, " block_on_error");
 
+	if (dm_log_userspace && support_delay_flush)
+		EMIT_PARAMS(pos, " support_delay_flush");
+
+
 	EMIT_PARAMS(pos, " %u ", seg->mirror_area_count);
 
 	if (_emit_areas_line(dmt, seg, params, paramsize, &pos) <= 0)
-- 
1.7.3.4

[PATCH 4/4] cluster LVM use flag DM_SUPPORT_DELAY_FLUSH

[dongmao zhang]

---
 lib/mirror/mirrored.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/lib/mirror/mirrored.c b/lib/mirror/mirrored.c
index c3eb27b..af69859 100644
--- a/lib/mirror/mirrored.c
+++ b/lib/mirror/mirrored.c
@@ -378,6 +378,10 @@ static int _add_log(struct dm_pool *mem, struct lv_segment *seg,
 	if (_block_on_error_available && !(seg->status & PVMOVE))
 		log_flags |= DM_BLOCK_ON_ERROR;
 
+	/*clustered log support delay flush*/
+	if(clustered)
+		log_flags |= DM_SUPPORT_DELAY_FLUSH;
+
 	return dm_tree_node_add_mirror_target_log(node, region_size, clustered, log_dlid, area_count, log_flags);
 }
 
-- 
1.7.3.4