arbiter: write performance improvement

Problem: The throughput for a 'dd' workload was much less for arbiter
configuration when compared to normal replica-3 volume. There were 2
issues:

i)arbiter_writev was using the request dict as response dict while
unwinding, leading to incorect GLUSTERFS_WRITE_IS_APPEND and
GLUSTERFS_OPEN_FD_COUNT values (=4), leading to immediate post-ops
because is_afr_delayed_changelog_post_op_needed() failed due to
afr_are_multiple_fds_opened() check.

ii) The arbiter code in afr was setting local->transaction.{start and len} =0
to take full file locks. What this meant was even for simultaenous but
non-overlapping writevs, afr_transaction_eager_lock_init() was not
happening because afr_locals_overlap() always stays true. Consequently
is_afr_delayed_changelog_post_op_needed() failed due to
local->delayed_post_op not being set.

Fix:
i) Send appropriate response dict values in arbiter_writev.
ii) Modify flock params instead of local->transaction.{start and len} to
take full file locks in the transaction.

Also changed _fill_writev_xdata() in posix to fill rsp_xdata for
whatever key is requested for.

Change-Id: I1c5fc5e98aba49ade540bb441a022e65b753432a
BUG: 1324004
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reported-by: Robert Rauch <robert.rauch@gns-systems.de>
Reported-by: Russel Purinton <russell.purinton@gmail.com>
Reviewed-on: http://review.gluster.org/13906
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Smoke: Gluster Build System <jenkins@build.gluster.com>
CentOS-regression: Gluster Build System <jenkins@build.gluster.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>

4 files changed, 82 insertions, 34 deletions

diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index e0928425c6a..15bae87a4f4 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -412,7 +412,6 @@ afr_do_writev (call_frame_t *frame, xlator_t *this)
 {
         call_frame_t    *transaction_frame = NULL;
         afr_local_t     *local             = NULL;
-        afr_private_t   *priv              = NULL;
         int             ret   = -1;
         int             op_errno = ENOMEM;
 
@@ -421,7 +420,6 @@ afr_do_writev (call_frame_t *frame, xlator_t *this)
                 goto out;
 
         local = frame->local;
-        priv = this->private;
         transaction_frame->local = local;
 	frame->local = NULL;
 
@@ -451,12 +449,6 @@ afr_do_writev (call_frame_t *frame, xlator_t *this)
                 local->transaction.start   = local->cont.writev.offset;
                 local->transaction.len     = iov_length (local->cont.writev.vector,
                                                          local->cont.writev.count);
-
-                /*Lock entire file to avoid network split brains.*/
-                if (priv->arbiter_count == 1) {
-                        local->transaction.start   = 0;
-                        local->transaction.len     = 0;
-                }
         }
 
         ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 92f68b91113..9c27f07b5d4 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -1356,16 +1356,24 @@ afr_post_lower_unlock_cbk (call_frame_t *frame, xlator_t *this)
 
 
 int
-afr_set_transaction_flock (afr_local_t *local)
+afr_set_transaction_flock (xlator_t *this, afr_local_t *local)
 {
         afr_internal_lock_t *int_lock = NULL;
         afr_inodelk_t       *inodelk  = NULL;
+        afr_private_t       *priv     = NULL;
 
         int_lock = &local->internal_lock;
         inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        priv = this->private;
 
-        inodelk->flock.l_len   = local->transaction.len;
-        inodelk->flock.l_start = local->transaction.start;
+        if (priv->arbiter_count) {
+                /*Lock entire file to avoid network split brains.*/
+                inodelk->flock.l_len   = 0;
+                inodelk->flock.l_start = 0;
+        } else {
+                inodelk->flock.l_len   = local->transaction.len;
+                inodelk->flock.l_start = local->transaction.start;
+        }
         inodelk->flock.l_type  = F_WRLCK;
 
         return 0;
@@ -1386,7 +1394,7 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this)
         switch (local->transaction.type) {
         case AFR_DATA_TRANSACTION:
         case AFR_METADATA_TRANSACTION:
-                afr_set_transaction_flock (local);
+                afr_set_transaction_flock (this, local);
 
                 int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk;
 
diff --git a/xlators/features/arbiter/src/arbiter.c b/xlators/features/arbiter/src/arbiter.c
index b404597be9d..786f60b7bc9 100644
--- a/xlators/features/arbiter/src/arbiter.c
+++ b/xlators/features/arbiter/src/arbiter.c
@@ -128,8 +128,7 @@ arbiter_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
         }
         buf = ctx->iattbuf;
 unwind:
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, buf, buf,
-                             xdata);
+        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, buf, buf, NULL);
         return 0;
 }
 
@@ -152,10 +151,48 @@ arbiter_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
         buf = ctx->iattbuf;
 unwind:
         STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno, buf, buf,
-                             xdata);
+                             NULL);
         return 0;
 }
 
+dict_t*
+arbiter_fill_writev_xdata (fd_t *fd, dict_t *xdata, xlator_t *this)
+{
+        dict_t  *rsp_xdata = NULL;
+        int32_t ret = 0;
+        int is_append = 1;
+
+        if (!fd || !fd->inode || gf_uuid_is_null (fd->inode->gfid)) {
+                goto out;
+        }
+
+        if (!xdata)
+                goto out;
+
+        rsp_xdata = dict_new();
+        if (!rsp_xdata)
+                goto out;
+
+        if (dict_get (xdata, GLUSTERFS_OPEN_FD_COUNT)) {
+                ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_OPEN_FD_COUNT,
+                                       fd->inode->fd_count);
+                if (ret < 0) {
+                        gf_msg_debug (this->name, 0, "Failed to set dict value"
+                                      " for GLUSTERFS_OPEN_FD_COUNT");
+                }
+        }
+        if (dict_get (xdata, GLUSTERFS_WRITE_IS_APPEND)) {
+                ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_WRITE_IS_APPEND,
+                                       is_append);
+                if (ret < 0) {
+                        gf_msg_debug (this->name, 0, "Failed to set dict value"
+                                      " for GLUSTERFS_WRITE_IS_APPEND");
+                }
+        }
+out:
+        return rsp_xdata;
+}
+
 int32_t
 arbiter_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
                 struct iovec *vector, int32_t count, off_t off, uint32_t flags,
@@ -163,6 +200,7 @@ arbiter_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
 {
         arbiter_inode_ctx_t *ctx      = NULL;
         struct iatt         *buf      = NULL;
+        dict_t           *rsp_xdata   = NULL;
         int                  op_ret   = 0;
         int                  op_errno = 0;
 
@@ -174,8 +212,12 @@ arbiter_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         }
         buf = ctx->iattbuf;
         op_ret = iov_length (vector, count);
+        rsp_xdata = arbiter_fill_writev_xdata (fd, xdata, this);
 unwind:
-        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, buf, buf, xdata);
+        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, buf, buf,
+                             rsp_xdata);
+        if (rsp_xdata)
+                dict_unref (rsp_xdata);
         return 0;
 }
 
@@ -196,8 +238,7 @@ arbiter_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd,
         }
         buf = ctx->iattbuf;
 unwind:
-        STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, buf, buf,
-                            xdata);
+        STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, buf, buf, NULL);
         return 0;
 }
 
@@ -218,7 +259,7 @@ arbiter_discard (call_frame_t *frame, xlator_t *this, fd_t *fd,
         }
         buf = ctx->iattbuf;
 unwind:
-        STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, buf, buf, xdata);
+        STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, buf, buf, NULL);
         return 0;
 }
 
@@ -239,7 +280,7 @@ arbiter_zerofill (call_frame_t *frame, xlator_t *this, fd_t *fd,
         }
         buf = ctx->iattbuf;
 unwind:
-        STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, buf, buf, xdata);
+        STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, buf, buf, NULL);
         return 0;
 }
 
diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c
index f72c13d9066..8cce4be1c34 100644
--- a/xlators/storage/posix/src/posix.c
+++ b/xlators/storage/posix/src/posix.c
@@ -3055,28 +3055,35 @@ _fill_writev_xdata (fd_t *fd, dict_t *xdata, xlator_t *this, int is_append)
                 goto out;
         }
 
-        if (!xdata || !dict_get (xdata, GLUSTERFS_OPEN_FD_COUNT))
+        if (!xdata)
                 goto out;
 
         rsp_xdata = dict_new();
         if (!rsp_xdata)
                 goto out;
 
-        ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_OPEN_FD_COUNT,
-                               fd->inode->fd_count);
-        if (ret < 0) {
-                gf_msg (this->name, GF_LOG_WARNING, 0, P_MSG_DICT_SET_FAILED,
-                        "%s: Failed to set dictionary value for %s",
-                        uuid_utoa (fd->inode->gfid), GLUSTERFS_OPEN_FD_COUNT);
+        if (dict_get (xdata, GLUSTERFS_OPEN_FD_COUNT)) {
+                ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_OPEN_FD_COUNT,
+                                       fd->inode->fd_count);
+                if (ret < 0) {
+                        gf_msg (this->name, GF_LOG_WARNING, 0,
+                                P_MSG_DICT_SET_FAILED, "%s: Failed to set "
+                                "dictionary value for %s",
+                                uuid_utoa (fd->inode->gfid),
+                                GLUSTERFS_OPEN_FD_COUNT);
+                }
         }
 
-        ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_WRITE_IS_APPEND,
-                               is_append);
-        if (ret < 0) {
-                gf_msg (this->name, GF_LOG_WARNING, 0, P_MSG_DICT_SET_FAILED,
-                        "%s: Failed to set dictionary value for %s",
-                        uuid_utoa (fd->inode->gfid),
-                        GLUSTERFS_WRITE_IS_APPEND);
+        if (dict_get (xdata, GLUSTERFS_WRITE_IS_APPEND)) {
+                ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_WRITE_IS_APPEND,
+                                       is_append);
+                if (ret < 0) {
+                        gf_msg (this->name, GF_LOG_WARNING, 0,
+                                P_MSG_DICT_SET_FAILED, "%s: Failed to set "
+                                "dictionary value for %s",
+                                uuid_utoa (fd->inode->gfid),
+                                GLUSTERFS_WRITE_IS_APPEND);
+                }
         }
 out:
         return rsp_xdata;