ec: Optimize read/write performance

This patch significantly improves performance of read/write
operations on a dispersed volume by reusing previous inodelk/
entrylk operations on the same inode/entry. This reduces the
latency of each individual operation considerably.

Inode version and size are also updated when needed instead
of on each request. This gives an additional boost.

Change-Id: I4b98d5508c86b53032e16e295f72a3f83fd8fcac
BUG: 1122586
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
Reviewed-on: http://review.gluster.org/8369
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Dan Lambright <dlambrig@redhat.com>

1 files changed, 71 insertions, 25 deletions

diff --git a/xlators/cluster/ec/src/ec-dir-write.c b/xlators/cluster/ec/src/ec-dir-write.c
index dc1d94e37e8..a28b7ad1937 100644
--- a/xlators/cluster/ec/src/ec-dir-write.c
+++ b/xlators/cluster/ec/src/ec-dir-write.c
@@ -181,7 +181,8 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state)
         /* Fall through */
 
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_DISPATCH;
 
@@ -245,11 +246,7 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state)
                                  cbk->xdata);
             }
 
-            if (cbk->op_ret >= 0)
-            {
-                return EC_STATE_UPDATE_SIZE_AND_VERSION;
-            }
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_DISPATCH:
@@ -263,14 +260,14 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state)
                                  NULL, NULL, NULL, NULL, NULL, NULL);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
-        case EC_STATE_UPDATE_SIZE_AND_VERSION:
-            ec_update_size_version(fop);
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 1);
 
             return EC_STATE_UNLOCK;
 
-        case -EC_STATE_UPDATE_SIZE_AND_VERSION:
         case -EC_STATE_UNLOCK:
         case EC_STATE_UNLOCK:
             ec_unlock(fop);
@@ -468,7 +465,8 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state)
             // Parent entry of fop->loc[0] should be locked, but I don't
             // receive enough information to do it (fop->loc[0].parent is
             // NULL).
-            ec_lock_entry(fop, &fop->loc[1]);
+            ec_lock_prepare_entry(fop, &fop->loc[1]);
+            ec_lock(fop);
 
             return EC_STATE_GET_SIZE_AND_VERSION;
 
@@ -531,7 +529,7 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state)
                                &cbk->iatt[1], &cbk->iatt[2], cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_GET_SIZE_AND_VERSION:
@@ -546,6 +544,12 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state)
                                NULL, NULL, NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -732,7 +736,8 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_DISPATCH;
 
@@ -785,7 +790,7 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state)
                                 &cbk->iatt[1], &cbk->iatt[2], cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_DISPATCH:
@@ -799,6 +804,12 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state)
                                 NULL, NULL, NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -982,7 +993,8 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_DISPATCH;
 
@@ -1035,7 +1047,7 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state)
                                 &cbk->iatt[1], &cbk->iatt[2], cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_DISPATCH:
@@ -1049,6 +1061,12 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state)
                                 NULL, NULL, NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -1230,8 +1248,9 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
-            ec_lock_entry(fop, &fop->loc[1]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[1]);
+            ec_lock(fop);
 
             return EC_STATE_GET_SIZE_AND_VERSION;
 
@@ -1292,7 +1311,7 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state)
                                  cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_GET_SIZE_AND_VERSION:
@@ -1307,6 +1326,12 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state)
                                  NULL, NULL, NULL, NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -1479,7 +1504,8 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_DISPATCH;
 
@@ -1524,7 +1550,7 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state)
                                 cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_DISPATCH:
@@ -1538,6 +1564,12 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state)
                                 NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -1719,7 +1751,8 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_DISPATCH;
 
@@ -1772,7 +1805,7 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state)
                                   &cbk->iatt[1], &cbk->iatt[2], cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_DISPATCH:
@@ -1786,6 +1819,12 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state)
                                   NULL, NULL, NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK:
@@ -1963,7 +2002,8 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state)
     {
         case EC_STATE_INIT:
         case EC_STATE_LOCK:
-            ec_lock_entry(fop, &fop->loc[0]);
+            ec_lock_prepare_entry(fop, &fop->loc[0]);
+            ec_lock(fop);
 
             return EC_STATE_GET_SIZE_AND_VERSION;
 
@@ -2013,7 +2053,7 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state)
                                  cbk->xdata);
             }
 
-            return EC_STATE_UNLOCK;
+            return EC_STATE_LOCK_REUSE;
 
         case -EC_STATE_LOCK:
         case -EC_STATE_GET_SIZE_AND_VERSION:
@@ -2028,6 +2068,12 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state)
                                  NULL, NULL, NULL);
             }
 
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop, 0);
+
             return EC_STATE_UNLOCK;
 
         case -EC_STATE_UNLOCK: