Posix: Optimize posix code to improve file creation

Problem: Before executing a fop in POSIX xlator it builds an internal
         path based on GFID.To validate the path it call's (l)stat
         system call and while .glusterfs is heavily loaded kernel takes
         time to lookup inode and due to that performance drops

Solution: In this patch we followed two ways to improve the performance.
          1) Keep open fd specific to first level directory(gfid[0])
             in .glusterfs, it would force to kernel keep the inodes
             from all those files in cache. In case of memory pressure
             kernel won't uncache first level inodes. We need to open
             256 fd's per brick to access the entry faster.
          2) Use at based call's to access relative path to reduce
             path based lookup time.

Note: To verify the patch we have executed kernel untar 100 times on 6
      different clients after enabling metadata group-cache and some
      other option.We were getting more than 20 percent improvement in
      kenel untar after applying the patch.

Credits: Xavi Hernandez <xhernandez@redhat.com>
Change-Id: I1643e6b01ed669b2bb148d02f4e6a8e08da45343
updates: #891
Signed-off-by: Mohit Agrawal <moagrawal@redhat.com>

1 files changed, 97 insertions, 4 deletions

diff --git a/xlators/storage/posix/src/posix-common.c b/xlators/storage/posix/src/posix-common.c
index 20a8d231181..1fc0eae2056 100644
--- a/xlators/storage/posix/src/posix-common.c
+++ b/xlators/storage/posix/src/posix-common.c
@@ -543,6 +543,30 @@ posix_create_unlink_dir(xlator_t *this)
     return 0;
 }
 
+int
+posix_create_open_directory_based_fd(xlator_t *this, int pdirfd, char *dir_name)
+{
+    int ret = -1;
+
+    ret = sys_openat(pdirfd, dir_name, (O_DIRECTORY | O_RDONLY), 0);
+    if (ret < 0 && errno == ENOENT) {
+        ret = sys_mkdirat(pdirfd, dir_name, 0700);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, P_MSG_HANDLE_CREATE,
+                   "Creating directory %s failed", dir_name);
+            goto out;
+        }
+        ret = sys_openat(pdirfd, dir_name, (O_DIRECTORY | O_RDONLY), 0);
+        if (ret < 0 && errno != EEXIST) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, P_MSG_HANDLE_CREATE,
+                   "error mkdir hash-1 %s ", dir_name);
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
 /**
  * init -
  */
@@ -579,6 +603,14 @@ posix_init(xlator_t *this)
     int force_directory = -1;
     int create_mask = -1;
     int create_directory_mask = -1;
+    char dir_handle[PATH_MAX] = {
+        0,
+    };
+    int i;
+    char fhash[4] = {
+        0,
+    };
+    int hdirfd = -1;
 
     dir_data = dict_get(this->options, "directory");
 
@@ -621,6 +653,11 @@ posix_init(xlator_t *this)
     _private->base_path = gf_strdup(dir_data->data);
     _private->base_path_length = dir_data->len - 1;
 
+    _private->dirfd = -1;
+    _private->mount_lock = -1;
+    for (i = 0; i < 256; i++)
+        _private->arrdfd[i] = -1;
+
     ret = dict_get_str(this->options, "hostname", &_private->hostname);
     if (ret) {
         _private->hostname = GF_CALLOC(256, sizeof(char), gf_common_mt_char);
@@ -893,8 +930,9 @@ posix_init(xlator_t *this)
     /* performing open dir on brick dir locks the brick dir
      * and prevents it from being unmounted
      */
-    _private->mount_lock = sys_opendir(dir_data->data);
-    if (!_private->mount_lock) {
+    _private->mount_lock = sys_open(dir_data->data, (O_DIRECTORY | O_RDONLY),
+                                    0);
+    if (_private->mount_lock < 0) {
         ret = -1;
         op_errno = errno;
         gf_msg(this->name, GF_LOG_ERROR, 0, P_MSG_DIR_OPERATION_FAILED,
@@ -938,6 +976,28 @@ posix_init(xlator_t *this)
     }
 
     this->private = (void *)_private;
+    snprintf(dir_handle, sizeof(dir_handle), "%s/%s", _private->base_path,
+             GF_HIDDEN_PATH);
+    hdirfd = posix_create_open_directory_based_fd(this, _private->mount_lock,
+                                                  dir_handle);
+    if (hdirfd < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, P_MSG_HANDLE_CREATE,
+               "error open directory failed for dir %s", dir_handle);
+        ret = -1;
+        goto out;
+    }
+    _private->dirfd = hdirfd;
+    for (i = 0; i < 256; i++) {
+        snprintf(fhash, sizeof(fhash), "%02x", i);
+        _private->arrdfd[i] = posix_create_open_directory_based_fd(this, hdirfd,
+                                                                   fhash);
+        if (_private->arrdfd[i] < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, P_MSG_HANDLE_CREATE,
+                   "error openat failed for file %s", fhash);
+            ret = -1;
+            goto out;
+        }
+    }
 
     op_ret = posix_handle_init(this);
     if (op_ret == -1) {
@@ -1101,9 +1161,27 @@ posix_init(xlator_t *this)
                    out);
 
     GF_OPTION_INIT("ctime", _private->ctime, bool, out);
+
 out:
     if (ret) {
         if (_private) {
+            if (_private->dirfd >= 0) {
+                sys_close(_private->dirfd);
+                _private->dirfd = -1;
+            }
+
+            for (i = 0; i < 256; i++) {
+                if (_private->arrdfd[i] >= 0) {
+                    sys_close(_private->arrdfd[i]);
+                    _private->arrdfd[i] = -1;
+                }
+            }
+            /*unlock brick dir*/
+            if (_private->mount_lock >= 0) {
+                (void)sys_close(_private->mount_lock);
+                _private->mount_lock = -1;
+            }
+
             GF_FREE(_private->base_path);
 
             GF_FREE(_private->hostname);
@@ -1124,6 +1202,7 @@ posix_fini(xlator_t *this)
     struct posix_private *priv = this->private;
     gf_boolean_t health_check = _gf_false;
     int ret = 0;
+    int i = 0;
 
     if (!priv)
         return;
@@ -1134,6 +1213,18 @@ posix_fini(xlator_t *this)
     }
     UNLOCK(&priv->lock);
 
+    if (priv->dirfd >= 0) {
+        sys_close(priv->dirfd);
+        priv->dirfd = -1;
+    }
+
+    for (i = 0; i < 256; i++) {
+        if (priv->arrdfd[i] >= 0) {
+            sys_close(priv->arrdfd[i]);
+            priv->arrdfd[i] = -1;
+        }
+    }
+
     if (health_check) {
         (void)gf_thread_cleanup_xint(priv->health_check);
         priv->health_check = 0;
@@ -1161,8 +1252,10 @@ posix_fini(xlator_t *this)
         priv->fsyncer = 0;
     }
     /*unlock brick dir*/
-    if (priv->mount_lock)
-        (void)sys_closedir(priv->mount_lock);
+    if (priv->mount_lock >= 0) {
+        (void)sys_close(priv->mount_lock);
+        priv->mount_lock = -1;
+    }
 
     GF_FREE(priv->base_path);
     LOCK_DESTROY(&priv->lock);