DAOS-4763 dfs: New API for anchor split to allow parallelizing dfs_readdir

src/client/dfs/dfs.c

@@ -30,6 +30,7 @@
 #include <daos/common.h>
 #include <daos/event.h>
 #include <daos/container.h>
+#include <daos/object.h>
 #include <daos/array.h>
 
 #include "daos.h"
@@ -3956,3 +3957,49 @@ dfs_umount_root_cont(dfs_t *dfs)
 	rc = daos_cont_close(coh, NULL);
 	return daos_der2errno(rc);
 }
+
+int
+dfs_obj_anchor_split(dfs_obj_t *obj, uint32_t *nr, daos_anchor_t *anchors)
+{
+	struct daos_obj_layout	*layout;
+	int			rc;
+
+	if (obj == NULL || nr == NULL || !S_ISDIR(obj->mode))
+		return EINVAL;
+
+	rc = dc_obj_layout_get(obj->oh, &layout);
+	if (rc)
+		return rc;
+
+	/** TBD - support more than per shard iteration */
+	if (*nr !=0 && *nr != layout->ol_nr) {
+		D_ERROR("For now, num anchors should be the same as what is"
+			"reported as optimal\n");
+		D_GOTO(out, rc = ENOSYS);
+	}
+
+	*nr = layout->ol_nr;
+
+	if (anchors) {
+		uint32_t i;
+
+		for (i = 0; i < layout->ol_nr; i++) {
+			daos_anchor_set_zero(&anchors[i]);
+			dc_obj_shard2anchor(&anchors[i], i);
+			daos_anchor_set_flags(&anchors[i], DIOF_TO_SPEC_SHARD);
+		}
+	}
+out:
+	daos_obj_layout_free(layout);
+	return rc;
+}
+
+int
+dfs_obj_anchor_set(dfs_obj_t *obj, uint32_t index, daos_anchor_t *anchor)
+{
+	/** TBD - support more than per shard iteration */
+	daos_anchor_set_zero(anchor);
+	dc_obj_shard2anchor(anchor, index);
+	daos_anchor_set_flags(anchor, DIOF_TO_SPEC_SHARD);
+	return 0;
+}

src/include/daos_fs.h

@@ -460,6 +460,44 @@ int
 dfs_iterate(dfs_t *dfs, dfs_obj_t *obj, daos_anchor_t *anchor,
 	    uint32_t *nr, size_t size, dfs_filler_cb_t op, void *udata);
 
+/**
+ * Provide a function for large directories to split an anchor to be able to
+ * execute a parallel readdir or iterate. This routine suggests the optimal
+ * number of anchors to use instead of just 1 and optionally returns all those
+ * anchors. The user would allocate the array of anchors after querying the
+ * number of anchors needed. Alternatively, user does not provide an array and
+ * can call dfs_obj_anchor_set() for every anchor to set.
+ *
+ * The user could suggest how many anchors to split the iteration over. This
+ * feature is not supported yet.
+ *
+ * \param[in]	obj	Dir object to split anchor for.
+ * \param[in/out]
+ *		nr	[in]: Number of anchors requested and allocated in
+ *			\a anchors. Pass 0 for DAOS to recommend split num.
+ *			[out]: Number of anchors recommended if 0 is passed in.
+ * \param[in]	anchors	Optional array of anchors that are split.
+ *
+ * \return		0 on success, errno code on failure.
+ */
+int
+dfs_obj_anchor_split(dfs_obj_t *obj, uint32_t *nr, daos_anchor_t *anchors);
+
+/**
+ * Set an anchor with an index based on split done with dfs_obj_anchor_split.
+ * The anchor passed will be re-intialized and set to start and finish iteration
+ * based on the specified index. 
+ *
+ * \param[in]   obj     Dir object to split anchor for.
+ * \param[in]	index	Index of set this anchor for iteration.
+ * \param[in,out]
+ *		anchor	Hash anchor to set.
+ *
+ * \return              0 on success, errno code on failure.
+ */
+int
+dfs_obj_anchor_set(dfs_obj_t *obj, uint32_t index, daos_anchor_t *anchor);
+
 /**
  * Create a directory.
  *

src/object/cli_obj.c

@@ -2155,7 +2155,8 @@ obj_list_dkey_cb(tse_task_t *task, struct obj_list_arg *arg, unsigned int opc)
 
 	if (!daos_anchor_is_eof(anchor)) {
 		D_DEBUG(DB_IO, "More keys in shard %d\n", shard);
-	} else if ((shard < obj->cob_shards_nr - grp_size)) {
+	} else if (!(daos_anchor_get_flags(anchor) & DIOF_TO_SPEC_SHARD) &&
+		   (shard < obj->cob_shards_nr - grp_size)) {
 		shard += grp_size;
 		D_DEBUG(DB_IO, "next shard %d grp %d nr %u\n",
 			shard, grp_size, obj->cob_shards_nr);