DAOS-3114 vos: Consolidate timestamp cache APIs (#3200)

Simplify the timestamp cache some by consolidating APIs. There is now one API for checking for conflicts and one API for updating the read timestamps. This is preparation for adding more read timestamp updates for iteration and query APIs and for adding epoch uncertainty checks. More simplification is probably needed but this is a step in that direction. Update the mvcc test to enable same transaction tests. A few changes were required for this to work *Change the code so that it only calls vts_dtx_begin/end/commit/abort once per transaction and uses sequence number. *Disable 5 cases that require the punch model change patch that is blocked by a rebuild bug. *Fixes a bug that was clearing the uuids in the timestamp cache *Disable cases where the first operation is a failing conditional update or punch in the same transaction. These cases require more work in *VOS to support *Encapsulate vos_publish/cancel inside of vos_tx_end *Defer free for same transaction overwrites *Add full dtx_id to timestamps for comparisons Signed-off-by: Jeff Olivier <[email protected]>
daos-stack · Nov 11, 2020 · 0fd7fb7 · 0fd7fb7
1 parent 88fbcb8
commit 0fd7fb7
Show file tree

Hide file tree

Showing 28 changed files with 907 additions and 658 deletions.
diff --git a/ci/patch_src_in_place b/ci/patch_src_in_place
@@ -1,7 +1,7 @@
 #!/bin/sh
 
 # shellcheck disable=SC2046,SC2035
-codespell -w --ignore-words-list nd,uint,ths,ba,creat,te,cas,mapp,pres,crashers,dout,tre,reord,mimick,cloneable,keypair,bject,tread,cancelled --builtin clear,rare,informal,names,en-GB_to_en-US --skip *.png,*.PNG,*.pyc,src/rdb/raft/*,src/control/vendor/*,RSA.golden $(git ls-tree --full-tree --name-only HEAD)
+codespell -w --ignore-words-list dedup,nd,uint,ths,ba,creat,te,cas,mapp,pres,crashers,dout,tre,reord,mimick,cloneable,keypair,bject,tread,cancelled --builtin clear,rare,informal,names,en-GB_to_en-US --skip *.png,*.PNG,*.pyc,src/rdb/raft/*,src/control/vendor/*,RSA.golden $(git ls-tree --full-tree --name-only HEAD)
 
 # The return code of codespell is the number of works it could not correct
 # because of multiple options.  We could report on these but they're rare

diff --git a/ci/unit/test_main_node.sh b/ci/unit/test_main_node.sh
@@ -3,7 +3,7 @@
 # This is a script to be run by the unit/test_main.sh to run a test
 # on a CI node.
 
-set -ex
+set -x
 
 sudo bash -c 'echo 1 > /proc/sys/kernel/sysrq'
 if grep /mnt/daos\  /proc/mounts; then

diff --git a/src/common/mem.c b/src/common/mem.c
@@ -263,6 +263,15 @@ pmem_tx_commit(struct umem_instance *umm)
 	return rc ? umem_tx_errno(rc) : 0;
 }
 
+static void
+pmem_defer_free(struct umem_instance *umm, umem_off_t off,
+		struct pobj_action *act)
+{
+	PMEMoid	id = umem_off2id(umm, off);
+
+	pmemobj_defer_free(umm->umm_pool, id, act);
+}
+
 static umem_off_t
 pmem_reserve(struct umem_instance *umm, struct pobj_action *act, size_t size,
 	     unsigned int type_num)
@@ -359,6 +368,7 @@ static umem_ops_t	pmem_ops = {
 	.mo_tx_begin		= pmem_tx_begin,
 	.mo_tx_commit		= pmem_tx_commit,
 	.mo_reserve		= pmem_reserve,
+	.mo_defer_free		= pmem_defer_free,
 	.mo_cancel		= pmem_cancel,
 	.mo_tx_publish		= pmem_tx_publish,
 	.mo_tx_add_callback	= pmem_tx_add_callback,

diff --git a/src/container/cli_internal.h b/src/container/cli_internal.h
@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016 Intel Corporation.
+ * (C) Copyright 2016-2020 Intel Corporation.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

diff --git a/src/dtx/dtx_common.c b/src/dtx/dtx_common.c
@@ -313,23 +313,13 @@ dtx_handle_init(struct dtx_id *dti, daos_handle_t coh, struct dtx_epoch *epoch,
 	dth->dth_modification_cnt = sub_modification_cnt;
 
 	dth->dth_op_seq = 0;
-	dth->dth_rsrvd_cnt = 0;
 	dth->dth_oid_cnt = 0;
 	dth->dth_oid_cap = 0;
 	dth->dth_oid_array = NULL;
 
 	dth->dth_dkey_hash = 0;
 
-	if (sub_modification_cnt <= 1) {
-		dth->dth_rsrvds = &dth->dth_rsrvd_inline;
-		return 0;
-	}
-
-	D_ALLOC_ARRAY(dth->dth_rsrvds, sub_modification_cnt);
-	if (dth->dth_rsrvds == NULL)
-		return -DER_NOMEM;
-
-	return 0;
+	return vos_dtx_rsrvd_init(dth);
 }
 
 static int
@@ -764,8 +754,7 @@ dtx_leader_end(struct dtx_leader_handle *dlh, struct ds_cont_child *cont,
 	D_FREE(dlh->dlh_subs);
 	D_FREE(dth->dth_oid_array);
 
-	if (dth->dth_rsrvds != &dth->dth_rsrvd_inline)
-		D_FREE(dth->dth_rsrvds);
+	vos_dtx_rsrvd_fini(dth);
 
 	return result;
 }

diff --git a/src/dtx/dtx_internal.h b/src/dtx/dtx_internal.h
@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2019 Intel Corporation.
+ * (C) Copyright 2019-2020 Intel Corporation.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -117,4 +117,5 @@ int dtx_abort(uuid_t po_uuid, uuid_t co_uuid, daos_epoch_t epoch,
 	      struct dtx_entry **dtes, int count);
 int dtx_check(uuid_t po_uuid, uuid_t co_uuid, struct dtx_entry *dte);
 
+
 #endif /* __DTX_INTERNAL_H__ */
diff --git a/src/include/daos/mem.h b/src/include/daos/mem.h
@@ -232,6 +232,18 @@ typedef struct {
 				       struct pobj_action *act, size_t size,
 				       unsigned int type_num);
 
+	/**
+	 * Defer free til commit.  For use with reserved extents that are not
+	 * yet published.  For VMEM, it just calls free.
+	 *
+	 * \param umm	[IN]		umem class instance.
+	 * \param off	[IN]		offset of allocation
+	 * \param act	[IN|OUT]	action used for later cancel/publish.
+	 */
+	void		 (*mo_defer_free)(struct umem_instance *umm,
+					  umem_off_t off,
+					  struct pobj_action *act);
+
 	/**
 	 * Cancel the reservation.
 	 *
@@ -494,6 +506,20 @@ umem_reserve(struct umem_instance *umm, struct pobj_action *act, size_t size)
 	return UMOFF_NULL;
 }
 
+static inline void
+umem_defer_free(struct umem_instance *umm, umem_off_t off,
+		struct pobj_action *act)
+{
+	if (umm->umm_ops->mo_defer_free)
+		return umm->umm_ops->mo_defer_free(umm, off, act);
+
+	/** Go ahead and free immediately.  The purpose of this function
+	 *  is to allow reserve/publish pair to execute on commit
+	 */
+	umem_free(umm, off);
+}
+
+
 static inline void
 umem_cancel(struct umem_instance *umm, struct pobj_action *actv, int actv_cnt)
 {

diff --git a/src/include/daos_obj_class.h b/src/include/daos_obj_class.h
@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2015-2019 Intel Corporation.
+ * (C) Copyright 2015-2020 Intel Corporation.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

diff --git a/src/include/daos_srv/dtx_srv.h b/src/include/daos_srv/dtx_srv.h
@@ -87,7 +87,8 @@ struct dtx_handle {
 	/** The flags, see dtx_entry_flags. */
 	uint32_t			 dth_flags;
 	/** The count of reserved items in the dth_rsrvds array. */
-	uint32_t			 dth_rsrvd_cnt;
+	uint16_t			 dth_rsrvd_cnt;
+	uint16_t			 dth_deferred_cnt;
 	/** The total sub modifications count. */
 	uint16_t			 dth_modification_cnt;
 	/** Modification sequence in the distributed transaction. */
@@ -105,6 +106,7 @@ struct dtx_handle {
 
 	struct dtx_rsrvd_uint		 dth_rsrvd_inline;
 	struct dtx_rsrvd_uint		*dth_rsrvds;
+	void				**dth_deferred;
 };
 
 /* Each sub transaction handle to manage each sub thandle */

diff --git a/src/include/daos_srv/vos.h b/src/include/daos_srv/vos.h
@@ -37,6 +37,23 @@
 #include <daos_srv/dtx_srv.h>
 #include <daos_srv/vos_types.h>
 
+/** Initialize the vos reserve/cancel related fields in dtx handle
+ *
+ * \param dth	[IN]	The dtx handle
+ *
+ * \return	0 on success
+ *		-DER_NOMEM on failure
+ */
+int
+vos_dtx_rsrvd_init(struct dtx_handle *dth);
+
+/** Finalize the vos reserve/cancel related fields in dtx handle
+ *
+ * \param dth	[IN]	The dtx handle
+ */
+void
+vos_dtx_rsrvd_fini(struct dtx_handle *dth);
+
 /**
  * Check the specified DTX's status, and related epoch, pool map version
  * information if required.

diff --git a/src/tests/suite/daos_test.c b/src/tests/suite/daos_test.c
@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016-2019 Intel Corporation.
+ * (C) Copyright 2016-2020 Intel Corporation.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

diff --git a/src/vos/tests/vts_common.h b/src/vos/tests/vts_common.h
@@ -145,8 +145,9 @@ vts_dtx_begin_ex(const daos_unit_oid_t *oid, daos_handle_t coh,
 	dth = *dthp;
 
 	dth->dth_modification_cnt = nmods;
-	D_ALLOC_ARRAY(dth->dth_rsrvds, nmods);
-	assert_ptr_not_equal(dth->dth_rsrvds, NULL);
+
+	/** first call in vts_dtx_begin will have set this to inline */
+	assert_int_equal(vos_dtx_rsrvd_init(dth), 0);
 }
 
 void

diff --git a/src/vos/tests/vts_dtx.c b/src/vos/tests/vts_dtx.c
@@ -86,22 +86,21 @@ vts_dtx_begin(const daos_unit_oid_t *oid, daos_handle_t coh, daos_epoch_t epoch,
 	dth->dth_modification_cnt = 1;
 
 	dth->dth_op_seq = 1;
-	dth->dth_rsrvd_cnt = 0;
 	dth->dth_oid_cnt = 0;
 	dth->dth_oid_cap = 0;
 	dth->dth_oid_array = NULL;
 
 	dth->dth_dkey_hash = dkey_hash;
-	dth->dth_rsrvds = &dth->dth_rsrvd_inline;
+
+	vos_dtx_rsrvd_init(dth);
 
 	*dthp = dth;
 }
 
 void
 vts_dtx_end(struct dtx_handle *dth)
 {
-	if (dth->dth_modification_cnt > 1)
-		D_FREE(dth->dth_rsrvds);
+	vos_dtx_rsrvd_fini(dth);
 	D_FREE(dth->dth_dte.dte_mbs);
 	D_FREE_PTR(dth);
 }