(1) fix NULL trx when mtr::commit, (2) dynamic size for pref array of pe in TT

Author: trdtnguyen

storage/innobase/fil/fil0fil.cc

@@ -5847,11 +5847,11 @@ fil_io(
 #else /* UNIV_HOTBACKUP */
 	/* Queue the aio request */
 //tdnguyen test
-	if (page_id.space() == 0){
-		if (page_id.page_no() == 6){
-			pm_ppl_test_write_rseg(req_type, page_id, (byte*) buf, page_size.physical());
-		}
-	}
+	//if (page_id.space() == 0){
+	//	if (page_id.page_no() == 6){
+	//		pm_ppl_test_write_rseg(req_type, page_id, (byte*) buf, page_size.physical());
+	//	}
+	//}
 //end tdnguyen test
 	err = os_aio(
 		req_type,
@@ -5970,7 +5970,9 @@ pm_ppl_test_write_rseg(
 	}
 	//Now follow the logic in trx_undo_mem_create_at_db_start() to get other info
 	page_header = undo_page + TRX_UNDO_PAGE_HDR;
-	type = mtr_read_ulint(page_header + TRX_UNDO_PAGE_TYPE, MLOG_2BYTES, &mtr);
+
+	//type = mtr_read_ulint(page_header + TRX_UNDO_PAGE_TYPE, MLOG_2BYTES, &mtr);
+	type = mach_read_ulint(page_header + TRX_UNDO_PAGE_TYPE, MLOG_2BYTES);
 
 	seg_header = undo_page + TRX_UNDO_SEG_HDR;
 

storage/innobase/fsp/fsp0fsp.cc

@@ -53,6 +53,10 @@ Created 11/29/1995 Heikki Tuuri
 
 #include <my_aes.h>
 
+#if defined (UNIV_PMEMOBJ_PART_PL)
+#include "trx0trx.h" //for trx_t
+#endif 
+
 /** Returns an extent to the free list of a space.
 @param[in]	page_id		page id in the extent
 @param[in]	page_size	page size
@@ -1677,6 +1681,15 @@ fsp_fill_free_list(
 				mtr_start(&ibuf_mtr);
 				ibuf_mtr.set_named_space(space);
 
+#if defined (UNIV_PMEMOBJ_PART_PL)
+				//type 59 MLOG_INIT_FILE_PAGE2
+				trx_t* trx = mtr->pmemlog_get_parent_trx();
+				if (trx != NULL){
+					ibuf_mtr.pmemlog_set_parent_trx(trx);
+					ibuf_mtr.pmemlog_set_trx_id(trx->id);
+				}
+#endif
+
 				/* Avoid logging while truncate table
 				fix-up is active. */
 				if (space->purpose == FIL_TYPE_TEMPORARY

storage/innobase/include/my_pmemobj.h

@@ -424,12 +424,6 @@ struct __pmem_tx_log_block {
 
 };
 
-// Dirty Page Entry Ref in the log block
-struct __pmem_page_ref {
-	uint64_t	key; //fold of page_id_t.fold() in InnoDB
-	int64_t		idx; // index of the entry in DPT
-	uint64_t	pageLSN; //latest LSN on this page caused by the host transction, this value used in reclaiming logs
-};
 
 //Dirty Page Table Entry
 struct __pmem_dpt_entry {
@@ -487,14 +481,14 @@ struct __pmem_page_part_log {
 	// Transaction Table
 	TOID(PMEM_TT) tt; // transaction table	
 
-
 	bool		is_new;
 	/*log area as hash table*/
 	uint64_t			n_buckets; //# of buckets
 	TOID_ARRAY(TOID(PMEM_PAGE_LOG_HASHED_LINE)) buckets;
 
 	uint64_t			n_blocks_per_bucket; //# load_factor, of log block per bucket
 
+
 	/* 
 	 * DRAM objects, alloc every time the server start
 	 * */
@@ -683,6 +677,13 @@ pm_log_redoer_close(PMEM_LOG_REDOER*	redoer);
 
 /*Transaction Table Entry*/
 
+// Dirty Page Entry Ref in the log block
+struct __pmem_page_ref {
+	uint64_t	key; //fold of page_id_t.fold() in InnoDB
+	int64_t		idx; // index of the entry in DPT
+	uint64_t	pageLSN; //latest LSN on this page caused by the host transction, this value used in reclaiming logs
+};
+
 struct __pmem_tt_entry {
 	PMEMrwlock		lock;
 	uint64_t		eid; //block id
@@ -692,6 +693,7 @@ struct __pmem_tt_entry {
 	/*dirty page array consists of (key, bid) pair of dirty pages caused by this transaction*/
 	TOID_ARRAY(TOID(PMEM_PAGE_REF))		dp_arr;
 	uint64_t							n_dp_entries; 
+	uint64_t							max_dp_entries; //this value may differ to n_pref_per_entry in PMEM_TT due to the realloc mechanism
 };
 
 struct __pmem_tt_hashed_line {
@@ -704,9 +706,11 @@ struct __pmem_tt_hashed_line {
 };
 
 struct __pmem_tt {
+	TOID_ARRAY(TOID(PMEM_TT_HASHED_LINE)) buckets;
 	uint64_t	n_buckets; 
+
 	uint64_t	n_entries_per_bucket;
-	TOID_ARRAY(TOID(PMEM_TT_HASHED_LINE)) buckets;
+	uint64_t	n_pref_per_entry;
 };
 
 struct __pmem_log_group {
@@ -1210,7 +1214,17 @@ __update_page_log_block_on_commit(
 		PMEM_PAGE_REF*				pref,
 		int64_t						eid);
 void 
-__reset_TT_entry(PMEM_TT_ENTRY* pe);
+__reset_TT_entry(
+		PMEMobjpool*				pop,
+		PMEM_PAGE_PART_LOG*			ppl,
+		PMEM_TT_ENTRY*				pe);
+void 
+__realloc_TT_entry(
+		PMEMobjpool*				pop,
+		PMEM_PAGE_PART_LOG*			ppl,
+		PMEM_TT_ENTRY*				pe,
+		uint64_t					new_size);
+
 void 
 __reset_page_log_block(PMEM_PAGE_LOG_BLOCK* plog_block);
 

storage/innobase/include/row0purge.h

@@ -115,6 +115,9 @@ struct purge_node_t{
 				clustered index record */
 	ibool		done;	/* Debug flag */
 	trx_id_t	trx_id;	/*!< trx id for this purging record */
+#if defined (UNIV_PMEMOBJ_PART_PL)
+	trx_t*		trx; /*pointer to transaction*/
+#endif
 
 #ifdef UNIV_DEBUG
 	/***********************************************************//**

storage/innobase/mtr/mtr0mtr.cc

@@ -348,11 +348,7 @@ struct ReleaseBlocks {
 	void add_dirty_page_to_flush_list(mtr_memo_slot_t* slot) const
 	{
 		ut_ad(m_end_lsn > 0);
-#if defined (UNIV_PMEMOBJ_PL)
-		ut_ad(m_start_lsn >= 0);
-#else //original
 		ut_ad(m_start_lsn > 0);
-#endif
 
 		buf_block_t*	block;
 
@@ -599,6 +595,7 @@ mtr_t::start(bool sync, bool read_only)
 	m_impl.m_flush_observer = NULL;
 #if defined (UNIV_PMEMOBJ_PL)
 	m_impl.m_parent_trx = NULL;
+	m_impl.m_trx_id = 0;
 	m_impl.key_arr = (uint64_t*) calloc(512, sizeof(uint64_t));
 	m_impl.LSN_arr = (uint64_t*) calloc(512, sizeof(uint64_t));
 	m_impl.space_arr = (uint64_t*) calloc(512, sizeof(uint64_t));
@@ -1054,6 +1051,8 @@ mtr_t::Command::execute()
 
 	trx_t*			trx;
 
+	const mtr_buf_t::block_t*	front = m_impl->m_log.front();
+	byte* start_log_ptr = (byte*) front->begin(); 	
 
 	len	= m_impl->m_log.size();
 	n_recs	= m_impl->m_n_log_recs;
@@ -1067,37 +1066,24 @@ mtr_t::Command::execute()
 	}
 
 //tdnguyen test
+	mlog_id_t type;
+	byte* ptr;
+	ptr = start_log_ptr;
+	type = (mlog_id_t)((ulint)*ptr & ~MLOG_SINGLE_REC_FLAG);
 
 	if (trx == NULL){
 		if (len > 0){
-			ulint i;
-			mlog_id_t type;
-			ulint space;
-			ulint page_no;
-			ulint rec_size;
-			byte* ptr;
-			const mtr_buf_t::block_t*	front = m_impl->m_log.front();
-			byte* start_log_ptr = (byte*) front->begin(); 	
-			ptr = start_log_ptr;
-			type = (mlog_id_t)((ulint)*ptr & ~MLOG_SINGLE_REC_FLAG);
-			printf("mtr::exec trx is NULL while len > 0 n_recs %zu type %zu \n", n_recs, type);
-
-			//for (i = 0; i < n_recs; i++){
-			//	if (i == n_recs - 1){
-			//		rec_size = len - m_impl->size_arr[i];
-			//	}
-			//	else{
-			//		rec_size = m_impl->size_arr[i+1] - m_impl->size_arr[i];
-			//	}
-
-			//	//mlog_parse_initial_log_record(ptr, ptr + rec_size, &type, &space, &page_no);
-			//	printf("mtr::exec trx is NULL while len > 0 n_recs %zu rec %zu type %zu space %zu page %zu \n", n_recs, i, type, space, page_no);
+			//check what we miss from capture log
+			if (type > 8){
+				printf("mtr::exec trx is NULL while len > 0 n_recs %zu type %zu \n", n_recs, type);
+			}
+		}
+	}
+	else{
+		//test trx_id == 0
+		if (trx->id == 0 && len > 0){
+			printf("mtr::exec trx->id is ZERO while trx_id is %zu len > 0 n_recs %zu type %zu \n", m_impl->m_trx_id,  n_recs, type);
 
-			//	//if (n_recs > 1){
-			//	//	printf("==> TEST in mtr::execute() multi-rec rec %zu type %zu space %zu page %zu trx_id %zu\n", i, type, space, page_no, trx->id);
-			//	//}
-			//	ptr += rec_size;
-			//}
 		}
 	}
 	//end tdnguyen test
@@ -1186,8 +1172,8 @@ mtr_t::Command::execute()
 	if (len > 0){
 		if (trx != NULL){
 			// PART_PL WRITE LOG
-			const mtr_buf_t::block_t*	front = m_impl->m_log.front();
-			byte* start_log_ptr = (byte*) front->begin(); 	
+			//const mtr_buf_t::block_t*	front = m_impl->m_log.front();
+			//byte* start_log_ptr = (byte*) front->begin(); 	
 			////tdnguyen test
 			//ulint i;
 			//mlog_id_t type;
@@ -1207,6 +1193,8 @@ mtr_t::Command::execute()
 			//	ptr += rec_size;
 			//}
 			////end tdnguyen test
+
+			//fix node->trx->id == 0 even though node->trx_id != 0 in row_purge()
 			trx->pm_log_block_id = pm_ppl_write(
 					gb_pmw->pop,
 					gb_pmw->ppl,