reconcile analyze

python安装.txt

@@ -1 +1,11 @@
+/usr/bin/ld: /usr/local/lib64/libsnappy.a(snappy.cc.o): relocation R_X86_64_32 against `.rodata' can not be used when making a shared object; recompile with -fPIC
+/usr/bin/ld: /usr/local/lib64/libsnappy.a(snappy-sinksource.cc.o): relocation R_X86_64_32 against symbol `_ZTVN6snappy6SourceE' can not be used when making a shared object; recompile with -fPIC
+/usr/bin/ld: /usr/local/lib64/libsnappy.a(snappy-c.cc.o): relocation R_X86_64_PC32 against symbol `snappy_max_compressed_length' can not be used when making a shared object; recompile with -fPIC
+/usr/bin/ld: final link failed: Bad value
+collect2: error: ld returned 1 exit status
+
+如果编译报错，可以跳过snappy，对应命令如下:
+cmake -DENABLE_SNAPPY=0 ../.
+
+
 ./configure --prefix=/usr/local/python-3.8.8  --enable-shared CFLAGS=-fPIC --enable-optimizations  升级python时候源码编译一定带上，否则高版本wiredtiger编译的时候会报错

wiredtiger-11.1.0/wiredtiger-11.1.0/examples/c/ex_access.c

@@ -70,17 +70,17 @@ access_example(void)
 
     /* Open a connection to the database, creating it if necessary. */ 
     //error_check(wiredtiger_open(home, NULL, "create,statistics=(all),create,verbose=[evictserver=5,evict=5,split=5,evict_stuck=5]", &conn));
-   // error_check(wiredtiger_open(home, NULL, "create,cache_size=1M, statistics=(all),create,verbose=[vessrify=5, split=5, overflow=5, generation=5, block=5, write=5, evictserver=5, evict_stuck=5, block_cache=5, checkpoint_progress=5,  checkpoint=5, checkpoint_cleanup=5, block=5,overflow=5,reconcile=5,evictserver=5,evict=5,split=5,evict_stuck=5]", &conn));
+    //error_check(wiredtiger_open(home, NULL, "create,cache_size=1M, statistics=(all),create,verbose=[split=5, overflow=5, generation=5, block=5, write=5, evictserver=5, evict_stuck=5, block_cache=5, checkpoint_progress=5,  checkpoint=5, checkpoint_cleanup=5, block=5,overflow=5,reconcile=5,evictserver=5,evict=5,split=5,evict_stuck=5]", &conn));
 
     error_check(wiredtiger_open(home, NULL, "create,cache_size=1M, statistics=(all),create,verbose=[\
     backup=5, block=5, block_cache=5, checkpoint=5, checkpoint_cleanup=5,checkpoint_progress=5,compact=5,\
     compact_progress=5,error_returns=5,evict=5,evict_stuck=5,evictserver=5,fileops=5,generation=5,handleops=5,log=5,\
     hs=5, history_store_activity=5,lsm=5,lsm_manager=5,metadata=5,mutex=5,out_of_order=5,overflow=5,read=5,reconcile=5,recovery=5, \
     recovery_progress=5,rts=5, salvage=5, shared_cache=5,split=5,temporary=5,thread_group=5,timestamp=5,tiered=5,transaction=5,verify=5,\
-    version=5,write=5, config_all_verbos=1, api=5]  ", &conn));
+    version=5,write=5, config_all_verbos=1, api=-3, metadata=-3]  ", &conn));
 
     /* Open a session handle for the database. */
-    error_check(conn->open_session(conn, NULL, NULL,     &session));
+    error_check(conn->open_session(conn, NULL, NULL, &session));
     /*! [access example connection] */
 
 
@@ -103,7 +103,7 @@ access_example(void)
     value_item.size = strlen(value_item.data);
     __wt_random_init_seed(NULL, &rnd);
 
-    for (i=400;i > 0; i--) {
+    for (i=50;i > 0; i--) {
         rval = __wt_random(&rnd);
 
         cursor->set_key(cursor, i); /* Insert a record. */

wiredtiger-11.1.0/wiredtiger-11.1.0/src/btree/bt_split.c

@@ -1160,7 +1160,8 @@ __split_internal(WT_SESSION_IMPL *session, WT_PAGE *parent, WT_PAGE *page)
 /*
  * __split_internal_lock --
  *     Lock an internal page.
- */ //获取page对应的page_lock锁，也就是锁住ref->home这个internal page
+ */ 
+//获取page对应的page_lock锁，也就是锁住ref->home这个internal page, 并返回parent page
 static int
 __split_internal_lock(WT_SESSION_IMPL *session, WT_REF *ref, bool trylock, WT_PAGE **parentp)
 {
@@ -1670,6 +1671,8 @@ __split_multi_inmem_fail(WT_SESSION_IMPL *session, WT_PAGE *orig, WT_MULTI *mult
  * __wt_multi_to_ref --
  *     Move a multi-block entry into a WT_REF structure.
  */
+//__wt_evict->__evict_page_dirty_update->__wt_split_multi->__split_multi_lock->__split_multi->__wt_multi_to_ref
+//让page指向拆分后的磁盘元数据
 int
 __wt_multi_to_ref(WT_SESSION_IMPL *session, WT_PAGE *page, WT_MULTI *multi, WT_REF **refp,
   size_t *incrp, bool closing)
@@ -2103,7 +2106,9 @@ __split_insert_lock(WT_SESSION_IMPL *session, WT_REF *ref)
 /*
  * __wt_split_insert --
  *     Split a page's last insert list entries into a separate page.
- */ //__wt_evict
+ */ 
+//__wt_evict: inmem_split，内存中的page进行拆分，拆分后的还是在内存中不会写入磁盘
+//__evict_reconcile: 对page拆分为多个page后写入磁盘中
 int
 __wt_split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 {
@@ -2123,6 +2128,7 @@ __wt_split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
  * __split_multi --
  *     Split a page into multiple pages.
  */
+//__wt_evict->__evict_page_dirty_update->__wt_split_multi->__split_multi_lock->__split_multi
 static int
 __split_multi(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 {
@@ -2147,7 +2153,7 @@ __split_multi(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
      */
     WT_RET(__wt_calloc_def(session, new_entries, &ref_new));
     for (i = 0; i < new_entries; ++i)
-        WT_ERR(
+        WT_ERR( //让page指向拆分后的磁盘元数据
           __wt_multi_to_ref(session, page, &mod->mod_multi[i], &ref_new[i], &parent_incr, closing));
 
     /*
@@ -2189,13 +2195,15 @@ __split_multi(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
  * __split_multi_lock --
  *     Split a page into multiple pages.
  */
+//__wt_evict->__evict_page_dirty_update->__wt_split_multi->__split_multi_lock
 static int
 __split_multi_lock(WT_SESSION_IMPL *session, WT_REF *ref, int closing)
 {
     WT_DECL_RET;
     WT_PAGE *parent;
 
     /* Lock the parent page, then proceed with the split. */
+    //获取page对应的page_lock锁，也就是锁住ref->home这个internal page, 并返回parent page
     WT_RET(__split_internal_lock(session, ref, false, &parent));
     if ((ret = __split_multi(session, ref, closing)) != 0 || closing) {
         __split_internal_unlock(session, parent);
@@ -2213,6 +2221,7 @@ __split_multi_lock(WT_SESSION_IMPL *session, WT_REF *ref, int closing)
  * __wt_split_multi --
  *     Split a page into multiple pages.
  */
+//__wt_evict->__evict_page_dirty_update->__wt_split_multi
 int
 __wt_split_multi(WT_SESSION_IMPL *session, WT_REF *ref, int closing)
 {

wiredtiger-11.1.0/wiredtiger-11.1.0/src/conn/conn_api.c

@@ -2219,7 +2219,8 @@ __wt_verbose_config(WT_SESSION_IMPL *session, const char *cfg[], bool reconfig)
              */
             verbose_value = WT_VERBOSE_DEBUG_1;
             goto verbos_assign;
-        } else if (sval.type == WT_CONFIG_ITEM_NUM && sval.val >= WT_VERBOSE_INFO &&
+            //yang add todo xxxxx ,这里应该是
+        } else if (sval.type == WT_CONFIG_ITEM_NUM && sval.val >= WT_VERBOSE_ERROR &&
           sval.val <= WT_VERBOSE_DEBUG_5) {
             verbose_value = (WT_VERBOSE_LEVEL)sval.val;
             goto verbos_assign;

wiredtiger-11.1.0/wiredtiger-11.1.0/src/evict/evict_page.c

@@ -173,7 +173,8 @@ __wt_evict(WT_SESSION_IMPL *session, WT_REF *ref, uint8_t previous_state, uint32
      * Review the page for conditions that would block its eviction. If the check fails (for
      * example, we find a page with active children), quit. Make this check for clean pages, too:
      * while unlikely eviction would choose an internal page with children, it's not disallowed.
-     */ //判断是否可以进行mem split，结果存入inmem_split
+     */ 
+    //判断是否可以进行mem split，结果存入inmem_split
     WT_ERR(__evict_review(session, ref, flags, &inmem_split));
 
     /*
@@ -197,6 +198,7 @@ __wt_evict(WT_SESSION_IMPL *session, WT_REF *ref, uint8_t previous_state, uint32
      * Fail 0.1% of the time after we have done reconciliation. We should always evict the page of a
      * dead tree.
      */
+    //让0.1%的evict reconcile跳过后续流程，直接err
     if (!closing && !tree_dead &&
       __wt_failpoint(session, WT_TIMING_STRESS_FAILPOINT_EVICTION_FAIL_AFTER_RECONCILIATION, 10)) {
         ret = EBUSY;
@@ -395,6 +397,7 @@ __evict_page_clean_update(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t flags)
  * __evict_page_dirty_update --
  *     Update a dirty page's reference on eviction.
  */
+//__wt_evict->__evict_page_dirty_update
 static int
 __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t evict_flags)
 {
@@ -441,6 +444,7 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t evict_
             WT_ASSERT(session, closing == false);
             WT_RET(__wt_split_rewrite(session, ref, &mod->mod_multi[0]));
         } else
+            //一般走这里，一个page通过evict reconcile拆分为多个page
             WT_RET(__wt_split_multi(session, ref, closing));
         break;
     case WT_PM_REC_REPLACE:
@@ -717,6 +721,8 @@ __evict_review(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t evict_flags, bool
  * __evict_reconcile --
  *     Reconcile the page for eviction.
  */
+//__wt_evict: inmem_split，内存中的page进行拆分，拆分后的还是在内存中不会写入磁盘
+//__evict_reconcile: 对page拆分为多个page后写入磁盘中
 static int
 __evict_reconcile(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t evict_flags)
 {

wiredtiger-11.1.0/wiredtiger-11.1.0/src/include/btmem.h

@@ -31,6 +31,7 @@
 #define WT_REC_CHECKPOINT_RUNNING 0x008u
 #define WT_REC_CLEAN_AFTER_REC 0x010u
 #define WT_REC_EVICT 0x020u
+//__evict_reconcile中如果是leaf page设置该标识
 #define WT_REC_HS 0x040u
 #define WT_REC_IN_MEMORY 0x080u
 #define WT_REC_SCRUB 0x100u
@@ -135,7 +136,7 @@ __wt_page_header_byteswap(WT_PAGE_HEADER *dsk)
  * WT_PAGE_HEADER_BYTE_SIZE --
  *	The first usable data byte on the block (past the combined headers).
  */
-//page header + block header
+//page header(WT_PAGE_HEADER_SIZE) + block header(WT_BLOCK_HEADER_SIZE)
 #define WT_PAGE_HEADER_BYTE_SIZE(btree) ((u_int)(WT_PAGE_HEADER_SIZE + (btree)->block_header))
 //dsk开始跳过page header + block header
 #define WT_PAGE_HEADER_BYTE(btree, dsk) \
@@ -273,7 +274,7 @@ struct __wt_multi {
     /*
      * Block's key: either a column-store record number or a row-store variable length byte string.
      */
-    //page拆分后对应的ref key，参考__rec_split_write
+    //page拆分后对应的ref key，参考__rec_split_dump_keys的打印，赋值的地方在__rec_split_write
     union {
         uint64_t recno;
         WT_IKEY *ikey;
@@ -403,8 +404,10 @@ struct __wt_page_modify {
             uint32_t multi_entries; /* Multiple blocks element count */
         } m;
 #undef mod_multi
+//赋值参考__rec_write_wrapup， 一次reconcile结束后，reconcile的multi信息转存到mod->mod_multi中
 #define mod_multi u1.m.multi
 #undef mod_multi_entries  
+//赋值参考__rec_write_wrapup，一次reconcile结束后，reconcile的multi信息转存到mod->mod_multi中
 #define mod_multi_entries u1.m.multi_entries
     } u1;
 
@@ -464,13 +467,17 @@ struct __wt_page_modify {
             WT_INSERT_HEAD **insert;
 
             /* Updated items for row-stores. */
+            //pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
             WT_UPDATE **update;
         } row_leaf;
 #undef mod_row_insert
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
+
 //WT_PAGE_ALLOC_AND_SWAP、__split_insert分配空间   
 //WT_ROW_INSERT_SLOT获取对应的跳跃表，实际上insert是个数组，数组每个成员对应一个跳跃表，参考__wt_leaf_page_can_split
 #define mod_row_insert u2.row_leaf.insert
 #undef mod_row_update
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
 #define mod_row_update u2.row_leaf.update
     } u2;
 
@@ -505,6 +512,7 @@ struct __wt_page_modify {
  * WT_PAGE_DIRTY --
  *	Two or more updates have been added to the page.
  */
+//__wt_page_modify_clear
 #define WT_PAGE_CLEAN 0
 #define WT_PAGE_DIRTY_FIRST 1
 #define WT_PAGE_DIRTY 2
@@ -514,6 +522,7 @@ struct __wt_page_modify {
 #define WT_PM_REC_EMPTY 1      /* Reconciliation: no replacement */
 #define WT_PM_REC_MULTIBLOCK 2 /* Reconciliation: multiple blocks */
 #define WT_PM_REC_REPLACE 3    /* Reconciliation: single block */
+    //复制参考__rec_write_wrapup
     uint8_t rec_result;        /* Reconciliation state */
 
 #define WT_PAGE_RS_RESTORED 0x1
@@ -690,6 +699,7 @@ struct __wt_page {
         //指向该page存储的真实数据，见__wt_page_alloc
         WT_ROW *row; /* Key/value pairs */
 #undef pg_row
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
 //指向该page存储的真实数据，见__wt_page_alloc
 #define pg_row u.row
 
@@ -1023,6 +1033,7 @@ struct __wt_ref {
      * or NULL if page created in-memory.
      */
     //数据对应的磁盘地址???????
+    //例如evict reconcile流程中的__wt_multi_to_ref，指向该page对应的磁盘元数据信息
     void *addr;
 
     /*
@@ -1248,6 +1259,7 @@ struct __wt_row { /* On-page key, on-page cell, or off-page WT_IKEY */
  * WT_ROW_FOREACH --
  *	Walk the entries of an in-memory row-store leaf page.
  */
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
 #define WT_ROW_FOREACH(page, rip, i) \
     for ((i) = (page)->entries, (rip) = (page)->pg_row; (i) > 0; ++(rip), --(i))
 #define WT_ROW_FOREACH_REVERSE(page, rip, i)                                             \
@@ -1557,6 +1569,7 @@ struct __wt_insert_head {
  * following macros return an array entry if the array of pointers and the specific structure exist,
  * else NULL.
  */
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
 #define WT_ROW_INSERT_SLOT(page, slot)                                  \
     ((page)->modify == NULL || (page)->modify->mod_row_insert == NULL ? \
         NULL :                                                          \
@@ -1571,6 +1584,7 @@ struct __wt_insert_head {
  * insert array. That's because the insert array requires an extra slot to hold keys that sort
  * before any key found on the original page.
  */
+//pg_row指向磁盘KV相关数据，mod_row_insert指向内存相关KV数据，mod_row_update记录内存中同一个K的变更过程
 #define WT_ROW_INSERT_SMALLEST(page)                                    \
     ((page)->modify == NULL || (page)->modify->mod_row_insert == NULL ? \
         NULL :                                                          \

wiredtiger-11.1.0/wiredtiger-11.1.0/src/include/btree.h

@@ -130,6 +130,7 @@ split_pct - The percentage of the leaf_page_max we will fill on-disk pages up to
     uint32_t maxleafpage;      /* Leaf page max size */
     uint32_t maxleafkey;       /* Leaf page max key size */
     uint32_t maxleafvalue;     /* Leaf page max value size */
+    //注意cache_size这里是/1000而不是除100
     //memory_page_max配置默认5M,取MIN(5M, (conn->cache->eviction_dirty_trigger * cache_size) / 1000) example测试也就是默认2M
     uint64_t maxmempage;       /* In-memory page max size */
     //4 * WT_MAX(btree->maxintlpage, btree->maxleafpage);
@@ -198,7 +199,8 @@ split_pct - The percentage of the leaf_page_max we will fill on-disk pages up to
 
     //__wt_btree_open->__wt_blkcache_open
     WT_BM *bm;          /* Block manager reference */  //__wt_btree.bm
-    u_int block_header; /* WT_PAGE_HEADER_BYTE_SIZE */
+    //WT_BLOCK_HEADER_SIZE
+    u_int block_header; /* WT_PAGE_HEADER_BYTE_SIZE */  //yang add todo xxxxx  备注长度错了，容易误解
 
     //__rec_set_page_write_gen中自增
     uint64_t write_gen;      /* Write generation */

wiredtiger-11.1.0/wiredtiger-11.1.0/src/include/btree_inline.h

@@ -1715,7 +1715,7 @@ __wt_leaf_page_can_split(WT_SESSION_IMPL *session, WT_PAGE *page)
     WT_INSERT *ins;
     WT_INSERT_HEAD *ins_head;
     size_t size;
-    int count;
+    int count1 = 0, count2 = 0;
 
     btree = S2BT(session);
 
@@ -1744,7 +1744,7 @@ __wt_leaf_page_can_split(WT_SESSION_IMPL *session, WT_PAGE *page)
      * correctness (the page must be reconciled again before being evicted after the split,
      * information from a previous reconciliation will be wrong, so we can't evict immediately).
      */
-    //这里是大阈值，如果page内存找过这个阈值，则不能进行memsplite,而是__evict_reconcile
+    //这里是大阈值，如果page内存超过这个阈值，则不能进行memsplite,而是__evict_reconcile
     if (page->memory_footprint < btree->splitmempage)
         return (false);
     if (WT_PAGE_IS_INTERNAL(page))
@@ -1774,14 +1774,14 @@ __wt_leaf_page_can_split(WT_SESSION_IMPL *session, WT_PAGE *page)
 //page消耗的内存较高，大于maxleafpage * 2，并且至少有5个KV
 #define WT_MAX_SPLIT_COUNT 5
     if (page->memory_footprint > (size_t)btree->maxleafpage * 2) {
-        for (count = 0, ins = ins_head->head[0]; ins != NULL; ins = ins->next[0]) {
-            if (++count < WT_MAX_SPLIT_COUNT)
+        for (count1 = 0, ins = ins_head->head[0]; ins != NULL; ins = ins->next[0]) {
+            if (++count1 < WT_MAX_SPLIT_COUNT)
                 continue;
 
             //当一个page使用内存较高的时候一般从这里返回
             WT_STAT_CONN_DATA_INCR(session, cache_inmem_splittable);
             printf("yang test ......111111........page->entries:%d..........__wt_leaf_page_can_split......count:%d..........\r\n", 
-                (int)page->entries, count);
+                (int)page->entries, count1);
             return (true);
         }
 
@@ -1797,17 +1797,20 @@ __wt_leaf_page_can_split(WT_SESSION_IMPL *session, WT_PAGE *page)
 #define WT_MIN_SPLIT_COUNT 30
 #define WT_MIN_SPLIT_MULTIPLIER 16 /* At level 2, we see 1/16th entries */
     //通过level2大概算一下有多少KV及使用的内存,这样可以避免扫描所有KV影响性能
-    for (count = 0, size = 0, ins = ins_head->head[WT_MIN_SPLIT_DEPTH]; ins != NULL;
+    for (count2 = 0, size = 0, ins = ins_head->head[WT_MIN_SPLIT_DEPTH]; ins != NULL;
          ins = ins->next[WT_MIN_SPLIT_DEPTH]) {
-        count += WT_MIN_SPLIT_MULTIPLIER;
+        count2 += WT_MIN_SPLIT_MULTIPLIER;
         size += WT_MIN_SPLIT_MULTIPLIER * (WT_INSERT_KEY_SIZE(ins) + WT_UPDATE_MEMSIZE(ins->upd));
-        if (count > WT_MIN_SPLIT_COUNT && size > (size_t)btree->maxleafpage) {
+        if (count2 > WT_MIN_SPLIT_COUNT && size > (size_t)btree->maxleafpage) {
             WT_STAT_CONN_DATA_INCR(session, cache_inmem_splittable);
-            printf("yang test ..__wt_leaf_page_can_split....sssssssssssssssss........count:%d..........size:%d................\r\n", (int)count, (int)size);
+            printf("yang test ..__wt_leaf_page_can_split....sssssssssssssssss........count2:%d..........size:%d................\r\n", (int)count2, (int)size);
             return (true);
         }
     }
-    printf("yang test ..__wt_leaf_page_can_split....11111111111111........count:%d..........size:%d................\r\n", (int)count, (int)size);
+
+    printf("yang test ..__wt_leaf_page_can_split....11111111111111...page->memory_footprint:%d \
+    ..btree->splitmempage:%d...count1:%d..........size:%d.....count2:%d...........\r\n", 
+    (int)page->memory_footprint, (int)btree->splitmempage, (int)count1, (int)size, (int)count2);
     return (false);
 }
 

wiredtiger-11.1.0/wiredtiger-11.1.0/src/include/cache.h

@@ -106,6 +106,7 @@ struct __wt_cache {
     uint64_t app_waits;  /* User threads waited for cache */
     uint64_t app_evicts; /* Pages evicted by user threads */
 
+    //也就是历史最大的page->memory_footprint
     uint64_t evict_max_page_size; /* Largest page seen at eviction */
     struct timespec stuck_time;   /* Stuck time */