Revert "WT-6816 Design write gen scheme to allow dhandles with active…

… history to get closed/re-opened (#6085)" (#6111)

This reverts commit 9c0d479.

dist/s_string.ok

@@ -346,7 +346,6 @@ Redistributions
 Refactor
 Resize
 RocksDB
-Runtime
 SIMD
 SLIST
 SLVG

src/btree/bt_handle.c

@@ -522,45 +522,16 @@ __btree_conf(WT_SESSION_IMPL *session, WT_CKPT *ckpt)
     btree->checkpoint_gen = __wt_gen(session, WT_GEN_CHECKPOINT); /* Checkpoint generation */
 
     /*
-     * The first time we open a btree, we'll be initializing the write gen to the connection-wide
-     * base write generation since this is the largest of all btree write generations from the
-     * previous run. This has the nice property of ensuring that the range of write generations used
-     * by consecutive runs do not overlap which aids with debugging.
+     * In the regular case, we'll be initializing to the connection-wide base write generation since
+     * this is the largest of all btree write generations from the previous run. This has the nice
+     * property of ensuring that the range of write generations used by consecutive runs do not
+     * overlap which aids with debugging.
      *
-     * If we're reopening a btree or importing a new one to a running system, the btree write
-     * generation from the last run may actually be ahead of the connection-wide base write
-     * generation. In that case, we should initialize our write gen just ahead of our btree specific
-     * write generation.
-     *
-     * The runtime write generation is important since it's going to determine what we're going to
-     * use as the base write generation (and thus what pages to wipe transaction ids from). The idea
-     * is that we want to initialize it once the first time we open the btree during a run and then
-     * for every subsequent open, we want to reuse it. This so that we're still able to read
-     * transaction ids from the previous time a btree was open in the same run.
-     *
-     * FIXME-WT-6819: When we begin discarding dhandles more aggressively, we need to check that
-     * updates aren't having their transaction ids wiped after reopening the dhandle. The runtime
-     * write generation is relevant here since it should remain static across the entire run.
-     */
-    btree->write_gen = WT_MAX(ckpt->write_gen, conn->base_write_gen) + 1;
-    WT_ASSERT(session, ckpt->write_gen >= ckpt->run_write_gen);
-
-    /* If this is the first time opening the tree this run. */
-    if (F_ISSET(session, WT_SESSION_IMPORT) || ckpt->run_write_gen <= conn->base_write_gen)
-        btree->base_write_gen = btree->run_write_gen = btree->write_gen;
-    else
-        btree->base_write_gen = btree->run_write_gen = ckpt->run_write_gen;
-
-    /*
-     * We've just overwritten the runtime write generation based off the fact that know that we're
-     * importing and therefore, the checkpoint data's runtime write generation is meaningless. We
-     * need to ensure that the underlying dhandle doesn't get discarded without being included in a
-     * subsequent checkpoint including the new overwritten runtime write generation. Otherwise,
-     * we'll reopen, won't know that we're in the import case and will incorrectly use the old
-     * system's runtime write generation.
+     * In the import case, the btree write generation from the last run may actually be ahead of the
+     * connection-wide base write generation. In that case, we should initialize our write gen just
+     * ahead of our btree specific write generation.
      */
-    if (F_ISSET(session, WT_SESSION_IMPORT))
-        btree->modified = true;
+    btree->write_gen = btree->base_write_gen = WT_MAX(ckpt->write_gen + 1, conn->base_write_gen);
 
     return (0);
 }

src/include/btree.h

@@ -168,7 +168,6 @@ struct __wt_btree {
 
     uint64_t write_gen;      /* Write generation */
     uint64_t base_write_gen; /* Write generation on startup. */
-    uint64_t run_write_gen;  /* Runtime write generation. */
     uint64_t rec_max_txn;    /* Maximum txn seen (clean trees) */
     wt_timestamp_t rec_max_timestamp;
 

src/include/meta.h

@@ -137,8 +137,7 @@ struct __wt_ckpt {
 
     uint64_t size; /* Checkpoint size */
 
-    uint64_t write_gen;     /* Write generation */
-    uint64_t run_write_gen; /* Runtime write generation. */
+    uint64_t write_gen; /* Write generation */
 
     char *block_metadata;   /* Block-stored metadata */
     char *block_checkpoint; /* Block-stored checkpoint */

src/include/session.h

@@ -170,32 +170,31 @@ struct __wt_session_impl {
 #define WT_SESSION_CACHE_CURSORS 0x00000004u
 #define WT_SESSION_CAN_WAIT 0x00000008u
 #define WT_SESSION_IGNORE_CACHE_SIZE 0x00000010u
-#define WT_SESSION_IMPORT 0x00000020u
-#define WT_SESSION_IMPORT_REPAIR 0x00000040u
-#define WT_SESSION_INSTANTIATE_PREPARE 0x00000080u
-#define WT_SESSION_INTERNAL 0x00000100u
-#define WT_SESSION_LOCKED_CHECKPOINT 0x00000200u
-#define WT_SESSION_LOCKED_HANDLE_LIST_READ 0x00000400u
-#define WT_SESSION_LOCKED_HANDLE_LIST_WRITE 0x00000800u
-#define WT_SESSION_LOCKED_HOTBACKUP_READ 0x00001000u
-#define WT_SESSION_LOCKED_HOTBACKUP_WRITE 0x00002000u
-#define WT_SESSION_LOCKED_METADATA 0x00004000u
-#define WT_SESSION_LOCKED_PASS 0x00008000u
-#define WT_SESSION_LOCKED_SCHEMA 0x00010000u
-#define WT_SESSION_LOCKED_SLOT 0x00020000u
-#define WT_SESSION_LOCKED_TABLE_READ 0x00040000u
-#define WT_SESSION_LOCKED_TABLE_WRITE 0x00080000u
-#define WT_SESSION_LOCKED_TURTLE 0x00100000u
-#define WT_SESSION_LOGGING_INMEM 0x00200000u
-#define WT_SESSION_NO_DATA_HANDLES 0x00400000u
-#define WT_SESSION_NO_LOGGING 0x00800000u
-#define WT_SESSION_NO_RECONCILE 0x01000000u
-#define WT_SESSION_NO_SCHEMA_LOCK 0x02000000u
-#define WT_SESSION_QUIET_CORRUPT_FILE 0x04000000u
-#define WT_SESSION_READ_WONT_NEED 0x08000000u
-#define WT_SESSION_RESOLVING_TXN 0x10000000u
-#define WT_SESSION_ROLLBACK_TO_STABLE 0x20000000u
-#define WT_SESSION_SCHEMA_TXN 0x40000000u
+#define WT_SESSION_IMPORT_REPAIR 0x00000020u
+#define WT_SESSION_INSTANTIATE_PREPARE 0x00000040u
+#define WT_SESSION_INTERNAL 0x00000080u
+#define WT_SESSION_LOCKED_CHECKPOINT 0x00000100u
+#define WT_SESSION_LOCKED_HANDLE_LIST_READ 0x00000200u
+#define WT_SESSION_LOCKED_HANDLE_LIST_WRITE 0x00000400u
+#define WT_SESSION_LOCKED_HOTBACKUP_READ 0x00000800u
+#define WT_SESSION_LOCKED_HOTBACKUP_WRITE 0x00001000u
+#define WT_SESSION_LOCKED_METADATA 0x00002000u
+#define WT_SESSION_LOCKED_PASS 0x00004000u
+#define WT_SESSION_LOCKED_SCHEMA 0x00008000u
+#define WT_SESSION_LOCKED_SLOT 0x00010000u
+#define WT_SESSION_LOCKED_TABLE_READ 0x00020000u
+#define WT_SESSION_LOCKED_TABLE_WRITE 0x00040000u
+#define WT_SESSION_LOCKED_TURTLE 0x00080000u
+#define WT_SESSION_LOGGING_INMEM 0x00100000u
+#define WT_SESSION_NO_DATA_HANDLES 0x00200000u
+#define WT_SESSION_NO_LOGGING 0x00400000u
+#define WT_SESSION_NO_RECONCILE 0x00800000u
+#define WT_SESSION_NO_SCHEMA_LOCK 0x01000000u
+#define WT_SESSION_QUIET_CORRUPT_FILE 0x02000000u
+#define WT_SESSION_READ_WONT_NEED 0x04000000u
+#define WT_SESSION_RESOLVING_TXN 0x08000000u
+#define WT_SESSION_ROLLBACK_TO_STABLE 0x10000000u
+#define WT_SESSION_SCHEMA_TXN 0x20000000u
     /* AUTOMATIC FLAG VALUE GENERATION STOP */
     uint32_t flags;
 

src/meta/meta_ckpt.c

@@ -697,16 +697,6 @@ __ckpt_load(WT_SESSION_IMPL *session, WT_CONFIG_ITEM *k, WT_CONFIG_ITEM *v, WT_C
         goto format;
     ckpt->write_gen = (uint64_t)a.val;
 
-    /*
-     * If runtime write generation isn't supplied, this means that we're doing an upgrade and that
-     * we're opening the tree for the first time. We should just leave it as 0 so it is recognized
-     * as part of a previous run.
-     */
-    ret = __wt_config_subgets(session, v, "run_write_gen", &a);
-    WT_RET_NOTFOUND_OK(ret);
-    if (ret != WT_NOTFOUND && a.len != 0)
-        ckpt->run_write_gen = (uint64_t)a.val;
-
     return (0);
 
 format:
@@ -799,13 +789,12 @@ __wt_meta_ckptlist_to_meta(WT_SESSION_IMPL *session, WT_CKPT *ckptbase, WT_ITEM
           "=(addr=\"%.*s\",order=%" PRId64 ",time=%" PRIu64 ",size=%" PRId64
           ",newest_start_durable_ts=%" PRId64 ",oldest_start_ts=%" PRId64 ",newest_txn=%" PRId64
           ",newest_stop_durable_ts=%" PRId64 ",newest_stop_ts=%" PRId64 ",newest_stop_txn=%" PRId64
-          ",prepare=%d,write_gen=%" PRId64 ",run_write_gen=%" PRId64 ")",
+          ",prepare=%d,write_gen=%" PRId64 ")",
           (int)ckpt->addr.size, (char *)ckpt->addr.data, ckpt->order, ckpt->sec,
           (int64_t)ckpt->size, (int64_t)ckpt->ta.newest_start_durable_ts,
           (int64_t)ckpt->ta.oldest_start_ts, (int64_t)ckpt->ta.newest_txn,
           (int64_t)ckpt->ta.newest_stop_durable_ts, (int64_t)ckpt->ta.newest_stop_ts,
-          (int64_t)ckpt->ta.newest_stop_txn, (int)ckpt->ta.prepare, (int64_t)ckpt->write_gen,
-          (int64_t)ckpt->run_write_gen));
+          (int64_t)ckpt->ta.newest_stop_txn, (int)ckpt->ta.prepare, (int64_t)ckpt->write_gen));
     }
     WT_RET(__wt_buf_catfmt(session, buf, ")"));
 

src/schema/schema_create.c

@@ -798,8 +798,6 @@ __schema_create(WT_SESSION_IMPL *session, const char *uri, const char *config)
      * back it all out.
      */
     WT_RET(__wt_meta_track_on(session));
-    if (import)
-        F_SET(session, WT_SESSION_IMPORT);
 
     if (WT_PREFIX_MATCH(uri, "colgroup:"))
         ret = __create_colgroup(session, uri, exclusive, config);
@@ -818,7 +816,6 @@ __schema_create(WT_SESSION_IMPL *session, const char *uri, const char *config)
         ret = __wt_bad_object_type(session, uri);
 
     session->dhandle = NULL;
-    F_CLR(session, WT_SESSION_IMPORT);
     WT_TRET(__wt_meta_track_off(session, true, ret != 0));
 
     return (ret);

src/txn/txn_ckpt.c

@@ -1580,7 +1580,6 @@ __wt_checkpoint_tree_reconcile_update(WT_SESSION_IMPL *session, WT_TIME_AGGREGAT
     WT_CKPT_FOREACH (ckptbase, ckpt)
         if (F_ISSET(ckpt, WT_CKPT_ADD)) {
             ckpt->write_gen = btree->write_gen;
-            ckpt->run_write_gen = btree->run_write_gen;
             WT_TIME_AGGREGATE_COPY(&ckpt->ta, ta);
         }
 }