diff options
Diffstat (limited to 'src/txn/txn_ckpt.c')
| -rw-r--r-- | src/txn/txn_ckpt.c | 80 |
1 files changed, 47 insertions, 33 deletions
diff --git a/src/txn/txn_ckpt.c b/src/txn/txn_ckpt.c index 3aad95f5a9f..0557e6ce60c 100644 --- a/src/txn/txn_ckpt.c +++ b/src/txn/txn_ckpt.c @@ -314,7 +314,7 @@ __checkpoint_update_generation(WT_SESSION_IMPL *session) * __checkpoint_reduce_dirty_cache -- * Release clean trees from the list cached for checkpoints. */ -static int +static void __checkpoint_reduce_dirty_cache(WT_SESSION_IMPL *session) { WT_CACHE *cache; @@ -332,9 +332,9 @@ __checkpoint_reduce_dirty_cache(WT_SESSION_IMPL *session) /* Give up if scrubbing is disabled. */ if (cache->eviction_checkpoint_target == 0 || cache->eviction_checkpoint_target >= cache->eviction_dirty_trigger) - return (0); + return; - WT_RET(__wt_epoch(session, &start)); + __wt_epoch(session, &start); last = start; bytes_written_last = 0; bytes_written_start = cache->bytes_written; @@ -345,7 +345,7 @@ __checkpoint_reduce_dirty_cache(WT_SESSION_IMPL *session) * cache via reconfigure. This avoids potential divide by zero. */ if (cache_size < 10 * WT_MEGABYTE) - return (0); + return; stepdown_us = 10000; work_us = 0; progress = false; @@ -371,7 +371,7 @@ __checkpoint_reduce_dirty_cache(WT_SESSION_IMPL *session) break; __wt_sleep(0, stepdown_us / 10); - WT_RET(__wt_epoch(session, &stop)); + __wt_epoch(session, &stop); current_us = WT_TIMEDIFF_US(stop, last); total_ms = WT_TIMEDIFF_MS(stop, start); bytes_written_total = @@ -427,14 +427,12 @@ __checkpoint_reduce_dirty_cache(WT_SESSION_IMPL *session) WT_MAX(cache->eviction_dirty_target, current_dirty - delta); WT_STAT_CONN_SET(session, txn_checkpoint_scrub_target, cache->eviction_scrub_limit); - WT_RET(__wt_epoch(session, &last)); + __wt_epoch(session, &last); } - WT_RET(__wt_epoch(session, &stop)); + __wt_epoch(session, &stop); total_ms = WT_TIMEDIFF_MS(stop, start); WT_STAT_CONN_SET(session, txn_checkpoint_scrub_time, total_ms); - - return (0); } /* @@ -497,7 +495,7 @@ __checkpoint_stats( * __checkpoint_verbose_track -- * Output a verbose message with timing information */ -static int +static void __checkpoint_verbose_track(WT_SESSION_IMPL *session, const char *msg, struct timespec *start) { @@ -506,9 +504,9 @@ __checkpoint_verbose_track(WT_SESSION_IMPL *session, uint64_t msec; if (!WT_VERBOSE_ISSET(session, WT_VERB_CHECKPOINT)) - return (0); + return; - WT_RET(__wt_epoch(session, &stop)); + __wt_epoch(session, &stop); /* * Get time diff in microseconds. @@ -526,7 +524,6 @@ __checkpoint_verbose_track(WT_SESSION_IMPL *session, WT_UNUSED(msg); WT_UNUSED(start); #endif - return (0); } /* @@ -576,7 +573,7 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) conn->cache->evict_max_page_size = 0; /* Initialize the verbose tracking timer */ - WT_ERR(__wt_epoch(session, &verb_timer)); + __wt_epoch(session, &verb_timer); /* * Update the global oldest ID so we do all possible cleanup. @@ -594,18 +591,18 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) * Try to reduce the amount of dirty data in cache so there is less * work do during the critical section of the checkpoint. */ - WT_ERR(__checkpoint_reduce_dirty_cache(session)); + __checkpoint_reduce_dirty_cache(session); /* Tell logging that we are about to start a database checkpoint. */ if (full && logging) WT_ERR(__wt_txn_checkpoint_log( session, full, WT_TXN_LOG_CKPT_PREPARE, NULL)); - WT_ERR(__checkpoint_verbose_track(session, - "starting transaction", &verb_timer)); + __checkpoint_verbose_track(session, + "starting transaction", &verb_timer); if (full) - WT_ERR(__wt_epoch(session, &start)); + __wt_epoch(session, &start); /* * Start the checkpoint for real. @@ -666,6 +663,14 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) WT_ERR(__wt_txn_id_check(session)); /* + * Mark the connection as clean. If some data gets modified after + * generating checkpoint transaction id, connection will be reset to + * dirty when reconciliation marks the btree dirty on encountering the + * dirty page. + */ + conn->modified = false; + + /* * Save the checkpoint session ID. * * We never do checkpoints in the default session (with id zero). @@ -689,7 +694,7 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) */ WT_ASSERT(session, WT_TXNID_LE(txn_global->oldest_id, txn_state->id) && - WT_TXNID_LE(txn_global->oldest_id, txn_state->snap_min)); + WT_TXNID_LE(txn_global->oldest_id, txn_state->pinned_id)); /* * Clear our entry from the global transaction session table. Any @@ -698,7 +703,7 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) * can safely ignore the checkpoint ID (see the visible all check for * details). */ - txn_state->id = txn_state->snap_min = WT_TXN_NONE; + txn_state->id = txn_state->pinned_id = WT_TXN_NONE; __wt_writeunlock(session, txn_global->scan_rwlock); /* @@ -739,23 +744,22 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) WT_ERR(__checkpoint_apply(session, cfg, __checkpoint_presync)); __wt_evict_server_wake(session); - WT_ERR(__checkpoint_verbose_track(session, - "committing transaction", &verb_timer)); + __checkpoint_verbose_track(session, + "committing transaction", &verb_timer); /* * Checkpoints have to hit disk (it would be reasonable to configure for * lazy checkpoints, but we don't support them yet). */ - WT_ERR(__wt_epoch(session, &fsync_start)); + __wt_epoch(session, &fsync_start); WT_ERR(__checkpoint_apply(session, cfg, __wt_checkpoint_sync)); - WT_ERR(__wt_epoch(session, &fsync_stop)); + __wt_epoch(session, &fsync_stop); fsync_duration_usecs = WT_TIMEDIFF_US(fsync_stop, fsync_start); WT_STAT_CONN_INCR(session, txn_checkpoint_fsync_post); WT_STAT_CONN_SET(session, txn_checkpoint_fsync_post_duration, fsync_duration_usecs); - WT_ERR(__checkpoint_verbose_track(session, - "sync completed", &verb_timer)); + __checkpoint_verbose_track(session, "sync completed", &verb_timer); /* * Commit the transaction now that we are sure that all files in the @@ -793,8 +797,8 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) ret = __wt_checkpoint_sync(session, NULL)); WT_ERR(ret); - WT_ERR(__checkpoint_verbose_track(session, - "metadata sync completed", &verb_timer)); + __checkpoint_verbose_track(session, + "metadata sync completed", &verb_timer); } else WT_WITH_DHANDLE(session, WT_SESSION_META_DHANDLE(session), @@ -808,7 +812,7 @@ __txn_checkpoint(WT_SESSION_IMPL *session, const char *cfg[]) txn_global->checkpoint_pinned = WT_TXN_NONE; if (full) { - WT_ERR(__wt_epoch(session, &stop)); + __wt_epoch(session, &stop); __checkpoint_stats(session, &start, &stop); } @@ -825,6 +829,9 @@ err: /* * overwritten the checkpoint, so what ends up on disk is not * consistent. */ + if (ret != 0 && !conn->modified) + conn->modified = true; + session->isolation = txn->isolation = WT_ISO_READ_UNCOMMITTED; if (tracking) WT_TRET(__wt_meta_track_off(session, false, ret != 0)); @@ -1352,9 +1359,13 @@ __checkpoint_tree( * out of sync with the set of dirty pages (modify is set, but there * are no dirty pages), we perform a checkpoint without any writes, no * checkpoint is created, and then things get bad. + * While marking the root page as dirty, we do not want to dirty the + * btree because we are marking the btree as clean just after this call. + * Also, marking the btree dirty at this stage will unnecessarily mark + * the connection as dirty causing checkpoint-skip code to fail. */ WT_ERR(__wt_page_modify_init(session, btree->root.page)); - __wt_page_modify_set(session, btree->root.page); + __wt_page_only_modify_set(session, btree->root.page); /* * Clear the tree's modified flag; any changes before we clear the flag @@ -1366,7 +1377,7 @@ __checkpoint_tree( * it sets the modified flag itself. Use a full barrier so we get the * store done quickly, this isn't a performance path. */ - btree->modified = 0; + btree->modified = false; WT_FULL_BARRIER(); /* Tell logging that a file checkpoint is starting. */ @@ -1440,8 +1451,11 @@ err: /* * If the checkpoint didn't complete successfully, make sure the * tree is marked dirty. */ - if (ret != 0 && !btree->modified && was_modified) - btree->modified = 1; + if (ret != 0 && !btree->modified && was_modified) { + btree->modified = true; + if (!S2C(session)->modified) + S2C(session)->modified = true; + } __wt_meta_ckptlist_free(session, ckptbase); btree->ckpt = NULL; |