Initial changes for per-table logging.

12 files changed, 86 insertions, 16 deletions

diff --git a/bench/wtperf/runners/medium-multi-btree-log-partial.wtperf b/bench/wtperf/runners/medium-multi-btree-log-partial.wtperf
new file mode 100644
index 00000000000..501cdcd9288
--- /dev/null
+++ b/bench/wtperf/runners/medium-multi-btree-log-partial.wtperf
@@ -0,0 +1,12 @@
+# wtperf options file: medium lsm configuration, with multiple tables.
+conn_config="cache_size=1G,log=(enabled,file_max=10MB)"
+table_config="type=file,os_cache_dirty_max=16MB"
+icount=5000000
+log_partial=true
+populate_threads=5
+populate_ops_per_txn=100
+#compact=true
+#threads=((count=8,read=1),(count=8,update=1))
+#run_time=180
+report_interval=5
+table_count=4
diff --git a/bench/wtperf/runners/medium-multi-btree-log.wtperf b/bench/wtperf/runners/medium-multi-btree-log.wtperf
new file mode 100644
index 00000000000..622d1ba823a
--- /dev/null
+++ b/bench/wtperf/runners/medium-multi-btree-log.wtperf
@@ -0,0 +1,11 @@
+# wtperf options file: medium lsm configuration, with multiple tables.
+conn_config="cache_size=1G,log=(enabled,file_max=10MB)"
+table_config="type=file,os_cache_dirty_max=16MB"
+icount=5000000
+populate_threads=5
+populate_ops_per_txn=100
+#compact=true
+#threads=((count=8,read=1),(count=8,update=1))
+#run_time=180
+report_interval=5
+table_count=4
diff --git a/bench/wtperf/wtperf.c b/bench/wtperf/wtperf.c
index 3a2e1709ddc..251857771c1 100644
--- a/bench/wtperf/wtperf.c
+++ b/bench/wtperf/wtperf.c
@@ -32,6 +32,7 @@
 static const CONFIG default_cfg = {
 	"WT_TEST",			/* home */
 	"WT_TEST",			/* monitor dir */
+	NULL,				/* partial logging */
 	NULL,				/* base_uri */
 	NULL,				/* uris */
 	NULL,				/* helium_mount */
@@ -1673,7 +1674,14 @@ create_tables(CONFIG *cfg)
 	}
 
 	for (i = 0; i < cfg->table_count; i++) {
-		if ((ret = session->create(
+		if (cfg->log_partial && i > 0) {
+			if (((ret = session->create(session,
+			    cfg->uris[i], cfg->partial_config)) != 0)) {
+				lprintf(cfg, ret, 0,
+				    "Error creating table %s", cfg->uris[i]);
+				return (ret);
+			}
+		} else if ((ret = session->create(
 		    session, cfg->uris[i], cfg->table_config)) != 0) {
 			lprintf(cfg, ret, 0,
 			    "Error creating table %s", cfg->uris[i]);
@@ -2155,6 +2163,16 @@ main(int argc, char *argv[])
 		if ((ret = config_opt_str(cfg, "table_config", tc_buf)) != 0)
 			goto err;
 	}
+	if (cfg->log_partial && cfg->table_count > 1) {
+		req_len = strlen(cfg->table_config) +
+		    strlen(LOG_PARTIAL_CONFIG) + 1;
+		if ((cfg->partial_config = calloc(req_len, 1)) == NULL) {
+			ret = enomem(cfg);
+			goto err;
+		}
+		snprintf(cfg->partial_config, req_len, "%s%s",
+		    cfg->table_config, LOG_PARTIAL_CONFIG);
+	}
 
 	/* Sanity-check the configuration. */
 	if ((ret = config_sanity(cfg)) != 0)
diff --git a/bench/wtperf/wtperf.h b/bench/wtperf/wtperf.h
index f176f62320e..03b19f765ea 100644
--- a/bench/wtperf/wtperf.h
+++ b/bench/wtperf/wtperf.h
@@ -98,6 +98,7 @@ typedef struct {
 	uint8_t ops[100];		/* Operation schedule */
 } WORKLOAD;
 
+#define	LOG_PARTIAL_CONFIG	",logging=false"
 /*
  * NOTE:  If you add any fields to this structure here, you must also add
  * an initialization in wtperf.c in the default_cfg.
@@ -105,6 +106,7 @@ typedef struct {
 struct __config {			/* Configuration structure */
 	const char *home;		/* WiredTiger home */
 	const char *monitor_dir;	/* Monitor output dir */
+	const char *partial_config;	/* Config string for partial logging */
 	char *base_uri;			/* Object URI */
 	char **uris;			/* URIs if multiple tables */
 	const char *helium_mount;	/* Optional Helium mount point */
diff --git a/bench/wtperf/wtperf_opt.i b/bench/wtperf/wtperf_opt.i
index cc3fd34e227..6cb39ac3cc4 100644
--- a/bench/wtperf/wtperf_opt.i
+++ b/bench/wtperf/wtperf_opt.i
@@ -115,6 +115,7 @@ DEF_OPT_AS_BOOL(index, 0,
 DEF_OPT_AS_BOOL(insert_rmw, 0,
     "execute a read prior to each insert in workload phase")
 DEF_OPT_AS_UINT32(key_sz, 20, "key size")
+DEF_OPT_AS_BOOL(log_partial, 0, "perform partial logging on first table only.")
 DEF_OPT_AS_UINT32(min_throughput, 0,
     "abort if any throughput measured is less than this amount.  Requires "
     "sample_interval to be configured")
diff --git a/dist/api_data.py b/dist/api_data.py
index fbf71581fe9..fb755045322 100644
--- a/dist/api_data.py
+++ b/dist/api_data.py
@@ -236,6 +236,11 @@ file_config = format_meta + [
     Config('leaf_item_max', '0', r'''
         historic term for leaf_key_max and leaf_value_max''',
         min=0, undoc=True),
+    Config('logging', 'true', r'''
+        the logging setting for this table.  Only valid if logging is
+        enabled on the connection.  If logging is disabled this table
+        has checkpoint-level durability.''',
+        type='boolean'),
     Config('memory_page_max', '5MB', r'''
         the maximum size a page can grow to in memory before being
         reconciled to disk.  The specified size will be adjusted to a lower
diff --git a/src/btree/bt_handle.c b/src/btree/bt_handle.c
index 7c4a4a57e3a..7c2838293d2 100644
--- a/src/btree/bt_handle.c
+++ b/src/btree/bt_handle.c
@@ -255,15 +255,25 @@ __btree_conf(WT_SESSION_IMPL *session, WT_CKPT *ckpt)
 	/* Page sizes */
 	WT_RET(__btree_page_sizes(session));
 
-	/* Eviction; the metadata file is never evicted. */
-	if (WT_IS_METADATA(btree->dhandle))
+	/* 
+	 * Set special flags for the metadata file.
+	 * Eviction; the metadata file is never evicted.
+	 * Logging; the metadata file is always logged if possible.
+	 */
+	if (WT_IS_METADATA(btree->dhandle)) {
 		F_SET(btree, WT_BTREE_IN_MEMORY | WT_BTREE_NO_EVICTION);
-	else {
+		F_CLR(btree, WT_BTREE_NO_LOGGING);
+	} else {
 		WT_RET(__wt_config_gets(session, cfg, "cache_resident", &cval));
 		if (cval.val)
 			F_SET(btree, WT_BTREE_IN_MEMORY | WT_BTREE_NO_EVICTION);
 		else
 			F_CLR(btree, WT_BTREE_IN_MEMORY | WT_BTREE_NO_EVICTION);
+		WT_RET(__wt_config_gets(session, cfg, "logging", &cval));
+		if (cval.val)
+			F_CLR(btree, WT_BTREE_NO_LOGGING);
+		else
+			F_SET(btree, WT_BTREE_NO_LOGGING);
 	}
 
 	/* Checksums */
diff --git a/src/config/config_def.c b/src/config/config_def.c
index 845b4e65825..4365b37e1fd 100644
--- a/src/config/config_def.c
+++ b/src/config/config_def.c
@@ -228,6 +228,7 @@ static const WT_CONFIG_CHECK confchk_WT_SESSION_create[] = {
 	    NULL, "min=512B,max=512MB",
 	    NULL, 0 },
 	{ "leaf_value_max", "int", NULL, "min=0", NULL, 0 },
+	{ "logging", "boolean", NULL, NULL, NULL, 0 },
 	{ "lsm", "category",
 	    NULL, NULL,
 	    confchk_WT_SESSION_create_lsm_subconfigs, 11 },
@@ -344,6 +345,7 @@ static const WT_CONFIG_CHECK confchk_file_meta[] = {
 	    NULL, "min=512B,max=512MB",
 	    NULL, 0 },
 	{ "leaf_value_max", "int", NULL, "min=0", NULL, 0 },
+	{ "logging", "boolean", NULL, NULL, NULL, 0 },
 	{ "memory_page_max", "int",
 	    NULL, "min=512B,max=10TB",
 	    NULL, 0 },
@@ -786,14 +788,14 @@ static const WT_CONFIG_ENTRY config_entries[] = {
 	  "huffman_value=,immutable=0,internal_item_max=0,"
 	  "internal_key_max=0,internal_key_truncate=,internal_page_max=4KB,"
 	  "key_format=u,key_gap=10,leaf_item_max=0,leaf_key_max=0,"
-	  "leaf_page_max=32KB,leaf_value_max=0,lsm=(auto_throttle=,bloom=,"
-	  "bloom_bit_count=16,bloom_config=,bloom_hash_count=8,"
+	  "leaf_page_max=32KB,leaf_value_max=0,logging=,lsm=(auto_throttle="
+	  ",bloom=,bloom_bit_count=16,bloom_config=,bloom_hash_count=8,"
 	  "bloom_oldest=0,chunk_count_limit=0,chunk_max=5GB,chunk_size=10MB"
 	  ",merge_max=15,merge_min=0),memory_page_max=5MB,"
 	  "os_cache_dirty_max=0,os_cache_max=0,prefix_compression=0,"
 	  "prefix_compression_min=4,source=,split_deepen_min_child=0,"
 	  "split_deepen_per_child=0,split_pct=75,type=file,value_format=u",
-	  confchk_WT_SESSION_create, 39
+	  confchk_WT_SESSION_create, 40
 	},
 	{ "WT_SESSION.drop",
 	  "force=0,remove_files=",
@@ -853,11 +855,12 @@ static const WT_CONFIG_ENTRY config_entries[] = {
 	  "huffman_value=,id=,internal_item_max=0,internal_key_max=0,"
 	  "internal_key_truncate=,internal_page_max=4KB,key_format=u,"
 	  "key_gap=10,leaf_item_max=0,leaf_key_max=0,leaf_page_max=32KB,"
-	  "leaf_value_max=0,memory_page_max=5MB,os_cache_dirty_max=0,"
-	  "os_cache_max=0,prefix_compression=0,prefix_compression_min=4,"
-	  "split_deepen_min_child=0,split_deepen_per_child=0,split_pct=75,"
-	  "value_format=u,version=(major=0,minor=0)",
-	  confchk_file_meta, 36
+	  "leaf_value_max=0,logging=,memory_page_max=5MB,"
+	  "os_cache_dirty_max=0,os_cache_max=0,prefix_compression=0,"
+	  "prefix_compression_min=4,split_deepen_min_child=0,"
+	  "split_deepen_per_child=0,split_pct=75,value_format=u,"
+	  "version=(major=0,minor=0)",
+	  confchk_file_meta, 37
 	},
 	{ "index.meta",
 	  "app_metadata=,collator=,columns=,extractor=,immutable=0,"
diff --git a/src/docs/wtperf.dox b/src/docs/wtperf.dox
index 0bca48bd18c..100857a985d 100644
--- a/src/docs/wtperf.dox
+++ b/src/docs/wtperf.dox
@@ -179,6 +179,8 @@ Whether to create an index on the value field.
 execute a read prior to each insert in workload phase
 @par key_sz (unsigned int, default=20)
 key size
+@par log_partial (boolean, default=false)
+perform partial logging on first table only.
 @par min_throughput (unsigned int, default=0)
 abort if any throughput measured is less than this amount.  Requires
 sample_interval to be configured
diff --git a/src/include/btree.h b/src/include/btree.h
index 2db14e293ed..75fefa75949 100644
--- a/src/include/btree.h
+++ b/src/include/btree.h
@@ -147,9 +147,10 @@ struct __wt_btree {
 #define	WT_BTREE_BULK		0x00100	/* Bulk-load handle */
 #define	WT_BTREE_IN_MEMORY	0x00200	/* Cache-resident object */
 #define	WT_BTREE_NO_EVICTION	0x00400	/* Disable eviction */
-#define	WT_BTREE_SALVAGE	0x00800	/* Handle is for salvage */
-#define	WT_BTREE_UPGRADE	0x01000	/* Handle is for upgrade */
-#define	WT_BTREE_VERIFY		0x02000	/* Handle is for verify */
+#define	WT_BTREE_NO_LOGGING	0x00800	/* Disable logging */
+#define	WT_BTREE_SALVAGE	0x01000	/* Handle is for salvage */
+#define	WT_BTREE_UPGRADE	0x02000	/* Handle is for upgrade */
+#define	WT_BTREE_VERIFY		0x04000	/* Handle is for verify */
 	uint32_t flags;
 };
 
diff --git a/src/include/wiredtiger.in b/src/include/wiredtiger.in
index a6f80039c10..344bbdeecaf 100644
--- a/src/include/wiredtiger.in
+++ b/src/include/wiredtiger.in
@@ -1104,6 +1104,10 @@ struct __wt_session {
 	 * temporarily ignored when large values are written.  The default is
 	 * one-half the size of a newly split leaf page., an integer greater
 	 * than or equal to 0; default \c 0.}
+	 * @config{logging, the logging setting for this table.  Only valid if
+	 * logging is enabled on the connection.  If logging is disabled this
+	 * table has checkpoint-level durability., a boolean flag; default \c
+	 * true.}
 	 * @config{lsm = (, options only relevant for LSM data sources., a set
 	 * of related configuration options defined below.}
 	 * @config{&nbsp;&nbsp;&nbsp;&nbsp;auto_throttle, Throttle inserts into
diff --git a/src/txn/txn_log.c b/src/txn/txn_log.c
index d76725d13f2..0a4843a6d51 100644
--- a/src/txn/txn_log.c
+++ b/src/txn/txn_log.c
@@ -156,7 +156,8 @@ __wt_txn_log_op(WT_SESSION_IMPL *session, WT_CURSOR_BTREE *cbt)
 	WT_TXN_OP *op;
 
 	if (!FLD_ISSET(S2C(session)->log_flags, WT_CONN_LOG_ENABLED) ||
-	    F_ISSET(session, WT_SESSION_NO_LOGGING))
+	    F_ISSET(session, WT_SESSION_NO_LOGGING) ||
+	    F_ISSET(S2BT(session), WT_BTREE_NO_LOGGING))
 		return (0);
 
 	txn = &session->txn;