Add recycle stats and code to set a maximum number of recycle files.

8 files changed, 130 insertions, 57 deletions

diff --git a/dist/stat_data.py b/dist/stat_data.py
index df6c919f808..43285076630 100644
--- a/dist/stat_data.py
+++ b/dist/stat_data.py
@@ -169,8 +169,14 @@ connection_stats = [
     LogStat('log_bytes_payload', 'log bytes of payload data'),
     LogStat('log_bytes_written', 'log bytes written'),
     LogStat('log_close_yields', 'yields waiting for previous log file close'),
-    LogStat('log_max_filesize', 'maximum log file size', 'no_clear'),
+    LogStat('log_max_filesize', 'maximum log file size', 'no_clear,no_scale'),
     LogStat('log_reads', 'log read operations'),
+    LogStat('log_recycle_files', 'archived log files prepared for recycling'),
+    LogStat('log_recycle_max',
+        'maximum recycled log files to keep', 'no_clear,no_scale'),
+    LogStat('log_recycle_removed',
+        'archived log files removed over recycle limit'),
+    LogStat('log_recycle_reused', 'recycled log files reused'),
     LogStat('log_scan_records', 'records processed by log scan'),
     LogStat('log_scan_rereads', 'log scan records requiring two reads'),
     LogStat('log_scans', 'log scan operations'),
diff --git a/src/conn/conn_log.c b/src/conn/conn_log.c
index 06ab802bbfc..341ce4bd101 100644
--- a/src/conn/conn_log.c
+++ b/src/conn/conn_log.c
@@ -85,13 +85,13 @@ __log_archive_once(WT_SESSION_IMPL *session, uint32_t backup_file)
 	WT_DECL_RET;
 	WT_LOG *log;
 	uint32_t lognum, min_lognum;
-	u_int i, locked, logcount;
-	char **logfiles;
+	u_int i, locked, logcount, num_archived, reccount, recycled;
+	char **logfiles, **recfiles;
 
 	conn = S2C(session);
 	log = conn->log;
-	logcount = 0;
-	logfiles = NULL;
+	logcount = num_archived = reccount = 0;
+	logfiles = recfiles = NULL;
 
 	/*
 	 * If we're coming from a backup cursor we want the smaller of
@@ -118,10 +118,12 @@ __log_archive_once(WT_SESSION_IMPL *session, uint32_t backup_file)
 	__wt_spin_lock(session, &conn->hot_backup_lock);
 	locked = 1;
 	if (conn->hot_backup == 0 || backup_file != 0) {
+		num_archived = 0;
 		for (i = 0; i < logcount; i++) {
 			WT_ERR(__wt_log_extract_lognum(
 			    session, logfiles[i], &lognum));
 			if (lognum < min_lognum) {
+				num_archived++;
 				if (conn->log_recycle)
 					WT_ERR(__wt_log_recycle_rename(
 					    session, lognum));
@@ -139,14 +141,48 @@ __log_archive_once(WT_SESSION_IMPL *session, uint32_t backup_file)
 	 * them to be reused.  We've moved them aside while holding the backup
 	 * lock and now we can process them without the lock.
 	 */
-	if (conn->log_recycle)
+	if (conn->log_recycle) {
+		if (conn->log_recycle_max == 0) {
+			/*
+			 * If checkpoints based on log size are set, use that
+			 * as the number of log files to keep.  Otherwise
+			 * compute it based on how many we just archived.
+			 */
+			if (WT_CKPT_LOGSIZE(conn))
+				conn->log_recycle_max =
+				    (uint32_t)(conn->ckpt_logsize/
+				    conn->log_file_max);
+			else
+				conn->log_recycle_max = num_archived;
+		} else
+			conn->log_recycle_max = WT_MAX(conn->log_recycle_max,
+			    num_archived);
+		WT_STAT_FAST_CONN_SET(session,
+		    log_recycle_max, conn->log_recycle_max);
+		WT_ERR(__wt_dirlist(session, conn->log_path,
+		    WT_LOG_RECYCLENAME, WT_DIRLIST_INCLUDE,
+		    &recfiles, &reccount));
+		recycled = reccount;
+		__wt_log_files_free(session, recfiles, reccount);
+		recfiles = NULL;
+		reccount = 0;
 		for (i = 0; i < logcount; i++) {
 			WT_ERR(__wt_log_extract_lognum(
 			    session, logfiles[i], &lognum));
-			if (lognum < min_lognum)
-				WT_ERR(__wt_log_recycle(
-				    session, lognum));
+			if (lognum < min_lognum) {
+				if (recycled < conn->log_recycle_max) {
+					recycled++;
+					WT_ERR(__wt_log_recycle(
+					    session, lognum));
+				} else {
+					WT_STAT_FAST_CONN_INCR(session,
+					    log_recycle_removed);
+					WT_ERR(__wt_log_remove(
+					    session, WT_LOG_ARCHNAME, lognum));
+				}
+			}
 		}
+	}
 
 	__wt_log_files_free(session, logfiles, logcount);
 	logfiles = NULL;
@@ -166,6 +202,8 @@ err:		__wt_err(session, ret, "log archive server error");
 		__wt_spin_unlock(session, &conn->hot_backup_lock);
 	if (logfiles != NULL)
 		__wt_log_files_free(session, logfiles, logcount);
+	if (recfiles != NULL)
+		__wt_log_files_free(session, recfiles, reccount);
 	return (ret);
 }
 
diff --git a/src/include/connection.h b/src/include/connection.h
index 14ce115e399..d9300b319fa 100644
--- a/src/include/connection.h
+++ b/src/include/connection.h
@@ -257,7 +257,9 @@ struct __wt_connection_impl {
 	wt_off_t	 log_file_max;	/* Log file max size */
 	const char	*log_path;	/* Logging path format */
 	int		 log_recycle;	/* Log file recycle configuration */
-	uint32_t	txn_logsync;	/* Log sync configuration */
+					/* Maximum recycled log files */
+	uint32_t	 log_recycle_max;
+	uint32_t	 txn_logsync;	/* Log sync configuration */
 
 	WT_SESSION_IMPL *sweep_session;	/* Handle sweep session */
 	wt_thread_t	 sweep_tid;	/* Handle sweep thread */
diff --git a/src/include/stat.h b/src/include/stat.h
index ee2baa9037b..9c1bb493d4b 100644
--- a/src/include/stat.h
+++ b/src/include/stat.h
@@ -196,6 +196,10 @@ struct __wt_connection_stats {
 	WT_STATS log_close_yields;
 	WT_STATS log_max_filesize;
 	WT_STATS log_reads;
+	WT_STATS log_recycle_files;
+	WT_STATS log_recycle_max;
+	WT_STATS log_recycle_removed;
+	WT_STATS log_recycle_reused;
 	WT_STATS log_scan_records;
 	WT_STATS log_scan_rereads;
 	WT_STATS log_scans;
diff --git a/src/include/wiredtiger.in b/src/include/wiredtiger.in
index 08a0837fba5..41c08f9b58b 100644
--- a/src/include/wiredtiger.in
+++ b/src/include/wiredtiger.in
@@ -3186,100 +3186,108 @@ extern int wiredtiger_extension_terminate(WT_CONNECTION *connection);
 #define	WT_STAT_CONN_LOG_MAX_FILESIZE			1063
 /*! log: log read operations */
 #define	WT_STAT_CONN_LOG_READS				1064
+/*! log: archived log files prepared for recycling */
+#define	WT_STAT_CONN_LOG_RECYCLE_FILES			1065
+/*! log: maximum recycled log files to keep */
+#define	WT_STAT_CONN_LOG_RECYCLE_MAX			1066
+/*! log: archived log files removed over recycle limit */
+#define	WT_STAT_CONN_LOG_RECYCLE_REMOVED		1067
+/*! log: recycled log files reused */
+#define	WT_STAT_CONN_LOG_RECYCLE_REUSED			1068
 /*! log: records processed by log scan */
-#define	WT_STAT_CONN_LOG_SCAN_RECORDS			1065
+#define	WT_STAT_CONN_LOG_SCAN_RECORDS			1069
 /*! log: log scan records requiring two reads */
-#define	WT_STAT_CONN_LOG_SCAN_REREADS			1066
+#define	WT_STAT_CONN_LOG_SCAN_REREADS			1070
 /*! log: log scan operations */
-#define	WT_STAT_CONN_LOG_SCANS				1067
+#define	WT_STAT_CONN_LOG_SCANS				1071
 /*! log: consolidated slot closures */
-#define	WT_STAT_CONN_LOG_SLOT_CLOSES			1068
+#define	WT_STAT_CONN_LOG_SLOT_CLOSES			1072
 /*! log: logging bytes consolidated */
-#define	WT_STAT_CONN_LOG_SLOT_CONSOLIDATED		1069
+#define	WT_STAT_CONN_LOG_SLOT_CONSOLIDATED		1073
 /*! log: consolidated slot joins */
-#define	WT_STAT_CONN_LOG_SLOT_JOINS			1070
+#define	WT_STAT_CONN_LOG_SLOT_JOINS			1074
 /*! log: consolidated slot join races */
-#define	WT_STAT_CONN_LOG_SLOT_RACES			1071
+#define	WT_STAT_CONN_LOG_SLOT_RACES			1075
 /*! log: slots selected for switching that were unavailable */
-#define	WT_STAT_CONN_LOG_SLOT_SWITCH_FAILS		1072
+#define	WT_STAT_CONN_LOG_SLOT_SWITCH_FAILS		1076
 /*! log: record size exceeded maximum */
-#define	WT_STAT_CONN_LOG_SLOT_TOOBIG			1073
+#define	WT_STAT_CONN_LOG_SLOT_TOOBIG			1077
 /*! log: failed to find a slot large enough for record */
-#define	WT_STAT_CONN_LOG_SLOT_TOOSMALL			1074
+#define	WT_STAT_CONN_LOG_SLOT_TOOSMALL			1078
 /*! log: consolidated slot join transitions */
-#define	WT_STAT_CONN_LOG_SLOT_TRANSITIONS		1075
+#define	WT_STAT_CONN_LOG_SLOT_TRANSITIONS		1079
 /*! log: log sync operations */
-#define	WT_STAT_CONN_LOG_SYNC				1076
+#define	WT_STAT_CONN_LOG_SYNC				1080
 /*! log: log write operations */
-#define	WT_STAT_CONN_LOG_WRITES				1077
+#define	WT_STAT_CONN_LOG_WRITES				1081
 /*! LSM: sleep for LSM checkpoint throttle */
-#define	WT_STAT_CONN_LSM_CHECKPOINT_THROTTLE		1078
+#define	WT_STAT_CONN_LSM_CHECKPOINT_THROTTLE		1082
 /*! LSM: sleep for LSM merge throttle */
-#define	WT_STAT_CONN_LSM_MERGE_THROTTLE			1079
+#define	WT_STAT_CONN_LSM_MERGE_THROTTLE			1083
 /*! LSM: rows merged in an LSM tree */
-#define	WT_STAT_CONN_LSM_ROWS_MERGED			1080
+#define	WT_STAT_CONN_LSM_ROWS_MERGED			1084
 /*! LSM: application work units currently queued */
-#define	WT_STAT_CONN_LSM_WORK_QUEUE_APP			1081
+#define	WT_STAT_CONN_LSM_WORK_QUEUE_APP			1085
 /*! LSM: merge work units currently queued */
-#define	WT_STAT_CONN_LSM_WORK_QUEUE_MANAGER		1082
+#define	WT_STAT_CONN_LSM_WORK_QUEUE_MANAGER		1086
 /*! LSM: tree queue hit maximum */
-#define	WT_STAT_CONN_LSM_WORK_QUEUE_MAX			1083
+#define	WT_STAT_CONN_LSM_WORK_QUEUE_MAX			1087
 /*! LSM: switch work units currently queued */
-#define	WT_STAT_CONN_LSM_WORK_QUEUE_SWITCH		1084
+#define	WT_STAT_CONN_LSM_WORK_QUEUE_SWITCH		1088
 /*! LSM: tree maintenance operations scheduled */
-#define	WT_STAT_CONN_LSM_WORK_UNITS_CREATED		1085
+#define	WT_STAT_CONN_LSM_WORK_UNITS_CREATED		1089
 /*! LSM: tree maintenance operations discarded */
-#define	WT_STAT_CONN_LSM_WORK_UNITS_DISCARDED		1086
+#define	WT_STAT_CONN_LSM_WORK_UNITS_DISCARDED		1090
 /*! LSM: tree maintenance operations executed */
-#define	WT_STAT_CONN_LSM_WORK_UNITS_DONE		1087
+#define	WT_STAT_CONN_LSM_WORK_UNITS_DONE		1091
 /*! connection: memory allocations */
-#define	WT_STAT_CONN_MEMORY_ALLOCATION			1088
+#define	WT_STAT_CONN_MEMORY_ALLOCATION			1092
 /*! connection: memory frees */
-#define	WT_STAT_CONN_MEMORY_FREE			1089
+#define	WT_STAT_CONN_MEMORY_FREE			1093
 /*! connection: memory re-allocations */
-#define	WT_STAT_CONN_MEMORY_GROW			1090
+#define	WT_STAT_CONN_MEMORY_GROW			1094
 /*! connection: total read I/Os */
-#define	WT_STAT_CONN_READ_IO				1091
+#define	WT_STAT_CONN_READ_IO				1095
 /*! reconciliation: page reconciliation calls */
-#define	WT_STAT_CONN_REC_PAGES				1092
+#define	WT_STAT_CONN_REC_PAGES				1096
 /*! reconciliation: page reconciliation calls for eviction */
-#define	WT_STAT_CONN_REC_PAGES_EVICTION			1093
+#define	WT_STAT_CONN_REC_PAGES_EVICTION			1097
 /*! reconciliation: split bytes currently awaiting free */
-#define	WT_STAT_CONN_REC_SPLIT_STASHED_BYTES		1094
+#define	WT_STAT_CONN_REC_SPLIT_STASHED_BYTES		1098
 /*! reconciliation: split objects currently awaiting free */
-#define	WT_STAT_CONN_REC_SPLIT_STASHED_OBJECTS		1095
+#define	WT_STAT_CONN_REC_SPLIT_STASHED_OBJECTS		1099
 /*! connection: pthread mutex shared lock read-lock calls */
-#define	WT_STAT_CONN_RWLOCK_READ			1096
+#define	WT_STAT_CONN_RWLOCK_READ			1100
 /*! connection: pthread mutex shared lock write-lock calls */
-#define	WT_STAT_CONN_RWLOCK_WRITE			1097
+#define	WT_STAT_CONN_RWLOCK_WRITE			1101
 /*! session: open cursor count */
-#define	WT_STAT_CONN_SESSION_CURSOR_OPEN		1098
+#define	WT_STAT_CONN_SESSION_CURSOR_OPEN		1102
 /*! session: open session count */
-#define	WT_STAT_CONN_SESSION_OPEN			1099
+#define	WT_STAT_CONN_SESSION_OPEN			1103
 /*! transaction: transaction begins */
-#define	WT_STAT_CONN_TXN_BEGIN				1100
+#define	WT_STAT_CONN_TXN_BEGIN				1104
 /*! transaction: transaction checkpoints */
-#define	WT_STAT_CONN_TXN_CHECKPOINT			1101
+#define	WT_STAT_CONN_TXN_CHECKPOINT			1105
 /*! transaction: transaction checkpoint currently running */
-#define	WT_STAT_CONN_TXN_CHECKPOINT_RUNNING		1102
+#define	WT_STAT_CONN_TXN_CHECKPOINT_RUNNING		1106
 /*! transaction: transaction checkpoint max time (msecs) */
-#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_MAX		1103
+#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_MAX		1107
 /*! transaction: transaction checkpoint min time (msecs) */
-#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_MIN		1104
+#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_MIN		1108
 /*! transaction: transaction checkpoint most recent time (msecs) */
-#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_RECENT		1105
+#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_RECENT		1109
 /*! transaction: transaction checkpoint total time (msecs) */
-#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_TOTAL		1106
+#define	WT_STAT_CONN_TXN_CHECKPOINT_TIME_TOTAL		1110
 /*! transaction: transactions committed */
-#define	WT_STAT_CONN_TXN_COMMIT				1107
+#define	WT_STAT_CONN_TXN_COMMIT				1111
 /*! transaction: transaction failures due to cache overflow */
-#define	WT_STAT_CONN_TXN_FAIL_CACHE			1108
+#define	WT_STAT_CONN_TXN_FAIL_CACHE			1112
 /*! transaction: transaction range of IDs currently pinned */
-#define	WT_STAT_CONN_TXN_PINNED_RANGE			1109
+#define	WT_STAT_CONN_TXN_PINNED_RANGE			1113
 /*! transaction: transactions rolled back */
-#define	WT_STAT_CONN_TXN_ROLLBACK			1110
+#define	WT_STAT_CONN_TXN_ROLLBACK			1114
 /*! connection: total write I/Os */
-#define	WT_STAT_CONN_WRITE_IO				1111
+#define	WT_STAT_CONN_WRITE_IO				1115
 
 /*!
  * @}
diff --git a/src/log/log.c b/src/log/log.c
index 60c1c34bc9e..1982b7a54e7 100644
--- a/src/log/log.c
+++ b/src/log/log.c
@@ -426,6 +426,7 @@ __log_alloc_recycle(WT_SESSION_IMPL *session, uint32_t to_num)
 	WT_ERR(__wt_verbose(session, WT_VERB_LOG,
 	    "log_alloc_recycle: rename log %s to %s",
 	    (char *)from_path->data, (char *)to_path->data));
+	WT_STAT_FAST_CONN_INCR(session, log_recycle_reused);
 	WT_ERR(__wt_rename(session, from_path->data, to_path->data));
 
 err:	__wt_scr_free(&from_path);
@@ -588,6 +589,7 @@ __wt_log_recycle(WT_SESSION_IMPL *session, uint32_t lognum)
 	WT_ERR(__wt_verbose(session, WT_VERB_LOG,
 	    "log_recycle: recycle log %s to %s",
 	    (char *)from_path->data, (char *)to_path->data));
+	WT_STAT_FAST_CONN_INCR(session, log_recycle_files);
 	WT_ERR(__wt_rename(session, from_path->data, to_path->data));
 err:	__wt_scr_free(&from_path);
 	__wt_scr_free(&to_path);
diff --git a/src/support/stat.c b/src/support/stat.c
index 2b53746bb46..d3f9c1f06f8 100644
--- a/src/support/stat.c
+++ b/src/support/stat.c
@@ -404,6 +404,10 @@ __wt_stat_init_connection_stats(WT_CONNECTION_STATS *stats)
 	stats->write_io.desc = "connection: total write I/Os";
 	stats->dh_session_handles.desc = "data-handle: session dhandles swept";
 	stats->dh_session_sweeps.desc = "data-handle: session sweep attempts";
+	stats->log_recycle_files.desc =
+	    "log: archived log files prepared for recycling";
+	stats->log_recycle_removed.desc =
+	    "log: archived log files removed over recycle limit";
 	stats->log_slot_closes.desc = "log: consolidated slot closures";
 	stats->log_slot_races.desc = "log: consolidated slot join races";
 	stats->log_slot_transitions.desc =
@@ -422,8 +426,11 @@ __wt_stat_init_connection_stats(WT_CONNECTION_STATS *stats)
 	stats->log_writes.desc = "log: log write operations";
 	stats->log_slot_consolidated.desc = "log: logging bytes consolidated";
 	stats->log_max_filesize.desc = "log: maximum log file size";
+	stats->log_recycle_max.desc =
+	    "log: maximum recycled log files to keep";
 	stats->log_slot_toobig.desc = "log: record size exceeded maximum";
 	stats->log_scan_records.desc = "log: records processed by log scan";
+	stats->log_recycle_reused.desc = "log: recycled log files reused";
 	stats->log_slot_switch_fails.desc =
 	    "log: slots selected for switching that were unavailable";
 	stats->log_buffer_size.desc = "log: total log buffer size";
@@ -542,6 +549,8 @@ __wt_stat_refresh_connection_stats(void *stats_arg)
 	stats->write_io.v = 0;
 	stats->dh_session_handles.v = 0;
 	stats->dh_session_sweeps.v = 0;
+	stats->log_recycle_files.v = 0;
+	stats->log_recycle_removed.v = 0;
 	stats->log_slot_closes.v = 0;
 	stats->log_slot_races.v = 0;
 	stats->log_slot_transitions.v = 0;
@@ -558,6 +567,7 @@ __wt_stat_refresh_connection_stats(void *stats_arg)
 	stats->log_slot_consolidated.v = 0;
 	stats->log_slot_toobig.v = 0;
 	stats->log_scan_records.v = 0;
+	stats->log_recycle_reused.v = 0;
 	stats->log_slot_switch_fails.v = 0;
 	stats->log_close_yields.v = 0;
 	stats->lsm_rows_merged.v = 0;
diff --git a/tools/stat_data.py b/tools/stat_data.py
index 56218f497b7..93e0c03a9cb 100644
--- a/tools/stat_data.py
+++ b/tools/stat_data.py
@@ -8,6 +8,8 @@ no_scale_per_second_list = [
     'cache: tracked dirty bytes in the cache',
     'cache: tracked dirty pages in the cache',
     'connection: files currently open',
+    'log: maximum log file size',
+    'log: maximum recycled log files to keep',
     'log: total log buffer size',
     'LSM: application work units currently queued',
     'LSM: merge work units currently queued',
@@ -56,6 +58,7 @@ no_clear_list = [
     'cache: pages currently held in the cache',
     'connection: files currently open',
     'log: maximum log file size',
+    'log: maximum recycled log files to keep',
     'log: total log buffer size',
     'LSM: application work units currently queued',
     'LSM: merge work units currently queued',