Addresses #1792 refs[t:1792] Started adding show engine status

git-svn-id: file:///svn/toku/tokudb@14695 c7de825b-a66e-492c-adef-691d508d4ae1

include/db.h

@@ -42,6 +42,13 @@ typedef struct __toku_db_btree_stat64 {
   u_int64_t bt_dsize; /* how big are the keys+values (not counting the lengths) (an estimate, unless flattened) */
   u_int64_t bt_fsize; /* how big is the underlying file                                                         */
 } DB_BTREE_STAT64;
+typedef struct __toku_engine_status {
+  int              ydb_lock_held;           /* is some thread holding the ydb lock? */ 
+  u_int32_t        checkpoint_period;       /* delay between automatic checkpoints  */ 
+  u_int64_t        checkpoint_footprint;    /* state of checkpoint procedure        */ 
+  struct timeval   tbegin;                  /* time of last checkpoint begin        */ 
+  struct timeval   tend;                    /* time of last checkpoint end          */ 
+} ENGINE_STATUS;
 typedef enum {
  DB_BTREE=1,
  DB_UNKNOWN=5
@@ -145,6 +152,7 @@ struct __toku_db_env {
   int (*checkpointing_end_atomic_operation)   (DB_ENV*) /* End   a set of operations (that must be atomic as far as checkpoints are concerned). */;
   int (*set_default_bt_compare)  (DB_ENV*,int (*bt_compare) (DB *, const DBT *, const DBT *)) /* Set default (key) comparison function for all DBs in this environment.  Required for RECOVERY since you cannot open the DBs manually. */;
   int (*set_default_dup_compare) (DB_ENV*,int (*bt_compare) (DB *, const DBT *, const DBT *)) /* Set default (val) comparison function for all DBs in this environment.  Required for RECOVERY since you cannot open the DBs manually. */;
+  int (*get_engine_status)                    (DB_ENV*, ENGINE_STATUS*) /* Fill in status struct */;
   void *app_private;
   void *api1_internal;
   int  (*close) (DB_ENV *, u_int32_t);

newbrt/checkpoint.c

@@ -54,15 +54,20 @@
 #include "logger.h"
 #include "checkpoint.h"
 
-// footprint for debugging only
+// footprint for debugging and status reporting only
 static u_int64_t checkpoint_footprint = 0;
 
 static toku_pthread_rwlock_t checkpoint_safe_lock;
 static toku_pthread_rwlock_t multi_operation_lock;
 
+// Call through function pointers because this layer has no access to ydb lock functions.
 static void (*ydb_lock)(void)   = NULL;
 static void (*ydb_unlock)(void) = NULL;
 
+static BOOL initialized = FALSE;     // sanity check
+
+
+
 // Note following static functions are called from checkpoint internal logic only,
 // and use the "writer" calls for locking and unlocking.
 
@@ -149,7 +154,13 @@ toku_checkpoint_safe_client_unlock(void) {
 }
 
 
-static BOOL initialized = FALSE;
+u_int64_t 
+toku_checkpoint_get_footprint(void) {
+    return checkpoint_footprint;
+}
+
+
+
 
 // Initialize the checkpoint mechanism, must be called before any client operations.
 int 
@@ -185,22 +196,22 @@ toku_checkpoint(CACHETABLE ct, TOKULOGGER logger, char **error_string,
 		void (*callback2_f)(void*), void * extra2) {
     int r;
 
-    checkpoint_footprint = 1;
+    checkpoint_footprint = 10;
     assert(initialized);
     multi_operation_checkpoint_lock();
-    checkpoint_footprint = 2;
+    checkpoint_footprint = 20;
     checkpoint_safe_checkpoint_lock();
-    checkpoint_footprint = 3;
+    checkpoint_footprint = 30;
     ydb_lock();
     
-    checkpoint_footprint = 4;
+    checkpoint_footprint = 40;
     r = toku_cachetable_begin_checkpoint(ct, logger);
     LSN oldest_live_lsn = toku_logger_get_oldest_living_lsn(logger);
 
     multi_operation_checkpoint_unlock();
     ydb_unlock();
 
-    checkpoint_footprint = 5;
+    checkpoint_footprint = 50;
     if (r==0) {
 	if (callback_f) 
 	    callback_f(extra);      // callback is called with checkpoint_safe_lock still held
@@ -211,7 +222,7 @@ toku_checkpoint(CACHETABLE ct, TOKULOGGER logger, char **error_string,
         r = toku_logger_maybe_trim_log(logger, trim_lsn);
     }
 
-    checkpoint_footprint = 6;
+    checkpoint_footprint = 60;
     checkpoint_safe_checkpoint_unlock();
     checkpoint_footprint = 0;
 

newbrt/checkpoint.h

@@ -57,3 +57,15 @@ int toku_checkpoint_destroy(void);
 int toku_checkpoint(CACHETABLE ct, TOKULOGGER logger, char **error_string, 
 		    void (*callback_f)(void*),  void * extra,
 		    void (*callback2_f)(void*), void * extra2);
+
+
+
+/****** 
+ * These functions are called from the ydb level.
+ * They return status information and have no side effects.
+ * Some status information may be incorrect because no locks are taken to collect status.
+ * (If checkpoint is in progress, it may overwrite status info while it is being read.)
+ *****/
+
+u_int64_t toku_checkpoint_get_footprint(void);
+

src/elocks.c

@@ -18,9 +18,16 @@
 #include <sys/types.h>
 
 static toku_pthread_mutex_t ydb_big_lock = TOKU_PTHREAD_MUTEX_INITIALIZER;
+static int ydb_lock_held = 0;	// useful for debug at a live installation (int easier to pass to handlerton than BOOL)
+
+int 
+toku_ydb_lock_held(void) {
+    return ydb_lock_held;
+}
 
 int
 toku_ydb_lock_init(void) {
+    ydb_lock_held = 0;
     int r = toku_pthread_mutex_init(&ydb_big_lock, NULL); assert(r == 0);
     return r;
 }
@@ -33,9 +40,11 @@ toku_ydb_lock_destroy(void) {
 
 void toku_ydb_lock(void) {
     int r = toku_pthread_mutex_lock(&ydb_big_lock);   assert(r == 0);
+    ydb_lock_held = 1;
 }
 
 void toku_ydb_unlock(void) {
     int r = toku_pthread_mutex_unlock(&ydb_big_lock); assert(r == 0);
+    ydb_lock_held = 0;
 }
 

src/ydb-internal.h

@@ -69,6 +69,7 @@ int toku_ydb_lock_init(void);
 int toku_ydb_lock_destroy(void);
 void toku_ydb_lock(void);
 void toku_ydb_unlock(void);
+int toku_ydb_lock_held(void);
 
 /* *********************************************************
 

src/ydb.c

@@ -927,6 +927,26 @@ locked_env_set_default_bt_compare(DB_ENV * env, int (*bt_compare) (DB *, const D
     return r;
 }
 
+
+// Do not take ydb lock around or in this function.  
+// If the engine is blocked because some thread is holding the ydb lock, this function
+// can help diagnose the problem.
+// This function only collects information, and it does not matter if something gets garbled
+// because of a race condition.  
+static int
+env_get_engine_status(DB_ENV * env, ENGINE_STATUS * engstat) {
+    HANDLE_PANICKED_ENV(env);
+    int r = 0;
+    if (!env_opened(env)) r = EINVAL;
+    else {
+	engstat->ydb_lock_held = toku_ydb_lock_held();                     // is ydb lock held?
+	env_checkpointing_get_period(env, &(engstat->checkpoint_period));  // do not take ydb lock
+	engstat->checkpoint_footprint = toku_checkpoint_get_footprint();
+    }
+    return r;
+}
+
+
 static int locked_txn_begin(DB_ENV * env, DB_TXN * stxn, DB_TXN ** txn, u_int32_t flags);
 
 static int toku_db_lt_panic(DB* db, int r);
@@ -952,6 +972,7 @@ static int toku_env_create(DB_ENV ** envp, u_int32_t flags) {
     result->checkpointing_resume = env_checkpointing_resume;
     result->checkpointing_begin_atomic_operation = env_checkpointing_begin_atomic_operation;
     result->checkpointing_end_atomic_operation = env_checkpointing_end_atomic_operation;
+    result->get_engine_status = env_get_engine_status;
     result->open = locked_env_open;
     result->close = locked_env_close;
     result->txn_checkpoint = toku_env_txn_checkpoint;