MDEV-16159 Use atomic memory access for purge_sys

Thanks to Sergey Vojtovich for feedback and many ideas.

purge_state_t: Remove. The states are replaced with
purge_sys_t::enabled() and purge_sys_t::paused() as follows:
PURGE_STATE_INIT, PURGE_STATE_EXIT, PURGE_STATE_DISABLED: !enabled().
PURGE_STATE_RUN, PURGE_STATE_STOP: paused() distinguishes these.

purge_sys_t::m_paused: Renamed from purge_sys_t::n_stop.
Protected by atomic memory access only, not purge_sys_t::latch.

purge_sys_t::m_enabled: An atomically updated Boolean that
replaces purge_sys_t::state.

purge_sys_t::running: Remove, because it duplicates
srv_sys.n_threads_active[SRV_PURGE].

purge_sys_t::running(): Accessor for srv_sys.n_threads_active[SRV_PURGE].

purge_sys_t::stop(): Renamed from trx_purge_stop().

purge_sys_t::resume(): Renamed from trx_purge_run().
Do not acquire latch; solely rely on atomics.

purge_sys_t::is_initialised(), purge_sys_t::m_initialised: Remove.

purge_sys_t::create(), purge_sys_t::close(): Instead of invoking
is_initialised(), check whether event is NULL.

purge_sys_t::event: Move before latch, so that fields that are
protected by latch can reside on the same cache line with latch.

srv_start_wait_for_purge_to_start(): Merge to the only caller srv_start().

storage/innobase/include/trx0purge.h

@@ -60,32 +60,6 @@ trx_purge(
 	ulint	n_purge_threads,	/*!< in: number of purge tasks to
 					submit to task queue. */
 	bool	truncate);		/*!< in: truncate history if true */
-/*******************************************************************//**
-Stop purge and wait for it to stop, move to PURGE_STATE_STOP. */
-void
-trx_purge_stop(void);
-/*================*/
-/*******************************************************************//**
-Resume purge, move to PURGE_STATE_RUN. */
-void
-trx_purge_run(void);
-/*================*/
-
-/** Purge states */
-enum purge_state_t {
-	PURGE_STATE_INIT,		/*!< Purge instance created */
-	PURGE_STATE_RUN,		/*!< Purge should be running */
-	PURGE_STATE_STOP,		/*!< Purge should be stopped */
-	PURGE_STATE_EXIT,		/*!< Purge has been shutdown */
-	PURGE_STATE_DISABLED		/*!< Purge was never started */
-};
-
-/*******************************************************************//**
-Get the purge state.
-@return purge state. */
-purge_state_t
-trx_purge_state(void);
-/*=================*/
 
 /** Rollback segements from a given transaction with trx-no
 scheduled for purge. */
@@ -396,29 +370,20 @@ namespace undo {
 /** The control structure used in the purge operation */
 class purge_sys_t
 {
-  bool m_initialised;
 public:
+	/** signal state changes; os_event_reset() and os_event_set()
+	are protected by rw_lock_x_lock(latch) */
 	MY_ALIGNED(CACHE_LINE_SIZE)
-	rw_lock_t	latch;		/*!< The latch protecting the purge
-					view. A purge operation must acquire an
-					x-latch here for the instant at which
-					it changes the purge view: an undo
-					log operation can prevent this by
-					obtaining an s-latch here. It also
-					protects state and running */
+	os_event_t	event;
+	/** latch protecting view, m_enabled */
 	MY_ALIGNED(CACHE_LINE_SIZE)
-	os_event_t	event;		/*!< State signal event;
-					os_event_set() and os_event_reset()
-					are protected by purge_sys_t::latch
-					X-lock */
-	MY_ALIGNED(CACHE_LINE_SIZE)
-	ulint		n_stop;		/*!< Counter to track number stops */
-
-	volatile bool	running;	/*!< true, if purge is active,
-					we check this without the latch too */
-	volatile purge_state_t	state;	/*!< Purge coordinator thread states,
-					we check this in several places
-					without holding the latch. */
+	rw_lock_t	latch;
+private:
+	/** whether purge is enabled; protected by latch and my_atomic */
+	int32_t		m_enabled;
+	/** number of pending stop() calls without resume() */
+	int32_t		m_paused;
+public:
 	que_t*		query;		/*!< The query graph which will do the
 					parallelized purge operation */
 	MY_ALIGNED(CACHE_LINE_SIZE)
@@ -494,10 +459,7 @@ class purge_sys_t
     uninitialised. Real initialisation happens in create().
   */
 
-  purge_sys_t() : m_initialised(false) {}
-
-
-  bool is_initialised() const { return m_initialised; }
+  purge_sys_t() : event(NULL), m_enabled(false) {}
 
 
   /** Create the instance */
@@ -505,6 +467,49 @@ class purge_sys_t
 
   /** Close the purge system on shutdown */
   void close();
+
+  /** @return whether purge is enabled */
+  bool enabled()
+  {
+    return my_atomic_load32_explicit(&m_enabled, MY_MEMORY_ORDER_RELAXED);
+  }
+  /** @return whether purge is enabled */
+  bool enabled_latched()
+  {
+    ut_ad(rw_lock_own_flagged(&latch, RW_LOCK_FLAG_X | RW_LOCK_FLAG_S));
+    return bool(m_enabled);
+  }
+  /** @return whether the purge coordinator is paused */
+  bool paused()
+  { return my_atomic_load32_explicit(&m_paused, MY_MEMORY_ORDER_RELAXED); }
+  /** @return whether the purge coordinator is paused */
+  bool paused_latched()
+  {
+    ut_ad(rw_lock_own_flagged(&latch, RW_LOCK_FLAG_X | RW_LOCK_FLAG_S));
+    return m_paused != 0;
+  }
+
+  /** Enable purge at startup. Not protected by latch; the main thread
+  will wait for purge_sys.enabled() in srv_start() */
+  void coordinator_startup()
+  {
+    ut_ad(!enabled());
+    my_atomic_store32_explicit(&m_enabled, true, MY_MEMORY_ORDER_RELAXED);
+  }
+
+  /** Disable purge at shutdown */
+  void coordinator_shutdown()
+  {
+    ut_ad(enabled());
+    my_atomic_store32_explicit(&m_enabled, false, MY_MEMORY_ORDER_RELAXED);
+  }
+
+  /** @return whether the purge coordinator thread is active */
+  bool running();
+  /** Stop purge during FLUSH TABLES FOR EXPORT */
+  void stop();
+  /** Resume purge at UNLOCK TABLES after FLUSH TABLES FOR EXPORT */
+  void resume();
 };
 
 /** The global data structure coordinating a purge */

storage/innobase/lock/lock0lock.cc

@@ -4620,44 +4620,15 @@ lock_print_info_summary(
 
 	fprintf(file,
 		"Purge done for trx's n:o < " TRX_ID_FMT
-		" undo n:o < " TRX_ID_FMT " state: ",
+		" undo n:o < " TRX_ID_FMT " state: %s\n"
+		"History list length " ULINTPF "\n",
 		purge_sys.tail.trx_no(),
-		purge_sys.tail.undo_no);
-
-	/* Note: We are reading the state without the latch. One because it
-	will violate the latching order and two because we are merely querying
-	the state of the variable for display. */
-
-	switch (purge_sys.state){
-	case PURGE_STATE_INIT:
-		/* Should never be in this state while the system is running. */
-		ut_error;
-
-	case PURGE_STATE_EXIT:
-		fprintf(file, "exited");
-		break;
-
-	case PURGE_STATE_DISABLED:
-		fprintf(file, "disabled");
-		break;
-
-	case PURGE_STATE_RUN:
-		fprintf(file, "running");
-		/* Check if it is waiting for more data to arrive. */
-		if (!purge_sys.running) {
-			fprintf(file, " but idle");
-		}
-		break;
-
-	case PURGE_STATE_STOP:
-		fprintf(file, "stopped");
-		break;
-	}
-
-	fprintf(file, "\n");
-
-	fprintf(file,
-		"History list length " ULINTPF "\n", trx_sys.history_size());
+		purge_sys.tail.undo_no,
+		purge_sys.enabled()
+		? (purge_sys.running() ? "running"
+		   : purge_sys.paused() ? "stopped" : "running but idle")
+		: "disabled",
+		trx_sys.history_size());
 
 #ifdef PRINT_NUM_OF_LOCK_STRUCTS
 	fprintf(file,

storage/innobase/row/row0merge.cc

@@ -1959,15 +1959,6 @@ row_merge_read_clustered_index(
 				}
 			}
 
-#ifdef DBUG_OFF
-# define dbug_run_purge	false
-#else /* DBUG_OFF */
-			bool	dbug_run_purge = false;
-#endif /* DBUG_OFF */
-			DBUG_EXECUTE_IF(
-				"ib_purge_on_create_index_page_switch",
-				dbug_run_purge = true;);
-
 			/* Insert the cached spatial index rows. */
 			bool	mtr_committed = false;
 
@@ -1984,9 +1975,8 @@ row_merge_read_clustered_index(
 				goto scan_next;
 			}
 
-			if (dbug_run_purge
-			    || my_atomic_load32_explicit(&clust_index->lock.waiters,
-							 MY_MEMORY_ORDER_RELAXED)) {
+			if (my_atomic_load32_explicit(&clust_index->lock.waiters,
+						      MY_MEMORY_ORDER_RELAXED)) {
 				/* There are waiters on the clustered
 				index tree lock, likely the purge
 				thread. Store and restore the cursor
@@ -2007,18 +1997,6 @@ row_merge_read_clustered_index(
 				btr_pcur_store_position(&pcur, &mtr);
 				mtr_commit(&mtr);
 
-				if (dbug_run_purge) {
-					/* This is for testing
-					purposes only (see
-					DBUG_EXECUTE_IF above).  We
-					signal the purge thread and
-					hope that the purge batch will
-					complete before we execute
-					btr_pcur_restore_position(). */
-					trx_purge_run();
-					os_thread_sleep(1000000);
-				}
-
 				/* Give the waiters a chance to proceed. */
 				os_thread_yield();
 scan_next:

storage/innobase/row/row0quiesce.cc

@@ -525,7 +525,7 @@ row_quiesce_table_start(
 	ib::info() << "Sync to disk of " << table->name << " started.";
 
 	if (srv_undo_sources) {
-		trx_purge_stop();
+		purge_sys.stop();
 	}
 
 	for (ulint count = 0;
@@ -609,7 +609,7 @@ row_quiesce_table_complete(
 	}
 
 	if (srv_undo_sources) {
-		trx_purge_run();
+		purge_sys.resume();
 	}
 
 	dberr_t	err = row_quiesce_set_state(table, QUIESCE_NONE, trx);

storage/innobase/srv/srv0srv.cc

@@ -613,6 +613,12 @@ struct srv_sys_t{
 
 static srv_sys_t	srv_sys;
 
+/** @return whether the purge coordinator thread is active */
+bool purge_sys_t::running()
+{
+  return my_atomic_loadlint(&srv_sys.n_threads_active[SRV_PURGE]);
+}
+
 /** Event to signal srv_monitor_thread. Not protected by a mutex.
 Set after setting srv_print_innodb_monitor. */
 os_event_t	srv_monitor_event;
@@ -1896,19 +1902,8 @@ srv_get_active_thread_type(void)
 
 	srv_sys_mutex_exit();
 
-	if (ret == SRV_NONE && srv_shutdown_state != SRV_SHUTDOWN_NONE
-	    && purge_sys.is_initialised()) {
-		/* Check only on shutdown. */
-		switch (trx_purge_state()) {
-		case PURGE_STATE_RUN:
-		case PURGE_STATE_STOP:
-			ret = SRV_PURGE;
-			break;
-		case PURGE_STATE_INIT:
-		case PURGE_STATE_DISABLED:
-		case PURGE_STATE_EXIT:
-			break;
-		}
+	if (ret == SRV_NONE && purge_sys.enabled()) {
+		ret = SRV_PURGE;
 	}
 
 	return(ret);
@@ -1947,7 +1942,7 @@ srv_wake_purge_thread_if_not_active()
 {
 	ut_ad(!srv_sys_mutex_own());
 
-	if (purge_sys.state == PURGE_STATE_RUN
+	if (purge_sys.enabled() && !purge_sys.paused()
 	    && !my_atomic_loadlint(&srv_sys.n_threads_active[SRV_PURGE])
 	    && trx_sys.history_size()) {
 
@@ -2494,19 +2489,11 @@ DECLARE_THREAD(srv_worker_thread)(
 
 			srv_wake_purge_thread_if_not_active();
 		}
-
-		/* Note: we are checking the state without holding the
-		purge_sys.latch here. */
-	} while (purge_sys.state != PURGE_STATE_EXIT);
+	} while (purge_sys.enabled());
 
 	srv_free_slot(slot);
 
-	rw_lock_x_lock(&purge_sys.latch);
-
-	ut_a(!purge_sys.running);
-	ut_a(purge_sys.state == PURGE_STATE_EXIT);
-
-	rw_lock_x_unlock(&purge_sys.latch);
+	ut_ad(!purge_sys.enabled());
 
 #ifdef UNIV_DEBUG_THREAD_CREATION
 	ib::info() << "Purge worker thread exiting, id "
@@ -2604,8 +2591,7 @@ srv_do_purge(ulint* n_total_purged)
 			/* The previous round still did some work. */
 			continue;
 		}
-	} while (n_pages_purged > 0
-		 && purge_sys.state == PURGE_STATE_RUN
+	} while (n_pages_purged > 0 && !purge_sys.paused()
 		 && !srv_purge_should_exit());
 
 	return(rseg_history_len);
@@ -2633,12 +2619,6 @@ srv_purge_coordinator_suspend(
 	int64_t		sig_count = srv_suspend_thread(slot);
 
 	do {
-		rw_lock_x_lock(&purge_sys.latch);
-
-		purge_sys.running = false;
-
-		rw_lock_x_unlock(&purge_sys.latch);
-
 		/* We don't wait right away on the the non-timed wait because
 		we want to signal the thread that wants to suspend purge. */
 		const bool wait = stop
@@ -2651,13 +2631,10 @@ srv_purge_coordinator_suspend(
 
 		rw_lock_x_lock(&purge_sys.latch);
 
-		stop = (srv_shutdown_state == SRV_SHUTDOWN_NONE
-			&& purge_sys.state == PURGE_STATE_STOP);
+		stop = srv_shutdown_state == SRV_SHUTDOWN_NONE
+			&& purge_sys.paused_latched();
 
 		if (!stop) {
-			ut_a(purge_sys.n_stop == 0);
-			purge_sys.running = true;
-
 			if (timeout
 			    && rseg_history_len < 5000
 			    && rseg_history_len == trx_sys.history_size()) {
@@ -2671,8 +2648,6 @@ srv_purge_coordinator_suspend(
 				stop = true;
 			}
 		} else {
-			ut_a(purge_sys.n_stop > 0);
-
 			/* Signal that we are suspended. */
 			os_event_set(purge_sys.event);
 		}
@@ -2700,15 +2675,9 @@ DECLARE_THREAD(srv_purge_coordinator_thread)(
 
 	ut_ad(!srv_read_only_mode);
 	ut_a(srv_n_purge_threads >= 1);
-	ut_a(trx_purge_state() == PURGE_STATE_INIT);
 	ut_a(srv_force_recovery < SRV_FORCE_NO_BACKGROUND);
 
-	rw_lock_x_lock(&purge_sys.latch);
-
-	purge_sys.running = true;
-	purge_sys.state = PURGE_STATE_RUN;
-
-	rw_lock_x_unlock(&purge_sys.latch);
+	purge_sys.coordinator_startup();
 
 #ifdef UNIV_PFS_THREAD
 	pfs_register_thread(srv_purge_thread_key);
@@ -2729,8 +2698,7 @@ DECLARE_THREAD(srv_purge_coordinator_thread)(
 
 		if (srv_shutdown_state == SRV_SHUTDOWN_NONE
 		    && srv_undo_sources
-		    && (n_total_purged == 0
-			|| purge_sys.state == PURGE_STATE_STOP)) {
+		    && (n_total_purged == 0 || purge_sys.paused())) {
 
 			srv_purge_coordinator_suspend(slot, rseg_history_len);
 		}
@@ -2754,16 +2722,13 @@ DECLARE_THREAD(srv_purge_coordinator_thread)(
 
 	/* Note that we are shutting down. */
 	rw_lock_x_lock(&purge_sys.latch);
-
-	purge_sys.state = PURGE_STATE_EXIT;
+	purge_sys.coordinator_shutdown();
 
 	/* If there are any pending undo-tablespace truncate then clear
 	it off as we plan to shutdown the purge thread. */
 	purge_sys.undo_trunc.clear();
 
-	purge_sys.running = false;
-
-	/* Ensure that the wait in trx_purge_stop() will terminate. */
+	/* Ensure that the wait in purge_sys_t::stop() will terminate. */
 	os_event_set(purge_sys.event);
 
 	rw_lock_x_unlock(&purge_sys.latch);

storage/innobase/srv/srv0start.cc

@@ -1102,41 +1102,6 @@ srv_undo_tablespaces_init(bool create_new_db)
 	return(DB_SUCCESS);
 }
 
-/********************************************************************
-Wait for the purge thread(s) to start up. */
-static
-void
-srv_start_wait_for_purge_to_start()
-/*===============================*/
-{
-	/* Wait for the purge coordinator and master thread to startup. */
-
-	purge_state_t	state = trx_purge_state();
-
-	ut_a(state != PURGE_STATE_DISABLED);
-
-	while (srv_shutdown_state == SRV_SHUTDOWN_NONE
-	       && srv_force_recovery < SRV_FORCE_NO_BACKGROUND
-	       && state == PURGE_STATE_INIT) {
-
-		switch (state = trx_purge_state()) {
-		case PURGE_STATE_RUN:
-		case PURGE_STATE_STOP:
-			break;
-
-		case PURGE_STATE_INIT:
-			ib::info() << "Waiting for purge to start";
-
-			os_thread_sleep(50000);
-			break;
-
-		case PURGE_STATE_EXIT:
-		case PURGE_STATE_DISABLED:
-			ut_error;
-		}
-	}
-}
-
 /** Create the temporary file tablespace.
 @param[in]	create_new_db	whether we are creating a new database
 @return DB_SUCCESS or error code. */
@@ -2417,7 +2382,7 @@ dberr_t srv_start(bool create_new_db)
 			| SRV_START_STATE_MONITOR;
 
 		ut_ad(srv_force_recovery >= SRV_FORCE_NO_UNDO_LOG_SCAN
-		      || trx_purge_state() == PURGE_STATE_INIT);
+		      || !purge_sys.enabled());
 
 		if (srv_force_recovery < SRV_FORCE_NO_BACKGROUND) {
 			srv_undo_sources = true;
@@ -2494,11 +2459,14 @@ dberr_t srv_start(bool create_new_db)
 			thread_started[5 + i + SRV_MAX_N_IO_THREADS] = true;
 		}
 
-		srv_start_wait_for_purge_to_start();
+		while (srv_shutdown_state == SRV_SHUTDOWN_NONE
+		       && srv_force_recovery < SRV_FORCE_NO_BACKGROUND
+		       && !purge_sys.enabled()) {
+			ib::info() << "Waiting for purge to start";
+			os_thread_sleep(50000);
+		}
 
 		srv_start_state_set(SRV_START_STATE_PURGE);
-	} else {
-		purge_sys.state = PURGE_STATE_DISABLED;
 	}
 
 	srv_is_being_started = false;

storage/innobase/trx/trx0purge.cc

@@ -161,11 +161,11 @@ purge_graph_build()
 void purge_sys_t::create()
 {
   ut_ad(this == &purge_sys);
-  ut_ad(!is_initialised());
+  ut_ad(!enabled());
+  ut_ad(!event);
   event= os_event_create(0);
-  n_stop= 0;
-  running= false;
-  state= PURGE_STATE_INIT;
+  ut_ad(event);
+  m_paused= 0;
   query= purge_graph_build();
   n_submitted= 0;
   n_completed= 0;
@@ -178,29 +178,28 @@ void purge_sys_t::create()
   rw_lock_create(trx_purge_latch_key, &latch, SYNC_PURGE_LATCH);
   mutex_create(LATCH_ID_PURGE_SYS_PQ, &pq_mutex);
   undo_trunc.create();
-  m_initialised = true;
 }
 
 /** Close the purge subsystem on shutdown. */
 void purge_sys_t::close()
 {
-	ut_ad(this == &purge_sys);
-	if (!is_initialised()) return;
-
-	m_initialised = false;
-	trx_t* trx = query->trx;
-	que_graph_free(query);
-	ut_ad(!trx->id);
-	ut_ad(trx->state == TRX_STATE_ACTIVE);
-	trx->state = TRX_STATE_NOT_STARTED;
-	trx_free(trx);
-	rw_lock_free(&latch);
-	/* rw_lock_free() already called latch.~rw_lock_t(); tame the
-	debug assertions when the destructor will be called once more. */
-	ut_ad(latch.magic_n == 0);
-	ut_d(latch.magic_n = RW_LOCK_MAGIC_N);
-	mutex_free(&pq_mutex);
-	os_event_destroy(event);
+  ut_ad(this == &purge_sys);
+  if (!event) return;
+
+  m_enabled= false;
+  trx_t* trx = query->trx;
+  que_graph_free(query);
+  ut_ad(!trx->id);
+  ut_ad(trx->state == TRX_STATE_ACTIVE);
+  trx->state= TRX_STATE_NOT_STARTED;
+  trx_free(trx);
+  rw_lock_free(&latch);
+  /* rw_lock_free() already called latch.~rw_lock_t(); tame the
+  debug assertions when the destructor will be called once more. */
+  ut_ad(latch.magic_n == 0);
+  ut_d(latch.magic_n= RW_LOCK_MAGIC_N);
+  mutex_free(&pq_mutex);
+  os_event_destroy(event);
 }
 
 /*================ UNDO LOG HISTORY LIST =============================*/
@@ -274,11 +273,10 @@ trx_purge_add_undo_to_history(const trx_t* trx, trx_undo_t*& undo, mtr_t* mtr)
 	in THD::cleanup() invoked from unlink_thd(), and we may also
 	continue to execute user transactions. */
 	ut_ad(srv_undo_sources
-	      || ((srv_startup_is_before_trx_rollback_phase
-		   || trx_rollback_is_active)
-		  && purge_sys.state == PURGE_STATE_INIT)
-	      || (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND
-		  && purge_sys.state == PURGE_STATE_DISABLED)
+	      || (!purge_sys.enabled()
+		  && (srv_startup_is_before_trx_rollback_phase
+		      || trx_rollback_is_active
+		      || srv_force_recovery >= SRV_FORCE_NO_BACKGROUND))
 	      || ((trx->undo_no == 0 || trx->mysql_thd
 		   || trx->internal)
 		  && srv_fast_shutdown));
@@ -1591,110 +1589,63 @@ trx_purge(
 	return(n_pages_handled);
 }
 
-/*******************************************************************//**
-Get the purge state.
-@return purge state. */
-purge_state_t
-trx_purge_state(void)
-/*=================*/
-{
-	purge_state_t	state;
-
-	rw_lock_x_lock(&purge_sys.latch);
-
-	state = purge_sys.state;
-
-	rw_lock_x_unlock(&purge_sys.latch);
-
-	return(state);
-}
-
-/*******************************************************************//**
-Stop purge and wait for it to stop, move to PURGE_STATE_STOP. */
-void
-trx_purge_stop(void)
-/*================*/
+/** Stop purge during FLUSH TABLES FOR EXPORT */
+void purge_sys_t::stop()
 {
-	rw_lock_x_lock(&purge_sys.latch);
-
-	switch (purge_sys.state) {
-	case PURGE_STATE_INIT:
-	case PURGE_STATE_DISABLED:
-		ut_error;
-	case PURGE_STATE_EXIT:
-		/* Shutdown must have been initiated during
-		FLUSH TABLES FOR EXPORT. */
-		ut_ad(!srv_undo_sources);
-unlock:
-		rw_lock_x_unlock(&purge_sys.latch);
-		break;
-	case PURGE_STATE_STOP:
-		ut_ad(srv_n_purge_threads > 0);
-		++purge_sys.n_stop;
-		if (!purge_sys.running) {
-			goto unlock;
-		}
-		ib::info() << "Waiting for purge to stop";
-		do {
-			rw_lock_x_unlock(&purge_sys.latch);
-			os_thread_sleep(10000);
-			rw_lock_x_lock(&purge_sys.latch);
-		} while (purge_sys.running);
-		goto unlock;
-	case PURGE_STATE_RUN:
-		ut_ad(srv_n_purge_threads > 0);
-		++purge_sys.n_stop;
-		ib::info() << "Stopping purge";
-
-		/* We need to wakeup the purge thread in case it is suspended,
-		so that it can acknowledge the state change. */
-
-		const int64_t sig_count = os_event_reset(purge_sys.event);
-		purge_sys.state = PURGE_STATE_STOP;
-		srv_purge_wakeup();
-		rw_lock_x_unlock(&purge_sys.latch);
-		/* Wait for purge coordinator to signal that it
-		is suspended. */
-		os_event_wait_low(purge_sys.event, sig_count);
-	}
-
-	MONITOR_INC_VALUE(MONITOR_PURGE_STOP_COUNT, 1);
+  rw_lock_x_lock(&latch);
+
+  if (!enabled_latched())
+  {
+    /* Shutdown must have been initiated during FLUSH TABLES FOR EXPORT. */
+    ut_ad(!srv_undo_sources);
+    rw_lock_x_unlock(&latch);
+    return;
+  }
+
+  ut_ad(srv_n_purge_threads > 0);
+
+  if (0 == my_atomic_add32_explicit(&m_paused, 1, MY_MEMORY_ORDER_RELAXED))
+  {
+    /* We need to wakeup the purge thread in case it is suspended, so
+    that it can acknowledge the state change. */
+    const int64_t sig_count = os_event_reset(event);
+    rw_lock_x_unlock(&latch);
+    ib::info() << "Stopping purge";
+    srv_purge_wakeup();
+    /* Wait for purge coordinator to signal that it is suspended. */
+    os_event_wait_low(event, sig_count);
+    MONITOR_ATOMIC_INC(MONITOR_PURGE_STOP_COUNT);
+    return;
+  }
+
+  rw_lock_x_unlock(&latch);
+
+  if (running())
+  {
+    ib::info() << "Waiting for purge to stop";
+    while (running())
+      os_thread_sleep(10000);
+  }
 }
 
-/*******************************************************************//**
-Resume purge, move to PURGE_STATE_RUN. */
-void
-trx_purge_run(void)
-/*===============*/
+/** Resume purge at UNLOCK TABLES after FLUSH TABLES FOR EXPORT */
+void purge_sys_t::resume()
 {
-	rw_lock_x_lock(&purge_sys.latch);
-
-	switch (purge_sys.state) {
-	case PURGE_STATE_EXIT:
-		/* Shutdown must have been initiated during
-		FLUSH TABLES FOR EXPORT. */
-		ut_ad(!srv_undo_sources);
-		break;
-	case PURGE_STATE_INIT:
-	case PURGE_STATE_DISABLED:
-		ut_error;
-
-	case PURGE_STATE_RUN:
-		ut_a(!purge_sys.n_stop);
-		break;
-	case PURGE_STATE_STOP:
-		ut_a(purge_sys.n_stop);
-		if (--purge_sys.n_stop == 0) {
-
-			ib::info() << "Resuming purge";
-
-			purge_sys.state = PURGE_STATE_RUN;
-		}
-
-		MONITOR_INC_VALUE(MONITOR_PURGE_RESUME_COUNT, 1);
-	}
-
-	rw_lock_x_unlock(&purge_sys.latch);
-
-	srv_purge_wakeup();
+   if (!enabled())
+   {
+     /* Shutdown must have been initiated during FLUSH TABLES FOR EXPORT. */
+     ut_ad(!srv_undo_sources);
+     return;
+   }
+
+   int32_t paused= my_atomic_add32_explicit(&m_paused, -1,
+                                            MY_MEMORY_ORDER_RELAXED);
+   ut_a(paused);
+
+   if (paused == 1)
+   {
+     ib::info() << "Resuming purge";
+     srv_purge_wakeup();
+     MONITOR_ATOMIC_INC(MONITOR_PURGE_RESUME_COUNT);
+   }
 }

storage/innobase/trx/trx0trx.cc

@@ -715,7 +715,6 @@ trx_lists_init_at_db_start()
 {
 	ut_a(srv_is_being_started);
 	ut_ad(!srv_was_started);
-	ut_ad(!purge_sys.is_initialised());
 
 	if (srv_operation == SRV_OPERATION_RESTORE) {
 		/* mariabackup --prepare only deals with