- Fix to log a verbose message at operational notice level if a

thread is not responding, to stats requests. It is logged with
  thread identifiers.

config.h.in

@@ -298,6 +298,9 @@
 /* Define to 1 if you have the `getrlimit' function. */
 #undef HAVE_GETRLIMIT
 
+/* Define to 1 if you have the `gettid' function. */
+#undef HAVE_GETTID
+
 /* Define to 1 if you have the `glob' function. */
 #undef HAVE_GLOB
 
@@ -806,12 +809,18 @@
 /* Shared data */
 #undef SHARE_DIR
 
+/* The size of `pthread_t', as computed by sizeof. */
+#undef SIZEOF_PTHREAD_T
+
 /* The size of `size_t', as computed by sizeof. */
 #undef SIZEOF_SIZE_T
 
 /* The size of `time_t', as computed by sizeof. */
 #undef SIZEOF_TIME_T
 
+/* The size of `unsigned long', as computed by sizeof. */
+#undef SIZEOF_UNSIGNED_LONG
+
 /* define if (v)snprintf does not return length needed, (but length used) */
 #undef SNPRINTF_RET_BROKEN
 

configure

@@ -15256,6 +15256,74 @@ cat >>confdefs.h <<_ACEOF
 _ACEOF
 
 
+# The cast to long int works around a bug in the HP C Compiler
+# version HP92453-01 B.11.11.23709.GP, which incorrectly rejects
+# declarations like `int a3[[(sizeof (unsigned char)) >= 0]];'.
+# This bug is HP SR number 8606223364.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking size of unsigned long" >&5
+$as_echo_n "checking size of unsigned long... " >&6; }
+if ${ac_cv_sizeof_unsigned_long+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if ac_fn_c_compute_int "$LINENO" "(long int) (sizeof (unsigned long))" "ac_cv_sizeof_unsigned_long"        "$ac_includes_default"; then :
+
+else
+  if test "$ac_cv_type_unsigned_long" = yes; then
+     { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error 77 "cannot compute sizeof (unsigned long)
+See \`config.log' for more details" "$LINENO" 5; }
+   else
+     ac_cv_sizeof_unsigned_long=0
+   fi
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_sizeof_unsigned_long" >&5
+$as_echo "$ac_cv_sizeof_unsigned_long" >&6; }
+
+
+
+cat >>confdefs.h <<_ACEOF
+#define SIZEOF_UNSIGNED_LONG $ac_cv_sizeof_unsigned_long
+_ACEOF
+
+
+if test x_$ub_have_pthreads != x_yes; then
+	# The cast to long int works around a bug in the HP C Compiler
+# version HP92453-01 B.11.11.23709.GP, which incorrectly rejects
+# declarations like `int a3[[(sizeof (unsigned char)) >= 0]];'.
+# This bug is HP SR number 8606223364.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking size of pthread_t" >&5
+$as_echo_n "checking size of pthread_t... " >&6; }
+if ${ac_cv_sizeof_pthread_t+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if ac_fn_c_compute_int "$LINENO" "(long int) (sizeof (pthread_t))" "ac_cv_sizeof_pthread_t"        "$ac_includes_default"; then :
+
+else
+  if test "$ac_cv_type_pthread_t" = yes; then
+     { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error 77 "cannot compute sizeof (pthread_t)
+See \`config.log' for more details" "$LINENO" 5; }
+   else
+     ac_cv_sizeof_pthread_t=0
+   fi
+fi
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_sizeof_pthread_t" >&5
+$as_echo "$ac_cv_sizeof_pthread_t" >&6; }
+
+
+
+cat >>confdefs.h <<_ACEOF
+#define SIZEOF_PTHREAD_T $ac_cv_sizeof_pthread_t
+_ACEOF
+
+
+fi
 
 # add option to disable the evil rpath
 
@@ -20594,7 +20662,7 @@ if test "$ac_res" != no; then :
 
 fi
 
-for ac_func in tzset sigprocmask fcntl getpwnam endpwent getrlimit setrlimit setsid chroot kill chown sleep usleep random srandom recvmsg sendmsg writev socketpair glob initgroups strftime localtime_r setusercontext _beginthreadex endservent endprotoent fsync shmget accept4 getifaddrs if_nametoindex poll
+for ac_func in tzset sigprocmask fcntl getpwnam endpwent getrlimit setrlimit setsid chroot kill chown sleep usleep random srandom recvmsg sendmsg writev socketpair glob initgroups strftime localtime_r setusercontext _beginthreadex endservent endprotoent fsync shmget accept4 getifaddrs if_nametoindex poll gettid
 do :
   as_ac_var=`$as_echo "ac_cv_func_$ac_func" | $as_tr_sh`
 ac_fn_c_check_func "$LINENO" "$ac_func" "$as_ac_var"

configure.ac

@@ -487,6 +487,10 @@ AC_INCLUDES_DEFAULT
 #endif
 ])
 AC_CHECK_SIZEOF(size_t)
+AC_CHECK_SIZEOF([unsigned long])
+if test x_$ub_have_pthreads != x_yes; then
+	AC_CHECK_SIZEOF(pthread_t)
+fi
 
 # add option to disable the evil rpath
 ACX_ARG_RPATH
@@ -1644,7 +1648,7 @@ AC_LINK_IFELSE([AC_LANG_PROGRAM([
   AC_MSG_RESULT(no))
 
 AC_SEARCH_LIBS([setusercontext], [util])
-AC_CHECK_FUNCS([tzset sigprocmask fcntl getpwnam endpwent getrlimit setrlimit setsid chroot kill chown sleep usleep random srandom recvmsg sendmsg writev socketpair glob initgroups strftime localtime_r setusercontext _beginthreadex endservent endprotoent fsync shmget accept4 getifaddrs if_nametoindex poll])
+AC_CHECK_FUNCS([tzset sigprocmask fcntl getpwnam endpwent getrlimit setrlimit setsid chroot kill chown sleep usleep random srandom recvmsg sendmsg writev socketpair glob initgroups strftime localtime_r setusercontext _beginthreadex endservent endprotoent fsync shmget accept4 getifaddrs if_nametoindex poll gettid])
 AC_CHECK_FUNCS([setresuid],,[AC_CHECK_FUNCS([setreuid])])
 AC_CHECK_FUNCS([setresgid],,[AC_CHECK_FUNCS([setregid])])
 

daemon/stats.c

@@ -69,6 +69,10 @@
 #ifdef HAVE_SSL
 #include <openssl/ssl.h>
 #endif
+#include <poll.h> /* DEBUG */
+
+/** How long to wait for threads to transmit statistics, in msec. */
+#define STATS_THREAD_WAIT 60000
 
 /** add timers and the values do not overflow or become negative */
 static void
@@ -380,6 +384,28 @@ void server_stats_obtain(struct worker* worker, struct worker* who,
 		worker_send_cmd(who, worker_cmd_stats);
 	else 	worker_send_cmd(who, worker_cmd_stats_noreset);
 	verbose(VERB_ALGO, "wait for stats reply");
+	if(tube_wait_timeout(worker->cmd, STATS_THREAD_WAIT) == 0) {
+		verbose(VERB_OPS, "no response from thread %d"
+#ifdef HAVE_GETTID
+			" LWP %u"
+#endif
+#if defined(HAVE_PTHREAD) && defined(SIZEOF_PTHREAD_T) && defined(SIZEOF_UNSIGNED_LONG)
+#  if SIZEOF_PTHREAD_T == SIZEOF_UNSIGNED_LONG
+			" pthread 0x%lx"
+#  endif
+#endif
+			,
+			who->thread_num
+#ifdef HAVE_GETTID
+			, (unsigned)who->thread_tid
+#endif
+#if defined(HAVE_PTHREAD) && defined(SIZEOF_PTHREAD_T) && defined(SIZEOF_UNSIGNED_LONG)
+#  if SIZEOF_PTHREAD_T == SIZEOF_UNSIGNED_LONG
+			, (unsigned long)*((unsigned long*)&who->thr_id)
+#  endif
+#endif
+			);
+	}
 	if(!tube_read_msg(worker->cmd, &reply, &len, 0))
 		fatal_exit("failed to read stats over cmd channel");
 	if(len != (uint32_t)sizeof(*s))

daemon/worker.c

@@ -1903,6 +1903,9 @@ worker_init(struct worker* worker, struct config_file *cfg,
 	struct dt_env* dtenv = &worker->dtenv;
 #else
 	void* dtenv = NULL;
+#endif
+#ifdef HAVE_GETTID
+	worker->thread_tid = gettid();
 #endif
 	worker->need_to_exit = 0;
 	worker->base = comm_base_create(do_sigs);

daemon/worker.h

@@ -86,6 +86,10 @@ struct worker {
 	struct daemon* daemon;
 	/** thread id */
 	ub_thread_type thr_id;
+#ifdef HAVE_GETTID
+	/** thread tid, the LWP id. */
+	pid_t thread_tid;
+#endif
 	/** pipe, for commands for this worker */
 	struct tube* cmd;
 	/** the event base this worker works with */

doc/Changelog

@@ -3,6 +3,9 @@
 	  work on OpenBSD.
 	- Slow down log frequency of write wait failures.
 	- Fix to set out of file descriptor warning to operational verbosity.
+	- Fix to log a verbose message at operational notice level if a
+	  thread is not responding, to stats requests. It is logged with
+	  thread identifiers.
 
 31 August 2022: Wouter
 	- Fix to avoid process wide fcntl calls mixed with nonblocking

util/tube.c

@@ -424,6 +424,28 @@ int tube_wait(struct tube* tube)
 	return pollit(tube->sr, NULL);
 }
 
+int tube_wait_timeout(struct tube* tube, int msec)
+{
+	struct timeval t;
+	int fd = tube->sr;
+	fd_set r;
+	t.tv_sec = msec/1000;
+	t.tv_usec = (msec%1000)*1000;
+#ifndef S_SPLINT_S
+	FD_ZERO(&r);
+	FD_SET(FD_SET_T fd, &r);
+#endif
+	while(1) {
+		if(select(fd+1, &r, NULL, NULL, &t) == -1) {
+			if(errno == EAGAIN || errno == EINTR)
+				continue;
+			return -1;
+		}
+		break;
+	}
+	return (int)(FD_ISSET(fd, &r));
+}
+
 int tube_read_fd(struct tube* tube)
 {
 	return tube->sr;
@@ -649,6 +671,26 @@ int tube_wait(struct tube* tube)
 	return 1;
 }
 
+int tube_wait_timeout(struct tube* tube, int msec)
+{
+	/* block on eventhandle */
+	DWORD res = WSAWaitForMultipleEvents(
+		1 /* one event in array */,
+		&tube->event /* the event to wait for, our pipe signal */,
+		0 /* wait for all events is false */,
+		msec /* wait for timeout */,
+		0 /* we are not alertable for IO completion routines */
+		);
+	if(res == WSA_WAIT_TIMEOUT) {
+		return 0;
+	}
+	if(res == WSA_WAIT_IO_COMPLETION) {
+		/* a bit unexpected, since we were not alertable */
+		return -1;
+	}
+	return 1;
+}
+
 int tube_read_fd(struct tube* ATTR_UNUSED(tube))
 {
 	/* nothing sensible on Windows */

util/tube.h

@@ -204,6 +204,14 @@ int tube_poll(struct tube* tube);
  */
 int tube_wait(struct tube* tube);
 
+/**
+ * Wait for data to be ready with a timeout.
+ * @param tube: the tube to wait on.
+ * @param msec: timeout in milliseconds.
+ * @return 1 if there is something to read within timeout, readability.
+ * 	0 on a timeout. On failures -1, like errors. */
+int tube_wait_timeout(struct tube* tube, int msec);
+
 /**
  * Get FD that is readable when new information arrives.
  * @param tube