s4-repl: ensure we don't starve pending replication ops

when there was a continuous sequence of notify ops, we could leave the
replication ops starving. This ensures we run whichever was queued
first

4 files changed, 24 insertions, 6 deletions

diff --git a/source4/dsdb/repl/drepl_notify.c b/source4/dsdb/repl/drepl_notify.c
index 0178e97e55..25378129da 100644
--- a/source4/dsdb/repl/drepl_notify.c
+++ b/source4/dsdb/repl/drepl_notify.c
@@ -228,7 +228,7 @@ static void dreplsrv_notify_op_callback(struct tevent_req *subreq)
 
 	talloc_free(op);
 	s->ops.n_current = NULL;
-	dreplsrv_notify_run_ops(s);
+	dreplsrv_run_pending_ops(s);
 }
 
 /*
@@ -343,6 +343,7 @@ static WERROR dreplsrv_schedule_notify_sync(struct dreplsrv_service *service,
 	op->uSN           = uSN;
 	op->is_urgent	  = is_urgent;
 	op->replica_flags = replica_flags;
+	op->schedule_time = time(NULL);
 
 	DLIST_ADD_END(service->ops.notifies, op, struct dreplsrv_notify_operation *);
 	talloc_steal(service, op);
@@ -499,5 +500,4 @@ static void dreplsrv_notify_run(struct dreplsrv_service *service)
 	talloc_free(mem_ctx);
 
 	dreplsrv_run_pending_ops(service);
-	dreplsrv_notify_run_ops(service);
 }
diff --git a/source4/dsdb/repl/drepl_out_pull.c b/source4/dsdb/repl/drepl_out_pull.c
index 3e59ee95ac..0af232c67f 100644
--- a/source4/dsdb/repl/drepl_out_pull.c
+++ b/source4/dsdb/repl/drepl_out_pull.c
@@ -52,6 +52,7 @@ WERROR dreplsrv_schedule_partition_pull_source(struct dreplsrv_service *s,
 	op->fsmo_info   = fsmo_info;
 	op->callback    = callback;
 	op->cb_data	= cb_data;
+	op->schedule_time = time(NULL);
 
 	DLIST_ADD_END(s->ops.pending, op, struct dreplsrv_out_operation *);
 
@@ -126,17 +127,16 @@ done:
 	talloc_free(op);
 	s->ops.current = NULL;
 	dreplsrv_run_pending_ops(s);
-	dreplsrv_notify_run_ops(s);
 }
 
-void dreplsrv_run_pending_ops(struct dreplsrv_service *s)
+void dreplsrv_run_pull_ops(struct dreplsrv_service *s)
 {
 	struct dreplsrv_out_operation *op;
 	time_t t;
 	NTTIME now;
 	struct tevent_req *subreq;
 
-	if (s->ops.current || s->ops.n_current) {
+	if (s->ops.current) {
 		/* if there's still one running, we're done */
 		return;
 	}
diff --git a/source4/dsdb/repl/drepl_periodic.c b/source4/dsdb/repl/drepl_periodic.c
index 60f3581f5f..fe0fb6dd1c 100644
--- a/source4/dsdb/repl/drepl_periodic.c
+++ b/source4/dsdb/repl/drepl_periodic.c
@@ -116,5 +116,21 @@ static void dreplsrv_periodic_run(struct dreplsrv_service *service)
 	dreplsrv_ridalloc_check_rid_pool(service);
 
 	dreplsrv_run_pending_ops(service);
-	dreplsrv_notify_run_ops(service);
+}
+
+/*
+  run the next pending op, either a notify or a pull
+ */
+void dreplsrv_run_pending_ops(struct dreplsrv_service *s)
+{
+	if (!s->ops.notifies && !s->ops.pending) {
+		return;
+	}
+	if (!s->ops.notifies ||
+	    (s->ops.pending &&
+	     s->ops.notifies->schedule_time > s->ops.pending->schedule_time)) {
+		dreplsrv_run_pull_ops(s);
+	} else {
+		dreplsrv_notify_run_ops(s);
+	}
 }
diff --git a/source4/dsdb/repl/drepl_service.h b/source4/dsdb/repl/drepl_service.h
index 9ad62fd538..d69c641178 100644
--- a/source4/dsdb/repl/drepl_service.h
+++ b/source4/dsdb/repl/drepl_service.h
@@ -119,6 +119,7 @@ typedef void (*dreplsrv_extended_callback_t)(struct dreplsrv_service *,
 
 struct dreplsrv_out_operation {
 	struct dreplsrv_out_operation *prev, *next;
+	time_t schedule_time;
 
 	struct dreplsrv_service *service;
 
@@ -133,6 +134,7 @@ struct dreplsrv_out_operation {
 
 struct dreplsrv_notify_operation {
 	struct dreplsrv_notify_operation *prev, *next;
+	time_t schedule_time;
 
 	struct dreplsrv_service *service;
 	uint64_t uSN;