Remove last remnants of old trpc thread exit protocol, clear trpc queue

  * Remove the shutting_down status in the TRPC_INSTANCE
  * Clear the TRPC message queue after failed connections
  * Add a few comments

diff --git a/include/trp_internal.h b/include/trp_internal.h
index a52edb8..8401bfa 100644 (file)
--- a/include/trp_internal.h
+++ b/include/trp_internal.h
@@ -142,7 +142,6 @@ struct trpc_instance {
   unsigned int port;
   TRP_CONNECTION *conn;
   TR_MQ *mq; /* msgs from master to trpc */
-  int shutting_down; /* 0 unless the TRPC is in the shutdown process */
 };
 
 /* TRP Server Instance Data */

diff --git a/tr/tr_trp.c b/tr/tr_trp.c
index 8304300..79a9a27 100644 (file)
--- a/tr/tr_trp.c
+++ b/tr/tr_trp.c
@@ -573,6 +573,10 @@ struct trpc_thread_data {
  * a TR_MQMSG_ABORT message is received on trpc->mq, the thread sends a
  * TR_MQMSG_TRPC_DISCONNECTED message to the trps thread, then cleans up and
  * terminates.
+ *
+ * The trps may continue queueing messages for this client even when the
+ * connection is down. To prevent the queue from growing endlessly, this thread
+ * should clear its queue after failed connection attempts.
  */
 static void *tr_trpc_thread(void *arg)
 {
@@ -596,6 +600,7 @@ static void *tr_trpc_thread(void *arg)
     tr_notice("tr_trpc_thread: failed to initiate connection to %s:%d.",
               trpc_get_server(trpc),
               trpc_get_port(trpc));
+    trpc_mq_clear(trpc); /* clear the queue even though we did not connect */
   } else {
     /* Retrieve the GSS name used by the peer for authentication */
     peer_gssname=trp_connection_get_peer(trpc_get_conn(trpc));

diff --git a/trp/trpc.c b/trp/trpc.c
index 5985980..aef5f7b 100644 (file)
--- a/trp/trpc.c
+++ b/trp/trpc.c
@@ -59,7 +59,6 @@ TRPC_INSTANCE *trpc_new (TALLOC_CTX *mem_ctx)
     trpc->server=NULL;
     trpc->port=0;
     trpc->conn=NULL;
-    trpc->shutting_down = 0;
     trpc->mq=tr_mq_new(trpc);
     if (trpc->mq==NULL) {
       talloc_free(trpc);

diff --git a/trp/trps.c b/trp/trps.c
index 83461d9..b737aba 100644 (file)
--- a/trp/trps.c
+++ b/trp/trps.c
@@ -265,13 +265,13 @@ TRP_RC trps_send_msg(TRPS_INSTANCE *trps, TRP_PEER *peer, const char *msg)
 
   /* get the connection for this peer */
   trpc=trps_find_trpc(trps, peer);
-  /* instead, let's let that happen and then clear the queue when an attempt to
-   * connect fails */
+  /* The peer connection (trpc) usually exists even if the connection is down.
+   * We will queue messages even if the connection is down. To prevent this from
+   * endlessly increasing the size of the queue, the trpc handler needs to clear
+   * its queue periodically, even if it is unable to send the messages
+   */
   if (trpc==NULL) {
     tr_warning("trps_send_msg: skipping message queued for missing TRP client entry.");
-  } else if (trpc->shutting_down) {
-    tr_debug("trps_send_msg: skipping message because TRP client is shutting down.");
-    rc = TRP_SUCCESS; /* it's ok that this didn't get sent, the connection will be gone in a moment */
   } else {
     mq_msg=tr_mq_msg_new(tmp_ctx, TR_MQMSG_TRPC_SEND, TR_MQ_PRIO_NORMAL);
     msg_dup=talloc_strdup(mq_msg, msg); /* get local copy in mq_msg context */