Canceled RPC are a pain right now (until we move to a N/1-port model, they are likely...

[simgrid.git] / src / gras / Msg / rpc.c

diff --git a/src/gras/Msg/rpc.c b/src/gras/Msg/rpc.c
index 85d30d2..eace210 100644 (file)
--- a/src/gras/Msg/rpc.c
+++ b/src/gras/Msg/rpc.c
@@ -9,11 +9,10 @@
 
 #include "gras/Msg/msg_private.h"
                 
-//XBT_LOG_NEW_DEFAULT_SUBCATEGORY(gras_rpc,gras_msg,"RPCing");
-
 xbt_set_t _gras_rpctype_set = NULL;
+xbt_dynar_t _gras_rpc_cancelled = NULL;
 
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(gras_rpc,gras_msg,"RPC mecanism");
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(gras_msg_rpc,gras_msg,"RPC mecanism");
 
 
 /** @brief declare a new versionned RPC type of the given name and payloads
@@ -69,10 +68,26 @@ static int msgfilter_rpcID(gras_msg_t msg, void* ctx) {
   unsigned long int ID= *(unsigned long int*)ctx;
   int res = msg->ID == ID && 
     (msg->kind == e_gras_msg_kind_rpcanswer || msg->kind == e_gras_msg_kind_rpcerror);
+  int cursor;
+  gras_msg_cb_ctx_t rpc_ctx;
+     
 
   DEBUG5("Filter a message of ID %lu, type '%s' and kind '%s'. Waiting for ID=%lu. %s",
         msg->ID,msg->type->name,e_gras_msg_kind_names[msg->kind],ID,
         res?"take it": "reject");
+   
+  if (res && !_gras_rpc_cancelled)
+     return res;
+  
+  /* Check whether it is an old answer to a message we already canceled */
+  xbt_dynar_foreach(_gras_rpc_cancelled,cursor,rpc_ctx) {
+     if (msg->ID == rpc_ctx->ID && msg->kind==e_gras_msg_kind_rpcanswer) {
+       VERB1("Got an answer to the already canceled (timeouted?) RPC %ld. Ignore it (leaking the payload!).",msg->ID);
+       xbt_dynar_cursor_rm (_gras_rpc_cancelled, &cursor);
+       return 1;
+     }
+  }
+   
   return res;
 }
 
@@ -84,16 +99,25 @@ gras_msg_rpc_async_call(gras_socket_t server,
                        void *request) {
   gras_msg_cb_ctx_t ctx = xbt_new0(s_gras_msg_cb_ctx_t,1);
 
+  if (msgtype->ctn_type) {
+    xbt_assert1(request,
+               "RPC type '%s' convey a payload you must provide",
+               msgtype->name);
+  } else {
+    xbt_assert1(!request,
+               "No payload was declared for RPC type '%s'",
+               msgtype->name);
+  }
+
   ctx->ID = last_msg_ID++;
   ctx->expeditor = server;
   ctx->msgtype=msgtype;
   ctx->timeout=timeOut;
 
-  VERB5("Send to %s:%d a RPC of type '%s' (ID=%lu) (exception%s caught)",
+  VERB4("Send to %s:%d a RPC of type '%s' (ID=%lu)",
        gras_socket_peer_name(server),
        gras_socket_peer_port(server),
-       msgtype->name,ctx->ID,
-       (__xbt_ex_ctx()->ctx_caught?"":" not"));
+       msgtype->name,ctx->ID);
 
   gras_msg_send_ext(server, e_gras_msg_kind_rpccall, ctx->ID, msgtype, request);
 
@@ -103,11 +127,38 @@ gras_msg_rpc_async_call(gras_socket_t server,
 /** @brief Wait teh answer of a RPC call previously launched asynchronously */
 void gras_msg_rpc_async_wait(gras_msg_cb_ctx_t ctx,
                             void *answer) {
+  xbt_ex_t e;
   s_gras_msg_t received;
 
-  gras_msg_wait_ext(ctx->timeout,
-                   ctx->msgtype, NULL, msgfilter_rpcID, &ctx->ID,
-                   &received);
+  if (ctx->msgtype->answer_type) {
+    xbt_assert1(answer,
+               "Answers to RPC '%s' convey a payload you must accept",
+               ctx->msgtype->name);
+  } else {
+    xbt_assert1(!answer,
+               "No payload was declared for answers to RPC '%s'",
+               ctx->msgtype->name);
+  }
+
+  TRY {
+     /* The filter returns 1 when we eat an old RPC answer to something canceled */
+     do {
+       gras_msg_wait_ext(ctx->timeout,
+                         ctx->msgtype, NULL, msgfilter_rpcID, &ctx->ID,
+                         &received);
+     } while (received.ID != ctx->ID);
+     
+  } CATCH(e) {
+     if (!_gras_rpc_cancelled)
+       _gras_rpc_cancelled = xbt_dynar_new(sizeof(ctx),NULL);
+     xbt_dynar_push(_gras_rpc_cancelled,&ctx);
+     INFO5("canceled RPC %ld pushed onto the stack (%s from %s:%d) Reason: %s",
+          ctx->ID,ctx->msgtype->name,
+          gras_socket_peer_name(ctx->expeditor),gras_socket_peer_port(ctx->expeditor),
+          e.msg);
+     RETHROW;
+  }
+   
   free(ctx);
   if (received.kind == e_gras_msg_kind_rpcerror) {
     xbt_ex_t e;
@@ -121,6 +172,7 @@ void gras_msg_rpc_async_wait(gras_msg_cb_ctx_t ctx,
      __xbt_ex_ctx()->ctx_ex.remote   = 1;
      __xbt_ex_ctx()->ctx_ex.host     = e.host;
      __xbt_ex_ctx()->ctx_ex.procname = e.procname;
+     __xbt_ex_ctx()->ctx_ex.pid      = e.pid;
      __xbt_ex_ctx()->ctx_ex.file     = e.file;
      __xbt_ex_ctx()->ctx_ex.line     = e.line;
      __xbt_ex_ctx()->ctx_ex.func     = e.func;
@@ -154,6 +206,10 @@ void gras_msg_rpccall(gras_socket_t server,
  */
 
 void gras_msg_rpcreturn(double timeOut,gras_msg_cb_ctx_t ctx,void *answer) {
+  DEBUG5("Return to RPC '%s' from %s:%d (tOut=%f, payl=%p)",
+        ctx->msgtype->name,
+        gras_socket_peer_name(ctx->expeditor),gras_socket_peer_port(ctx->expeditor),
+        timeOut,answer);
   gras_msg_send_ext(ctx->expeditor, e_gras_msg_kind_rpcanswer, 
                    ctx->ID, ctx->msgtype, answer);
 }