scheduler race condition

svn path=/trunk/boinc/; revision=9666

checkin_notes

@@ -2906,3 +2906,17 @@ Rom    16 Mar 2006
     clientlib/win/
         boinc_dll.cpp
         NetworkTracker.cpp
+
+David  16 Mar 2006
+    - scheduler: when mark a result as sent, make sure its
+        server_state is what we expected (e.g. UNSENT).
+        This guards against two scheduler instances trying to
+        send the same result at about the same time.
+        (from Bruce Allen)
+
+    db/
+        boinc_db.C,h
+        db_base.C,h
+    sched/
+        sched_resend.C
+        sched_send.C

db/boinc_db.C

@@ -736,16 +736,22 @@ void DB_RESULT::db_print_values(char* buf){
     UNESCAPE(stderr_out);
 }
 
-// called from scheduler when dispatch this result
+// called from scheduler when dispatch this result.
+// The "... and server_state=%d" is a safeguard against
+// the case where another scheduler tries to send this result at the same time
 //
-int DB_RESULT::update_subset() {
+int DB_RESULT::mark_as_sent(int old_server_state) {
     char query[MAX_QUERY_LEN];
+    int retval;
 
     sprintf(query,
-        "update result set server_state=%d, hostid=%d, userid=%d, sent_time=%d, report_deadline=%d where id=%d",
-        server_state, hostid, userid, sent_time, report_deadline, id
+        "update result set server_state=%d, hostid=%d, userid=%d, sent_time=%d, report_deadline=%d where id=%d and server_state=%d",
+        server_state, hostid, userid, sent_time, report_deadline, id,
+        old_server_state
     );
-    return db->do_query(query);
+    retval = db->do_query(query);
+    if (retval) return retval;
+    if (db->affected_rows() != 1) return ERR_DB_NOT_FOUND;
 }
 
 void DB_RESULT::db_parse(MYSQL_ROW &r) {

db/boinc_db.h

@@ -560,7 +560,7 @@ class DB_RESULT : public DB_BASE, public RESULT {
 public:
     DB_RESULT(DB_CONN* p=0);
     int get_id();
-    int update_subset();
+    int mark_as_sent(int old_server_state);
     void db_print(char*);
     void db_print_values(char*);
     void db_parse(MYSQL_ROW &row);

db/db_base.C

@@ -63,6 +63,10 @@ int DB_CONN::do_query(const char* p) {
     return retval;
 }
 
+int DB_CONN::affected_rows() {
+    return mysql_affected_rows(mysql);
+}
+
 int DB_CONN::insert_id() {
     int retval;
     MYSQL_ROW row;

db/db_base.h

@@ -62,6 +62,7 @@ public:
     DB_CONN();
     int open(char* name, char* host, char* user, char* passwd);
     int do_query(const char*);
+    int affected_rows();
     void close();
     int insert_id();
     void print_error(const char*);

doc/boinc_news.inc

@@ -1,6 +1,16 @@
 <?
 
 $project_news = array(
+array("March 15, 2006",
+    "The <a href=http://attribution.cpdn.org/>Seasonal Attribution Project</a>,
+    a spin-off of Climateprediction.net,
+    seeks to determine the extent to which extreme weather events
+    are attributable to human-induced climate change.
+    Its application is a high-resolution simulation of the world's climate,
+    and is recommended only for computers
+    with at least 1 GB of RAM, and a 2.4GHz Pentium or faster processor.
+    "
+),
 array("March 5, 2006",
     "Tony/Knightrider/Chuggybus has created BOINC coinage.
     See the <a href=images/coins/>large</a>

doc/graphics.php

@@ -12,12 +12,12 @@ General comments about graphics:
 <ul>
 
 <li>
-The graphics often provide a
-visualization of what the application is doing scientifically.
-The graphics code must run concurrently with the science code,
-and must share data structures with it.
-This can be done using any of
-several <b>process structures</b>.
+It is desirable for the graphics to provide a
+visualization of what the application is currently doing.
+Usually the easiest way to do this is for
+the graphics code to run concurrently with the science code,
+and to share data structures with it.
+This can be done using any of several <b>process structures</b>.
 For example, you can use separate threads in a single address space,
 or shared memory between separate programs.
 BOINC supports various process structures (see below).

doc/links.php

@@ -112,6 +112,7 @@ language("French", array(
     site("http://boinc-quebec.org", "boinc-quebec.org")
 ));
 language("German", array(
+    site("http://www.seti-leipzig.de/", "SETI-Leipzig"),
     site("http://www.dc-gemeinschaft.de/", "DC - Gemeinschaft"),
     site("http://boinccast.podhost.de/", "BOINCcast (Podcast)"),
     site("http://www.boinc-team.de/", "BOINC@Heidelberg"),

sched/sched_resend.C

@@ -166,7 +166,7 @@ bool resend_lost_work(
                 possibly_give_result_new_deadline(result, wu, reply)
             ) {
                 result.report_deadline = time(0);
-                retval = result.update_subset();
+                retval = result.mark_as_sent(result.server_state);
                 if (retval) {
                     log_messages.printf(
                         SCHED_MSG_LOG::MSG_CRITICAL,

sched/sched_send.C

@@ -590,6 +590,7 @@ int add_result_to_reply(
     result.hostid = reply.host.id;
     result.userid = reply.user.id;
     result.sent_time = time(0);
+    int old_server_state = result.server_state;
 
     if (result.server_state != RESULT_SERVER_STATE_IN_PROGRESS) {
         // We are sending this result for the first time
@@ -617,7 +618,7 @@ int add_result_to_reply(
             result.id, reply.host.id
         );
     }
-    retval = result.update_subset();
+    retval = result.mark_as_sent(old_server_state);
     if (retval) {
         log_messages.printf(
             SCHED_MSG_LOG::MSG_CRITICAL,