checkin changes

svn path=/trunk/boinc/; revision=3948

sched/handle_request.C

@@ -813,7 +813,7 @@ void process_request(
     int last_rpc_dayofyear;
     int current_rpc_dayofyear;
     bool ok_to_send_work = true;
-
+    bool have_no_work;
 
     // if different major version of BOINC, just send a message
     //
@@ -831,8 +831,12 @@ void process_request(
     // this isn't an initial RPC,
     // and client is requesting work, return without accessing DB
     //
+    lock_sema();
+    have_no_work = ss.no_work(g_pid);
+    unlock_sema();
+
     if ((sreq.work_req_seconds > 0)
-        && ss.no_work()
+        && have_no_work
         && (sreq.results.size() == 0)
         && (sreq.hostid != 0)
     ) {
@@ -845,16 +849,18 @@ void process_request(
         return;
     }
 
+    // FROM HERE ON DON'T RETURN; goto leave instead
+
     // now open the database
     //
     retval = open_database();
     if (retval) {
         send_message("Server can't open database", 3600);
-        return;
+        goto leave;
     }
 
     retval = authenticate_user(sreq, reply);
-    if (retval) return;
+    if (retval) goto leave;
     if (reply.user.id == 0) {
         log_messages.printf(SCHED_MSG_LOG::CRITICAL, "No user ID!\n");
     }
@@ -891,7 +897,7 @@ void process_request(
             SCHED_MSG_LOG::CRITICAL, "[HOST#%d] platform '%s' not found\n",
             reply.host.id, sreq.platform_name
         );
-        return;
+        goto leave;
     }
 
     handle_global_prefs(sreq, reply);
@@ -906,7 +912,7 @@ void process_request(
 
     // if last RPC was within config.min_sendwork_interval, don't send work
     //
-    if (ok_to_send_work && sreq.work_req_seconds > 0) {
+    if (!have_no_work && ok_to_send_work && sreq.work_req_seconds > 0) {
         if (config.min_sendwork_interval) {
             double diff = dtime() - last_rpc_time;
             if (diff < config.min_sendwork_interval) {
@@ -935,6 +941,11 @@ void process_request(
     }
 
     update_host_record(reply.host);
+
+leave:
+    if (!have_no_work) {
+        ss.restore_work(g_pid);
+    }
 }
 
 void handle_request(

sched/main.C

@@ -66,6 +66,7 @@ bool use_files = false;     // use disk files for req/reply msgs (for debugging)
 
 SCHED_CONFIG config;
 key_t sema_key;
+int g_pid;
 
 void send_message(char* msg, int delay) {
     printf(
@@ -96,7 +97,7 @@ int open_database() {
 
 int main() {
     FILE* fin, *fout;
-    int i, retval, pid;
+    int i, retval;
     char req_path[256], reply_path[256], path[256];
     SCHED_SHMEM* ssp=0;
     void* p;
@@ -169,7 +170,7 @@ int main() {
         }
     }
 
-    pid = getpid();
+    g_pid = getpid();
 #ifdef _USING_FCGI_
     while(FCGI_Accept() >= 0) {
     counter++;
@@ -186,8 +187,8 @@ int main() {
         // (this makes it easy to save the input,
         // and to know the length of the output).
         //
-        sprintf(req_path, "%s%d_%u", REQ_FILE_PREFIX, pid, counter);
-        sprintf(reply_path, "%s%d_%u", REPLY_FILE_PREFIX, pid, counter);
+        sprintf(req_path, "%s%d_%u", REQ_FILE_PREFIX, g_pid, counter);
+        sprintf(reply_path, "%s%d_%u", REPLY_FILE_PREFIX, g_pid, counter);
         fout = fopen(req_path, "w");
         if (!fout) {
             log_messages.printf(SCHED_MSG_LOG::CRITICAL, "can't write request file\n");

sched/main.h

@@ -23,6 +23,9 @@
 
 extern SCHED_CONFIG config;
 extern key_t sema_key;
+extern int g_pid;
+extern void lock_sema();
+extern void unlock_sema();
 
 extern void send_message(char*, int);
 extern int open_database();

sched/sched_send.C

@@ -496,10 +496,8 @@ static void scan_work_array(
         // i.e. ones that don't require DB access
         // if any check fails, continue
 
-        switch (wu_result.state) {
-            case WR_STATE_EMPTY:
-            case WR_STATE_CHECKED_OUT:
-                continue;
+        if (wu_result.state != WR_STATE_PRESENT || wu_result.state != g_pid) {
+            continue;
         }
 
         if (wreq.infeasible_only && (wu_result.infeasible_count==0)) {

sched/sched_shmem.C

@@ -157,12 +157,26 @@ APP_VERSION* SCHED_SHMEM::lookup_app_version(
     return best_avp;
 }
 
-bool SCHED_SHMEM::no_work() {
+bool SCHED_SHMEM::no_work(int pid) {
     int i;
 
-    if (!ready) return false;
+    if (!ready) return true;
     for (i=0; i<max_wu_results; i++) {
-        if (wu_results[i].state == WR_STATE_PRESENT) return false;
+        if (wu_results[i].state == WR_STATE_PRESENT) {
+            wu_results[i].state = pid;
+            return false;
+        }
     }
     return true;
 }
+
+void SCHED_SHMEM::restore_work(int pid) {
+    int i;
+
+    for (i=0; i<max_wu_results; i++) {
+        if (wu_results[i].state == pid) {
+            wu_results[i].state = WR_STATE_PRESENT;
+            return;
+        }
+    }
+}

sched/sched_shmem.h

@@ -80,7 +80,8 @@ struct SCHED_SHMEM {
     int verify();
     int scan_tables();
     bool have_app(int);
-    bool no_work();
+    bool no_work(int pid);
+    void restore_work(int pid);
 
     APP* lookup_app(int);
     APP_VERSION* lookup_app_version(int appid, int platform, int version);

sched/show_shmem.C

@@ -62,6 +62,8 @@ int main() {
         case WR_STATE_CHECKED_OUT:
             printf("%d: checked out: result %d\n", i, wu_result.resultid);
             break;
+        default:
+            printf("%d: PID %d: result %d\n", i, wu_result.state, wu_result.resultid);
         }
     }
 }