no trickles while net comm suspended

svn path=/trunk/boinc/; revision=7329

checkin_notes

@@ -10570,4 +10570,19 @@ Bruce 13 Aug 2005
         BOINCGUIApp.h
         ViewWork.cpp
 
+David  14 Aug 2005
+    - Don't do scheduler RPC to send trickle-up message
+        if network comm is suspended.
+        (previously we were setting sched_rpc_pending
+        when have trickle up, same as user "update" command,
+        which overrides comm suspended).
+        Add new flag PROJECT::trickle_up_pending
+        (suggested by Glenn Dill)
+
+    client/
+        app_control.C
+        client_state.h
+        client_types.C,h
+        cs_scheduler.C
+        scheduler_op.h
 

client/app_control.C

@@ -833,7 +833,7 @@ bool ACTIVE_TASK::get_trickle_up_msg() {
         if (match_tag(msg_buf, "<have_new_trickle_up/>")) {
             retval = move_trickle_file();
             if (!retval) {
-                wup->project->sched_rpc_pending = true;
+                wup->project->trickle_up_pending = true;
             }
         }
         if (match_tag(msg_buf, "<have_new_upload_file/>")) {

client/client_state.h

@@ -325,6 +325,7 @@ private:
 
     PROJECT* find_project_with_overdue_results();
     PROJECT* next_project_sched_rpc_pending();
+    PROJECT* next_project_trickle_up_pending();
     //bool some_project_rpc_ok();
     bool scheduler_rpc_poll();
     double time_until_work_done(PROJECT*, int, double);

client/client_types.C

@@ -80,6 +80,7 @@ void PROJECT::init() {
     min_report_min_rpc_time = 0;
     master_url_fetch_pending = false;
     sched_rpc_pending = false;
+    trickle_up_pending = false;
     tentative = false;
     anonymous_platform = false;
     non_cpu_intensive = false;
@@ -98,11 +99,6 @@ void PROJECT::init() {
     duration_correction_factor = 1;
 }
 
-PROJECT::~PROJECT() {
-}
-
-
-
 // parse project fields from client_state.xml
 //
 int PROJECT::parse_state(MIOFILE& in) {
@@ -163,6 +159,7 @@ int PROJECT::parse_state(MIOFILE& in) {
         }
         else if (match_tag(buf, "<master_url_fetch_pending/>")) master_url_fetch_pending = true;
         else if (match_tag(buf, "<sched_rpc_pending/>")) sched_rpc_pending = true;
+        else if (match_tag(buf, "<trickle_up_pending/>")) trickle_up_pending = true;
         else if (match_tag(buf, "<send_file_list/>")) send_file_list = true;
         else if (match_tag(buf, "<non_cpu_intensive/>")) non_cpu_intensive = true;
         else if (match_tag(buf, "<suspended_via_gui/>")) suspended_via_gui = true;
@@ -248,6 +245,7 @@ int PROJECT::write_state(MIOFILE& out, bool gui_rpc) {
         duration_correction_factor,
         master_url_fetch_pending?"    <master_url_fetch_pending/>\n":"",
         sched_rpc_pending?"    <sched_rpc_pending/>\n":"",
+        trickle_up_pending?"    <trickle_up_pending/>\n":"",
         send_file_list?"    <send_file_list/>\n":"",
         non_cpu_intensive?"    <non_cpu_intensive/>\n":"",
         suspended_via_gui?"    <suspended_via_gui/>\n":"",
@@ -307,7 +305,7 @@ void PROJECT::copy_state_fields(PROJECT& p) {
     master_fetch_failures = p.master_fetch_failures;
     min_rpc_time = p.min_rpc_time;
     master_url_fetch_pending = p.master_url_fetch_pending;
-    sched_rpc_pending = p.sched_rpc_pending;
+    trickle_up_pending = p.trickle_up_pending;
     safe_strcpy(code_sign_key, p.code_sign_key);
     short_term_debt = p.short_term_debt;
     long_term_debt = p.long_term_debt;

client/client_types.h

@@ -208,6 +208,7 @@ public:
     bool master_url_fetch_pending;
                                 // need to fetch and parse the master URL
     bool sched_rpc_pending;     // contact scheduling server for preferences
+    bool trickle_up_pending;    // have trickle up to send
     bool tentative;             // master URL and account ID not confirmed
     bool anonymous_platform;    // app_versions.xml file found in project dir;
                             // use those apps rather then getting from server
@@ -291,7 +292,7 @@ public:
 #endif
 
     PROJECT();
-    ~PROJECT();
+    ~PROJECT(){}
     void init();
     void copy_state_fields(PROJECT&);
     char *get_project_name();

client/cs_scheduler.C

@@ -140,6 +140,21 @@ PROJECT* CLIENT_STATE::next_project_sched_rpc_pending() {
     return 0;
 }
 
+PROJECT* CLIENT_STATE::next_project_trickle_up_pending() {
+    unsigned int i;
+    PROJECT* p;
+
+    for (i=0; i<projects.size(); i++) {
+        p = projects[i];
+        if (p->waiting_until_min_rpc_time()) continue;
+        if (p->suspended_via_gui) continue;
+        if (p->trickle_up_pending) {
+            return p;
+        }
+    }
+    return 0;
+}
+
 // Return the best project to fetch work from, NULL if none
 //
 // Basically, pick the one with largest long term debt - amount of current work
@@ -530,17 +545,22 @@ int CLIENT_STATE::compute_work_requests() {
         p->work_request_urgency = WORK_FETCH_DONT_NEED;
         if (!p->contactable()) continue;
 
-        // if the projects have been running in round robin without resort to EDF, then
-        // all projects will have a LT debt greater than 
+        // if system has been running in round robin,
+        // then all projects will have a LT debt greater than 
         // -global_prefs.cpu_scheduling_period_minutes * 60
         // Therefore any project that has a LT debt greater than this 
         // is a candidate for more work.
         // Also if the global need is immediate, we need to get work from 
-        // someplace - anyplace that can be contacted, even if the LT debt
-        // is extremely negative.
-        if ((p->long_term_debt < -global_prefs.cpu_scheduling_period_minutes * 60) && (overall_work_fetch_urgency != WORK_FETCH_NEED_IMMEDIATELY)) continue;
+        // any contactable project, even if its LT debt is extremely negative.
+        //
+        if ((p->long_term_debt < -global_prefs.cpu_scheduling_period_minutes*60)
+            && (overall_work_fetch_urgency != WORK_FETCH_NEED_IMMEDIATELY)
+        ) {
+            continue;
+        }
 
         // if it is non cpu intensive and we have work, we don't need any more.
+        //
         if (p->non_cpu_intensive && p->runnable()) continue;
 
         int min_results = proj_min_results(p, prrs);
@@ -623,6 +643,12 @@ bool CLIENT_STATE::scheduler_rpc_poll() {
             break;
         }
         if (network_suspended || activities_suspended) break;
+        p = next_project_trickle_up_pending();
+        if (p) {
+            scheduler_op->init_op_project(p, REASON_TRICKLE_UP);
+            action = true;
+            break;
+        }
         
         // report overdue results
         //
@@ -967,6 +993,7 @@ int CLIENT_STATE::handle_scheduler_reply(
         project->send_file_list = true;
     }
     project->sched_rpc_pending = false;
+    project->trickle_up_pending = false;
 
     // handle delay request
     //
@@ -991,7 +1018,9 @@ bool CLIENT_STATE::should_get_work() {
     for (unsigned int i = 0; i < results.size();++i) {
         tot_cpu_time_remaining += results[i]->estimated_cpu_time_remaining();
     }
-    if (tot_cpu_time_remaining < global_prefs.work_buf_min_days * SECONDS_PER_DAY) return true;
+    if (tot_cpu_time_remaining < global_prefs.work_buf_min_days*SECONDS_PER_DAY) {
+        return true;
+    }
 
     // if the CPU started this time period overloaded,
     // let it process for a while to get out of the CPU overload state.
@@ -999,9 +1028,8 @@ bool CLIENT_STATE::should_get_work() {
     if (!work_fetch_no_new_work) {
         set_scheduler_modes();
     }
-    bool ret = !work_fetch_no_new_work;
 
-    return ret;
+    return !work_fetch_no_new_work;
 }
 
 void PROJECT::set_rrsim_proc_rate(double per_cpu_proc_rate, double rrs) {

client/scheduler_op.C

@@ -227,6 +227,7 @@ int SCHEDULER_OP::start_rpc(PROJECT* p) {
         case REASON_USER_REQ: why = "Requested by user"; break;
         case REASON_NEED_WORK: why = "To fetch work"; break;
         case REASON_RESULTS_DUE: why = "To report results"; break;
+        case REASON_TRICKLE_UP: why = "To send trickle-up message"; break;
         default: why = "Unknown";
         }
         msg_printf(p, MSG_INFO,  "Reason: %s", why);

client/scheduler_op.h

@@ -45,7 +45,8 @@
 typedef enum {
     REASON_USER_REQ,
     REASON_RESULTS_DUE,
-    REASON_NEED_WORK
+    REASON_NEED_WORK,
+    REASON_TRICKLE_UP
 } SCHEDULER_OP_REASON ;
 
 // default constants related to scheduler RPC policy

doc/boinc_dev.php

@@ -28,22 +28,15 @@ Also, read:
 
 The following medium-to-large development projects are available:
 <ul>
-<li> Core client: use the 
-<a href=http://curl.haxx.se/libcurl/>libcurl</a> library
-or some other open-source HTTP library,
-replacing the HTTP_OP and PROXY classes.
-This would let the core client use HTTPS.
-(Carl C. is working on this).
-
 <li> BOINC Manager:
 Change the Statistics tab to use a single graph
 with different colors for different projects.
 
 <li> BOINC Manager:
-add progress bars for file transfers and in-progress results.
+Use progress bars for file transfers and in-progress results.
 
 <li> BOINC Manager:
-add pie charts for disk usage
+Use pie charts for disk usage
 
 <li> Disk space management: prevent disk space usage from
 exceeding user preferences,
@@ -106,7 +99,7 @@ Various implementation notes:
 <li> <a href=trickle_impl.php>Trickle messages</a>
 <li> <a href=version_diff.txt>How to see what has changed
 between two versions of an executable</a>.
-<li> <a href=acct_mgt.php>Account management systems</a>
+<li> <a href=acct_mgt_new.php>Account management systems</a>
 <li> <a href=spec.txt>Spec info for RPMs</a>
 <li> <a href=stats_summary.php>Statistics summaries</a>
 </ul>

doc/boinc_news.inc

@@ -2,6 +2,13 @@
 
 $project_news = array(
 
+array("August 14, 2005",
+    "We are <a href=new_setup.php>redesigning BOINC</a> to use
+    a user-supplied identifier and password
+    instead of a system-supplied random key
+    for identifying an account.
+    Comments on the design are welcome."
+),
 array("August 9, 2005",
     "BOINC is transitioning to use <a href=http://curl.haxx.se/>libcurl</a>
     for HTTP operations.

doc/server_debug.php

@@ -24,7 +24,15 @@ for command-line apps it's written to your terminal).
 This is verbose but extremely useful for tracking down
 database-level problems.
 
-<h2>Scheduler single-stepping</h2>
+<h2>Getting core dumps from the scheduler</h2>
+<p>
+In sched/main.C put:
+<pre>
+#define DUMP_CORE_ON_SEGV 1
+</pre>
+and recompile.
+
+<h2>Running the scheduler under a debugger</h2>
 The scheduler is a CGI program.
 It reads from stdin and writes to stdout,
 so you can also run it with a command-line debugger like gdb: