Pause group after 5 consecutive lost WUs. #305, Ensure WU is saved to…

… DB in case shutdown Windows kills the process. #290

CHANGELOG.md

@@ -1,6 +1,10 @@
 Folding@home Client Changelog
 =============================
 
+## v8.4.7
+ - Pause group after 5 consecutive lost WUs. #305
+ - Ensure WU is saved to DB in case shutdown Windows kills the process. #290
+
 ## v8.4.6
  - Attempt more graceful shutdown in Windows.
  - Increase clock skew detection threshold from 15s to 5m.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "fah-client",
-  "version": "8.4.6",
+  "version": "8.4.7",
   "bin": {"fah-client": "./fah-client"},
   "author": "Joseph Coffland <[email protected]>",
   "homepage": "https://foldingathome.org/",

src/fah/client/Group.cpp

@@ -72,7 +72,9 @@ Group::Units Group::units() const {return Units(*app.getUnits(), name);}
 
 
 void Group::setState(const JSON::Value &msg) {
+  bool wasPaused = config->getPaused();
   config->setState(msg);
+  if (wasPaused && !config->getPaused()) clearErrors();
   triggerUpdate();
 }
 
@@ -115,14 +117,30 @@ void Group::shutdown(function<void ()> cb) {
 }
 
 
-void Group::unitComplete(bool success) {
-  if (success) {
-    failures = 0;
-    setWait(0);
+void Group::clearErrors() {
+  lostWUs  = 0;
+  failures = 0;
+  setWait(0);
+  insert("failed_wus", 0);
+  insert("lost_wus",   0);
+  insert("failed", "");
+}
+
 
-  } else {
-    failures++;
+void Group::unitComplete(bool success, bool downloaded) {
+  if (success) clearErrors();
+  else {
+    insert("failed_wus", ++failures);
     setWait(pow(2, std::min(failures, 10U)));
+
+    if (downloaded) {
+      insert("lost_wus", ++lostWUs);
+
+      if (4 < lostWUs) {
+        insert("failed", "Paused due too many failed Work Units.");
+        config->setPaused(true);
+      }
+    }
   }
 
   triggerUpdate();
@@ -178,8 +196,7 @@ void Group::update() {
   }
 
   // No further action if paused or idle
-  if (config->getPaused() || waitForIdle())
-    return setWait(0); // Pausing clears wait timer
+  if (config->getPaused() || waitForIdle()) return;
 
   // Wait on failures
   auto now = Time::now();

src/fah/client/Group.h

@@ -44,6 +44,7 @@ namespace FAH {
       cb::SmartPointer<Config> config;
 
       cb::Event::EventPtr event;
+      uint32_t lostWUs    = 0;
       uint32_t failures   = 0;
       uint64_t waitUntil  = 0;
 
@@ -95,9 +96,7 @@ namespace FAH {
       Group(App &app, const std::string &name);
 
       const std::string &getName() const {return name;}
-
       Config &getConfig() const {return *config;}
-
       Units units() const;
 
       void setState(const cb::JSON::Value &msg);
@@ -109,7 +108,8 @@ namespace FAH {
       bool hasUnrunWUs() const;
       void triggerUpdate();
       void shutdown(std::function<void ()> cb);
-      void unitComplete(bool success);
+      void clearErrors();
+      void unitComplete(bool success, bool downloaded);
 
       void save();
       void remove();

src/fah/client/Groups.cpp

@@ -115,12 +115,12 @@ void Groups::triggerUpdate() {
 void Groups::setState(const JSON::Value &msg) {
   if (msg.hasString("group")) {
     LOG_INFO(1, "Group state " << msg.getString("state"));
-    getGroup(msg.getString("group")).getConfig().setState(msg);
+    getGroup(msg.getString("group")).setState(msg);
 
   } else {
     LOG_INFO(1, "Machine state " << msg.getString("state"));
     for (auto &name: keys())
-      getGroup(name).getConfig().setState(msg);
+      getGroup(name).setState(msg);
   }
 }
 

src/fah/client/Unit.cpp

@@ -404,7 +404,7 @@ void Unit::dumpWU() {
 
 
 void Unit::save() {
-  if (getState() < UNIT_RUN || getState() == UNIT_DONE) return;
+  if (getState() < UNIT_CORE || getState() == UNIT_DONE) return;
 
   JSON::BufferWriter writer;
 
@@ -448,7 +448,8 @@ void Unit::next() {
       return monitorRun();
     }
 
-    return finalizeRun();
+    finalizeRun();
+    return save();
   }
 
   // Handle pause
@@ -853,7 +854,7 @@ void Unit::clean(const string &result) {
   TRY_CATCH_ERROR(app.getDB("units").unset(id));
 
   setState(UNIT_DONE);
-  group->unitComplete(success);
+  group->unitComplete(result == "credited", UNIT_CORE < getState());
 }
 
 
@@ -1080,7 +1081,7 @@ void Unit::downloadResponse(const JSON::ValuePtr &data) {
 
   setState(UNIT_CORE);
   this->data = data;
-  save(); // Not strictly necessary
+  save();
 }