hydra-queue-runner: --build-one: correctly handle a cached build

Previously, the build ID would never flow through channels which
exited.

This patch tracks the buildOne state as part of State and exits avoids
waiting forever for new work.

The code around buildOnly is a bit rough, making this a bit weird to
implement but since it is only used for testing the value of improving
it on its own is a bit questionable.

src/hydra-queue-runner/builder.cc

@@ -148,7 +148,8 @@ State::StepResult State::doBuildStep(nix::ref<Store> destStore,
             localStore->printStorePath(step->drvPath), repeats + 1, machine->sshName, buildId, (dependents.size() - 1));
     }
 
-    bool quit = buildId == buildOne && step->drvPath == *buildDrvPath;
+    if (!buildOneDone)
+        buildOneDone = buildId == buildOne && step->drvPath == *buildDrvPath;
 
     RemoteResult result;
     BuildOutput res;
@@ -265,7 +266,7 @@ State::StepResult State::doBuildStep(nix::ref<Store> destStore,
         if (retry) {
             auto mc = startDbUpdate();
             stepFinished = true;
-            if (quit) exit(1);
+            if (buildOneDone) exit(1);
             return sRetry;
         }
     }
@@ -376,7 +377,7 @@ State::StepResult State::doBuildStep(nix::ref<Store> destStore,
         }
 
     } else
-        failStep(*conn, step, buildId, result, machine, stepFinished, quit);
+        failStep(*conn, step, buildId, result, machine, stepFinished);
 
     // FIXME: keep stats about aborted steps?
     nrStepsDone++;
@@ -386,7 +387,7 @@ State::StepResult State::doBuildStep(nix::ref<Store> destStore,
     machine->state->totalStepTime += stepStopTime - stepStartTime;
     machine->state->totalStepBuildTime += result.stopTime - result.startTime;
 
-    if (quit) exit(0); // testing hack; FIXME: this won't run plugins
+    if (buildOneDone) exit(0); // testing hack; FIXME: this won't run plugins
 
     return sDone;
 }
@@ -398,8 +399,7 @@ void State::failStep(
     BuildID buildId,
     const RemoteResult & result,
     Machine::ptr machine,
-    bool & stepFinished,
+    bool & stepFinished)
-    bool & quit)
 {
     /* Register failure in the database for all Build objects that
        directly or indirectly depend on this step. */
@@ -481,7 +481,7 @@ void State::failStep(
             b->finishedInDB = true;
             builds_->erase(b->id);
             dependentIDs.push_back(b->id);
-            if (buildOne == b->id) quit = true;
+            if (!buildOneDone && buildOne == b->id) buildOneDone = true;
         }
     }
 

src/hydra-queue-runner/dispatcher.cc

@@ -374,7 +374,6 @@ void State::abortUnsupported()
             if (!build) build = *dependents.begin();
 
             bool stepFinished = false;
-            bool quit = false;
 
             failStep(
                 *conn, step, build->id,
@@ -385,9 +384,9 @@ void State::abortUnsupported()
                     .startTime = now2,
                     .stopTime = now2,
                 },
-                nullptr, stepFinished, quit);
+                nullptr, stepFinished);
 
-            if (quit) exit(1);
+            if (buildOneDone) exit(1);
         }
     }
 

src/hydra-queue-runner/queue-monitor.cc

@@ -35,14 +35,17 @@ void State::queueMonitorLoop()
 
     unsigned int lastBuildId = 0;
 
-    while (true) {
+    bool quit = false;
+    while (!quit) {
         localStore->clearPathInfoCache();
 
         bool done = getQueuedBuilds(*conn, destStore, lastBuildId);
 
+        if (buildOne && buildOneDone) quit = true;
+
         /* Sleep until we get notification from the database about an
            event. */
-        if (done) {
+        if (done && !quit) {
             conn->await_notification();
             nrQueueWakeups++;
         } else
@@ -65,6 +68,8 @@ void State::queueMonitorLoop()
             processJobsetSharesChange(*conn);
         }
     }
+
+    exit(0);
 }
 
 
@@ -160,6 +165,7 @@ bool State::getQueuedBuilds(Connection & conn,
 
             /* Some step previously failed, so mark the build as
                failed right away. */
+            if (!buildOneDone && build->id == buildOne) buildOneDone = true;
             printMsg(lvlError, "marking build %d as cached failure due to ‘%s’",
                 build->id, localStore->printStorePath(ex.step->drvPath));
             if (!build->finishedInDB) {
@@ -231,6 +237,7 @@ bool State::getQueuedBuilds(Connection & conn,
             auto mc = startDbUpdate();
             pqxx::work txn(conn);
             time_t now = time(0);
+            if (!buildOneDone && build->id == buildOne) buildOneDone = true;
             printMsg(lvlInfo, "marking build %1% as succeeded (cached)", build->id);
             markSucceededBuild(txn, build, res, true, now, now);
             notifyBuildFinished(txn, build->id, {});
@@ -289,6 +296,8 @@ bool State::getQueuedBuilds(Connection & conn,
         for (auto & r : newRunnable)
             makeRunnable(r);
 
+        if (buildOne && newRunnable.size() == 0) buildOneDone = true;
+
         nrBuildsRead += nrAdded;
 
         /* Stop after a certain time to allow priority bumps to be

src/hydra-queue-runner/state.hh

@@ -367,6 +367,7 @@ private:
 
     /* Specific build to do for --build-one (testing only). */
     BuildID buildOne;
+    bool buildOneDone = false;
 
     /* Statistics per machine type for the Hydra auto-scaler. */
     struct MachineType
@@ -485,8 +486,7 @@ private:
         BuildID buildId,
         const RemoteResult & result,
         Machine::ptr machine,
-        bool & stepFinished,
+        bool & stepFinished);
-        bool & quit);
 
     Jobset::ptr createJobset(pqxx::work & txn,
         const std::string & projectName, const std::string & jobsetName);