planner: implement compaction planner policy

faasm · Apr 21, 2024 · 6b3063d · 6b3063d
1 parent 5b65b3e
commit 6b3063d
Show file tree

Hide file tree

Showing 17 changed files with 1,136 additions and 14 deletions.
diff --git a/.env b/.env
@@ -1,4 +1,4 @@
-FAABRIC_VERSION=0.17.0
-FAABRIC_CLI_IMAGE=faasm.azurecr.io/faabric:0.17.0
+FAABRIC_VERSION=0.18.0
+FAABRIC_CLI_IMAGE=faasm.azurecr.io/faabric:0.18.0
 COMPOSE_PROJECT_NAME=faabric-dev
 CONAN_CACHE_MOUNT_SOURCE=./conan-cache/
diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
@@ -20,7 +20,7 @@ jobs:
     if: github.event.pull_request.draft == false
     runs-on: ubuntu-latest
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
     env:
       DEPLOYMENT_TYPE: gha-ci
     steps:
@@ -34,7 +34,7 @@ jobs:
     if: github.event.pull_request.draft == false
     runs-on: ubuntu-latest
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
     steps:
       - name: "Check out code"
         uses: actions/checkout@v4
@@ -45,7 +45,7 @@ jobs:
     if: github.event.pull_request.draft == false
     runs-on: ubuntu-latest
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
     steps:
       - name: "Check out code"
         uses: actions/checkout@v4
@@ -65,7 +65,7 @@ jobs:
       REDIS_QUEUE_HOST: redis
       REDIS_STATE_HOST: redis
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
       options: --privileged
     services:
       redis:
@@ -104,7 +104,7 @@ jobs:
       REDIS_QUEUE_HOST: redis
       REDIS_STATE_HOST: redis
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
       options: --privileged
     services:
       redis:
@@ -156,7 +156,7 @@ jobs:
       REDIS_QUEUE_HOST: redis
       REDIS_STATE_HOST: redis
     container:
-      image: faasm.azurecr.io/faabric:0.17.0
+      image: faasm.azurecr.io/faabric:0.18.0
     services:
       redis:
         image: redis

diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-0.17.0
+0.18.0
diff --git a/include/faabric/batch-scheduler/BatchScheduler.h b/include/faabric/batch-scheduler/BatchScheduler.h
@@ -84,9 +84,11 @@ class BatchScheduler
 
     static int numSlots(const Host& host) { return host->slots; }
 
+    static int numUsedSlots(const Host& host) { return host->usedSlots; }
+
     static int numSlotsAvailable(const Host& host)
     {
-        return std::max<int>(0, numSlots(host) - host->usedSlots);
+        return std::max<int>(0, numSlots(host) - numUsedSlots(host));
     }
 
     static void claimSlots(Host& host, int numSlotsToClaim)
@@ -120,4 +122,6 @@ class BatchScheduler
 std::shared_ptr<BatchScheduler> getBatchScheduler();
 
 void resetBatchScheduler();
+
+void resetBatchScheduler(const std::string& newMode);
 }
diff --git a/include/faabric/batch-scheduler/CompactScheduler.h b/include/faabric/batch-scheduler/CompactScheduler.h
@@ -0,0 +1,35 @@
+#pragma once
+
+#include <faabric/batch-scheduler/BatchScheduler.h>
+#include <faabric/batch-scheduler/SchedulingDecision.h>
+#include <faabric/util/batch.h>
+
+namespace faabric::batch_scheduler {
+
+// This batch scheduler behaves in the same way than BinPack for NEW and
+// SCALE_CHANGE requests, but for DIST_CHANGE requests it tries to compact
+// to the fewest number of VMs.
+class CompactScheduler final : public BatchScheduler
+{
+  public:
+    std::shared_ptr<SchedulingDecision> makeSchedulingDecision(
+      HostMap& hostMap,
+      const InFlightReqs& inFlightReqs,
+      std::shared_ptr<faabric::BatchExecuteRequest> req) override;
+
+  private:
+    bool isFirstDecisionBetter(
+      std::shared_ptr<SchedulingDecision> decisionA,
+      std::shared_ptr<SchedulingDecision> decisionB) override;
+
+    bool isFirstDecisionBetter(HostMap& hostMap,
+                               std::shared_ptr<SchedulingDecision> decisionA,
+                               std::shared_ptr<SchedulingDecision> decisionB);
+
+    std::vector<Host> getSortedHosts(
+      HostMap& hostMap,
+      const InFlightReqs& inFlightReqs,
+      std::shared_ptr<faabric::BatchExecuteRequest> req,
+      const DecisionType& decisionType) override;
+};
+}
diff --git a/include/faabric/planner/Planner.h b/include/faabric/planner/Planner.h
@@ -33,6 +33,8 @@ class Planner
 
     void printConfig() const;
 
+    void setPolicy(const std::string& newPolicy);
+
     // ----------
     // Util public API
     // ----------

diff --git a/src/batch-scheduler/BatchScheduler.cpp b/src/batch-scheduler/BatchScheduler.cpp
@@ -1,5 +1,6 @@
 #include <faabric/batch-scheduler/BatchScheduler.h>
 #include <faabric/batch-scheduler/BinPackScheduler.h>
+#include <faabric/batch-scheduler/CompactScheduler.h>
 #include <faabric/util/config.h>
 #include <faabric/util/logging.h>
 
@@ -20,6 +21,8 @@ std::shared_ptr<BatchScheduler> getBatchScheduler()
 
     if (mode == "bin-pack") {
         batchScheduler = std::make_shared<BinPackScheduler>();
+    } else if (mode == "compact") {
+        batchScheduler = std::make_shared<CompactScheduler>();
     } else {
         SPDLOG_ERROR("Unrecognised batch scheduler mode: {}", mode);
         throw std::runtime_error("Unrecognised batch scheduler mode");
@@ -33,6 +36,15 @@ void resetBatchScheduler()
     batchScheduler = nullptr;
 }
 
+void resetBatchScheduler(const std::string& newMode)
+{
+    resetBatchScheduler();
+
+    faabric::util::getSystemConfig().batchSchedulerMode = newMode;
+
+    getBatchScheduler();
+}
+
 DecisionType BatchScheduler::getDecisionType(
   const InFlightReqs& inFlightReqs,
   std::shared_ptr<faabric::BatchExecuteRequest> req)

diff --git a/src/batch-scheduler/CMakeLists.txt b/src/batch-scheduler/CMakeLists.txt
@@ -6,6 +6,7 @@ faabric_lib(scheduling_util
 faabric_lib(batch_scheduler
     BatchScheduler.cpp
     BinPackScheduler.cpp
+    CompactScheduler.cpp
 )
 
 target_link_libraries(batch_scheduler PRIVATE