Search in sources :

Example 1 with AvailableResources

use of org.bf2.performance.TestUtils.AvailableResources in project kas-fleetshard by bf2fc6cc711aee1a0c2a.

the class InstanceProfiler method setup.

private void setup() throws Exception {
    readResults();
    if (profilingResult.name == null) {
        profilingResult.name = "profile-" + Environment.DATE_FORMAT.format(LocalDateTime.now());
    }
    logDir = new File("target", profilingResult.name);
    Files.createDirectories(logDir.toPath());
    kafkaCluster = KubeClusterResource.connectToKubeCluster(PerformanceEnvironment.KAFKA_KUBECONFIG);
    profilingResult.kafkaNodeType = kafkaCluster.getWorkerNodes().get(0).getMetadata().getLabels().get("node.kubernetes.io/instance-type");
    kafkaProvisioner = ManagedKafkaProvisioner.create(kafkaCluster);
    kafkaProvisioner.setup();
    omb = new OMB(KubeClusterResource.connectToKubeCluster(PerformanceEnvironment.OMB_KUBECONFIG));
    omb.install(kafkaProvisioner.getTlsConfig());
    // TODO: if there is an existing result, make sure it's the same test setup
    profilingResult.ombNodeType = omb.getOmbCluster().getWorkerNodes().get(0).getMetadata().getLabels().get("node.kubernetes.io/instance-type");
    profilingResult.ombWorkerNodes = omb.getOmbCluster().getWorkerNodes().size();
    AvailableResources resources = getMinAvailableResources(omb.getOmbCluster().getWorkerNodes().stream());
    // use all available resources on the worker nodes with 2 workers per node
    // if (resources.memoryBytes > 16*ONE_GB || resources.memoryBytes < 8*ONE_GB) {
    // throw new IllegalStateException("Client instance types are expected to have 16 GB");
    // }
    // assume instead resources that will fit on 2xlarge or xlarge
    resources.cpuMillis = Math.min(6400, resources.cpuMillis);
    resources.memoryBytes = Math.min(12 * ONE_GB, resources.memoryBytes);
    omb.setWorkerCpu(Quantity.parse(resources.cpuMillis / 2 + "m"));
    omb.setWorkerContainerMemory(Quantity.parse(String.valueOf(resources.memoryBytes / 2)));
    profilingResult.ombWorkerCpu = omb.getWorkerCpu();
    profilingResult.ombWorkerMemory = omb.getWorkerContainerMemory();
    LOGGER.info("OMB Workers will use {} cpu and {} memory requests", omb.getWorkerCpu(), omb.getWorkerContainerMemory());
    if (profilingResult.completedStep == null) {
        installedProvisioner = true;
        kafkaProvisioner.install();
        writeResults(Step.SETUP);
    }
}
Also used : AvailableResources(org.bf2.performance.TestUtils.AvailableResources) File(java.io.File)

Example 2 with AvailableResources

use of org.bf2.performance.TestUtils.AvailableResources in project kas-fleetshard by bf2fc6cc711aee1a0c2a.

the class InstanceProfiler method sizeInstance.

protected void sizeInstance() throws Exception {
    Stream<Node> workerNodes = kafkaCluster.getWorkerNodes().stream();
    if (!collocateBrokerWithZookeeper) {
        kafkaProvisioner.validateClusterForBrokers(numberOfBrokers, false, workerNodes);
        workerNodes = kafkaCluster.getWorkerNodes().stream().filter(n -> n.getSpec().getTaints().stream().anyMatch(t -> t.getKey().equals(ManagedKafkaProvisioner.KAFKA_BROKER_TAINT_KEY)));
    }
    // note these number seem to change per release - 4.9 reports a different allocatable, than 4.8
    AvailableResources resources = getMinAvailableResources(workerNodes);
    long cpuMillis = resources.cpuMillis;
    long memoryBytes = resources.memoryBytes;
    Properties p = new Properties();
    try (InputStream is = InstanceProfiler.class.getResourceAsStream("/application.properties")) {
        p.load(is);
    }
    KafkaInstanceConfiguration defaults = Serialization.jsonMapper().convertValue(p, KafkaInstanceConfiguration.class);
    // when locating with ZK, then reduce the available resources accordingly
    if (collocateBrokerWithZookeeper) {
        // earlier code making a guess at the page cache size has been removed - until we can more reliably detect it's effect
        // there's no point in making a trade-off between extra container memory and JVM memory
        // TODO: could choose a memory size where we can fit even multiples of zookeepers
        long zookeeperBytes = Quantity.getAmountInBytes(Quantity.parse(defaults.getZookeeper().getContainerMemory())).longValue();
        long zookeeperCpu = Quantity.getAmountInBytes(Quantity.parse(defaults.getZookeeper().getContainerCpu())).movePointRight(3).longValue();
        List<Long> additionalPodCpu = new ArrayList<>();
        List<Long> additionalPodMemory = new ArrayList<>();
        additionalPodCpu.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getCanary().getContainerCpu())).movePointRight(3).longValue());
        additionalPodMemory.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getCanary().getContainerMemory())).longValue());
        additionalPodCpu.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getAdminserver().getContainerCpu())).movePointRight(3).longValue());
        additionalPodMemory.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getAdminserver().getContainerMemory())).longValue());
        additionalPodCpu.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getExporter().getContainerCpu())).movePointRight(3).longValue());
        additionalPodMemory.add(Quantity.getAmountInBytes(Quantity.parse(defaults.getExporter().getContainerMemory())).longValue());
        LOGGER.info("Total overhead of additional pods {} memory, {} cpu", additionalPodMemory.stream().collect(Collectors.summingLong(Long::valueOf)), additionalPodCpu.stream().collect(Collectors.summingLong(Long::valueOf)));
        // actual needs ~ 800Mi and 1075m/1575m cpu over 3 nodes, but worst case is over two. amountNeeded will
        // estimate that in a more targeted way - but still simplified
        memoryBytes = resources.memoryBytes - density * (zookeeperBytes + amountNeeded(additionalPodMemory));
        cpuMillis = resources.cpuMillis - density * (zookeeperCpu + amountNeeded(additionalPodCpu));
    // TODO account for possible ingress replica collocation
    }
    // and if there are eventually pods that need to be collocated, and we don't want to adjust the resources downward
    if (density == 1) {
        memoryBytes -= 2 * ONE_GB;
        cpuMillis -= 500;
    } else {
        // we can assume a much tighter resource utilization for density 2 - it can fluctuate between releases
        // or may require adjustments as other pods are added or pod resource adjustments are made
        memoryBytes -= 1 * ONE_GB;
        cpuMillis -= 200;
    }
    memoryBytes = memoryBytes / density;
    cpuMillis = cpuMillis / density;
    long maxVmBytes = Math.min(memoryBytes - getVMOverheadForContainer(memoryBytes), MAX_KAFKA_VM_SIZE);
    if (density > 1) {
        maxVmBytes -= 1 * ONE_GB;
    }
    if (!autoSize) {
        long defaultMemory = Quantity.getAmountInBytes(Quantity.parse(defaults.getKafka().getContainerMemory())).longValue();
        long defaultCpu = Quantity.getAmountInBytes(Quantity.parse(defaults.getKafka().getContainerCpu())).movePointRight(3).longValue();
        long defaultMaxVmBytes = Quantity.getAmountInBytes(Quantity.parse(defaults.getKafka().getJvmXms())).longValue();
        LOGGER.info("Calculated kafka sizing {} container memory, {} container cpu, and {} vm memory", memoryBytes, cpuMillis, maxVmBytes);
        memoryBytes = defaultMemory;
        cpuMillis = defaultCpu;
        maxVmBytes = defaultMaxVmBytes;
    }
    KafkaInstanceConfiguration toUse = new KafkaInstanceConfiguration();
    toUse.getKafka().setEnableQuota(false);
    AdopterProfile.openListenersAndAccess(toUse);
    toUse.getKafka().setContainerCpu(cpuMillis + "m");
    toUse.getKafka().setJvmXms(String.valueOf(maxVmBytes));
    toUse.getKafka().setContainerMemory(String.valueOf(memoryBytes));
    profilingResult.config = toUse;
    profilingResult.config.getKafka().setColocateWithZookeeper(collocateBrokerWithZookeeper);
    profilingResult.config.getKafka().setMaxConnections(Integer.MAX_VALUE);
    profilingResult.config.getKafka().setConnectionAttemptsPerSec(Integer.MAX_VALUE);
    profilingResult.config.getKafka().setMessageMaxBytes(11534336);
    profilingResult.config.getKafka().setStorageClass(storage.name().toLowerCase());
    profilingResult.config.getZookeeper().setVolumeSize(storage.zookeeperSize);
    // once we make the determination, create the instance
    // not used as quota is turned off
    profilingResult.capacity = kafkaProvisioner.defaultCapacity(40_000_000);
    profilingResult.capacity.setMaxDataRetentionSize(Quantity.parse((GIGS * numberOfBrokers / 3) + "Gi"));
    profilingResult.capacity.setMaxPartitions(defaults.getKafka().getPartitionCapacity() * numberOfBrokers / 3);
    Kafka kafka = profilingResult.config.getKafka();
    LOGGER.info("Running with kafka sizing {} container memory, {} container cpu, and {} vm memory", kafka.getContainerMemory(), kafka.getContainerCpu(), kafka.getJvmXms());
// if running on m5.4xlarge or greater and want to constrain resources like m5.2xlarge (fully dedicated)
// profilingResult.config.getKafka().setContainerMemory("29013426176");
// profilingResult.config.getKafka().setContainerCpu("6500m");
// to constrain resources like m5.xlarge (fully dedicated)
// profilingResult.config.getKafka().setContainerMemory("12453740544");
// profilingResult.config.getKafka().setContainerCpu("2500m");
}
Also used : Quantity(io.fabric8.kubernetes.api.model.Quantity) KubeClusterResource(org.bf2.performance.framework.KubeClusterResource) Arrays(java.util.Arrays) LocalDateTime(java.time.LocalDateTime) AvailableResources(org.bf2.performance.TestUtils.AvailableResources) HashMap(java.util.HashMap) KeyDistributorType(io.openmessaging.benchmark.utils.distributor.KeyDistributorType) ArrayList(java.util.ArrayList) Resource(io.fabric8.kubernetes.client.dsl.Resource) Workload(io.openmessaging.benchmark.Workload) Serialization(io.fabric8.kubernetes.client.utils.Serialization) Kafka(org.bf2.operator.operands.KafkaInstanceConfiguration.Kafka) Map(java.util.Map) BiConsumer(java.util.function.BiConsumer) Assertions.assertEquals(org.junit.jupiter.api.Assertions.assertEquals) Node(io.fabric8.kubernetes.api.model.Node) KubernetesClientException(io.fabric8.kubernetes.client.KubernetesClientException) Properties(java.util.Properties) Iterator(java.util.Iterator) Files(java.nio.file.Files) KafkaInstanceConfiguration(org.bf2.operator.operands.KafkaInstanceConfiguration) Environment(org.bf2.test.Environment) ManagedKafkaCapacity(org.bf2.operator.resources.v1alpha1.ManagedKafkaCapacity) IOException(java.io.IOException) Collectors(java.util.stream.Collectors) File(java.io.File) StandardCharsets(java.nio.charset.StandardCharsets) Consumer(java.util.function.Consumer) List(java.util.List) Logger(org.apache.logging.log4j.Logger) Stream(java.util.stream.Stream) TreeMap(java.util.TreeMap) Assertions.assertTrue(org.junit.jupiter.api.Assertions.assertTrue) TestResult(io.openmessaging.benchmark.TestResult) JsonInclude(com.fasterxml.jackson.annotation.JsonInclude) Comparator(java.util.Comparator) LogManager(org.apache.logging.log4j.LogManager) SuppressFBWarnings(org.bf2.common.SuppressFBWarnings) ManagedKafka(org.bf2.operator.resources.v1alpha1.ManagedKafka) InputStream(java.io.InputStream) InputStream(java.io.InputStream) Node(io.fabric8.kubernetes.api.model.Node) ArrayList(java.util.ArrayList) Kafka(org.bf2.operator.operands.KafkaInstanceConfiguration.Kafka) ManagedKafka(org.bf2.operator.resources.v1alpha1.ManagedKafka) KafkaInstanceConfiguration(org.bf2.operator.operands.KafkaInstanceConfiguration) AvailableResources(org.bf2.performance.TestUtils.AvailableResources) Properties(java.util.Properties)

Aggregations

File (java.io.File)2 AvailableResources (org.bf2.performance.TestUtils.AvailableResources)2 JsonInclude (com.fasterxml.jackson.annotation.JsonInclude)1 Node (io.fabric8.kubernetes.api.model.Node)1 Quantity (io.fabric8.kubernetes.api.model.Quantity)1 KubernetesClientException (io.fabric8.kubernetes.client.KubernetesClientException)1 Resource (io.fabric8.kubernetes.client.dsl.Resource)1 Serialization (io.fabric8.kubernetes.client.utils.Serialization)1 TestResult (io.openmessaging.benchmark.TestResult)1 Workload (io.openmessaging.benchmark.Workload)1 KeyDistributorType (io.openmessaging.benchmark.utils.distributor.KeyDistributorType)1 IOException (java.io.IOException)1 InputStream (java.io.InputStream)1 StandardCharsets (java.nio.charset.StandardCharsets)1 Files (java.nio.file.Files)1 LocalDateTime (java.time.LocalDateTime)1 ArrayList (java.util.ArrayList)1 Arrays (java.util.Arrays)1 Comparator (java.util.Comparator)1 HashMap (java.util.HashMap)1