mirror of
https://github.com/wassname/ray.git
synced 2026-06-28 04:07:01 +08:00
[Core] Multi-tenancy: Job isolation & implement per job config (except for env variables) (#9500)
This commit is contained in:
@@ -11,6 +11,7 @@ import io.ray.runtime.gcs.GcsClient;
|
||||
import io.ray.runtime.gcs.GcsClientOptions;
|
||||
import io.ray.runtime.gcs.RedisClient;
|
||||
import io.ray.runtime.generated.Common.WorkerType;
|
||||
import io.ray.runtime.generated.Gcs.JobConfig;
|
||||
import io.ray.runtime.object.NativeObjectStore;
|
||||
import io.ray.runtime.runner.RunManager;
|
||||
import io.ray.runtime.task.NativeTaskExecutor;
|
||||
@@ -106,6 +107,17 @@ public final class RayNativeRuntime extends AbstractRayRuntime {
|
||||
}
|
||||
int numWorkersPerProcess =
|
||||
rayConfig.workerMode == WorkerType.DRIVER ? 1 : rayConfig.numWorkersPerProcess;
|
||||
|
||||
byte[] serializedJobConfig = null;
|
||||
if (rayConfig.workerMode == WorkerType.DRIVER) {
|
||||
JobConfig.Builder jobConfigBuilder =
|
||||
JobConfig.newBuilder()
|
||||
.setNumJavaWorkersPerProcess(rayConfig.numWorkersPerProcess)
|
||||
.addAllJvmOptions(rayConfig.jvmOptionsForJavaWorker)
|
||||
.putAllWorkerEnv(rayConfig.workerEnv);
|
||||
serializedJobConfig = jobConfigBuilder.build().toByteArray();
|
||||
}
|
||||
|
||||
// TODO(qwang): Get object_store_socket_name and raylet_socket_name from Redis.
|
||||
nativeInitialize(rayConfig.workerMode.getNumber(),
|
||||
rayConfig.nodeIp, rayConfig.getNodeManagerPort(),
|
||||
@@ -113,7 +125,7 @@ public final class RayNativeRuntime extends AbstractRayRuntime {
|
||||
rayConfig.objectStoreSocketName, rayConfig.rayletSocketName,
|
||||
(rayConfig.workerMode == WorkerType.DRIVER ? rayConfig.getJobId() : JobId.NIL).getBytes(),
|
||||
new GcsClientOptions(rayConfig), numWorkersPerProcess,
|
||||
rayConfig.logDir, rayConfig.rayletConfigParameters);
|
||||
rayConfig.logDir, rayConfig.rayletConfigParameters, serializedJobConfig);
|
||||
|
||||
taskExecutor = new NativeTaskExecutor(this);
|
||||
workerContext = new NativeWorkerContext();
|
||||
@@ -201,7 +213,7 @@ public final class RayNativeRuntime extends AbstractRayRuntime {
|
||||
int workerMode, String ndoeIpAddress,
|
||||
int nodeManagerPort, String driverName, String storeSocket, String rayletSocket,
|
||||
byte[] jobId, GcsClientOptions gcsClientOptions, int numWorkersPerProcess,
|
||||
String logDir, Map<String, String> rayletConfigParameters);
|
||||
String logDir, Map<String, String> rayletConfigParameters, byte[] serializedJobConfig);
|
||||
|
||||
private static native void nativeRunTaskExecutor(TaskExecutor taskExecutor);
|
||||
|
||||
|
||||
@@ -3,6 +3,7 @@ package io.ray.runtime.config;
|
||||
import com.google.common.base.Preconditions;
|
||||
import com.google.common.base.Strings;
|
||||
import com.google.common.collect.ImmutableList;
|
||||
import com.google.common.collect.ImmutableMap;
|
||||
import com.typesafe.config.Config;
|
||||
import com.typesafe.config.ConfigException;
|
||||
import com.typesafe.config.ConfigFactory;
|
||||
@@ -29,6 +30,8 @@ public class RayConfig {
|
||||
public static final String DEFAULT_CONFIG_FILE = "ray.default.conf";
|
||||
public static final String CUSTOM_CONFIG_FILE = "ray.conf";
|
||||
|
||||
private static int DEFAULT_NUM_JAVA_WORKER_PER_PROCESS = 10;
|
||||
|
||||
private static final Random RANDOM = new Random();
|
||||
|
||||
private static final DateTimeFormatter DATE_TIME_FORMATTER =
|
||||
@@ -90,6 +93,9 @@ public class RayConfig {
|
||||
|
||||
public final int numWorkersPerProcess;
|
||||
|
||||
public final List<String> jvmOptionsForJavaWorker;
|
||||
public final Map<String, String> workerEnv;
|
||||
|
||||
private void validate() {
|
||||
if (workerMode == WorkerType.WORKER) {
|
||||
Preconditions.checkArgument(redisAddress != null,
|
||||
@@ -141,6 +147,17 @@ public class RayConfig {
|
||||
this.jobId = JobId.NIL;
|
||||
}
|
||||
|
||||
// jvm options for java workers of this job.
|
||||
jvmOptionsForJavaWorker = config.getStringList("ray.job.jvm-options");
|
||||
|
||||
ImmutableMap.Builder<String, String> workerEnvBuilder = ImmutableMap.builder();
|
||||
Config workerEnvConfig = config.getConfig("ray.job.worker-env");
|
||||
if (workerEnvConfig != null) {
|
||||
for (Map.Entry<String, ConfigValue> entry : workerEnvConfig.entrySet()) {
|
||||
workerEnvBuilder.put(entry.getKey(), workerEnvConfig.getString(entry.getKey()));
|
||||
}
|
||||
}
|
||||
workerEnv = workerEnvBuilder.build();
|
||||
updateSessionDir();
|
||||
// Object store configurations.
|
||||
objectStoreSize = config.getBytes("ray.object-store.size");
|
||||
@@ -206,7 +223,22 @@ public class RayConfig {
|
||||
jobResourcePath = null;
|
||||
}
|
||||
|
||||
numWorkersPerProcess = config.getInt("ray.raylet.config.num_workers_per_process_java");
|
||||
boolean enableMultiTenancy = false;
|
||||
if (config.hasPath("ray.raylet.config.enable_multi_tenancy")) {
|
||||
enableMultiTenancy =
|
||||
Boolean.valueOf(config.getString("ray.raylet.config.enable_multi_tenancy"));
|
||||
}
|
||||
|
||||
if (!enableMultiTenancy) {
|
||||
numWorkersPerProcess = config.getInt("ray.raylet.config.num_workers_per_process_java");
|
||||
} else {
|
||||
final int localNumWorkersPerProcess = config.getInt("ray.job.num-java-workers-per-process");
|
||||
if (localNumWorkersPerProcess <= 0) {
|
||||
numWorkersPerProcess = DEFAULT_NUM_JAVA_WORKER_PER_PROCESS;
|
||||
} else {
|
||||
numWorkersPerProcess = localNumWorkersPerProcess;
|
||||
}
|
||||
}
|
||||
|
||||
// Validate config.
|
||||
validate();
|
||||
|
||||
@@ -29,6 +29,15 @@ ray {
|
||||
// executing tasks from different jobs. E.g. if it's set to '/tm/job_resources',
|
||||
// the path for job 123 will be '/tmp/job_resources/123'.
|
||||
resource-path: ""
|
||||
/// The number of java worker per worker process.
|
||||
num-java-workers-per-process: 10
|
||||
/// The jvm options for java workers of the job.
|
||||
jvm-options: []
|
||||
// Environment variables to be set on worker processes.
|
||||
worker-env {
|
||||
// key1 : "value1"
|
||||
// key2 : "value2"
|
||||
}
|
||||
}
|
||||
|
||||
// Configurations about logging.
|
||||
|
||||
Reference in New Issue
Block a user