[qob] cancel stage if any partitions fail.

hail-is · Oct 25, 2024 · a0b5577 · a0b5577
1 parent 71b00f3
commit a0b5577
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 6 deletions.
diff --git a/hail/src/main/scala/is/hail/backend/service/ServiceBackend.scala b/hail/src/main/scala/is/hail/backend/service/ServiceBackend.scala
@@ -1,6 +1,6 @@
 package is.hail.backend.service
 
-import is.hail.{CancellingExecutorService, HailContext, HailFeatureFlags}
+import is.hail.{HailContext, HailFeatureFlags}
 import is.hail.annotations._
 import is.hail.asm4s._
 import is.hail.backend._
@@ -202,6 +202,7 @@ class ServiceBackend(
           batch_id = batchConfig.batchId,
           absolute_parent_id = batchConfig.jobGroupId,
           token = token,
+          cancel_after_n_failures = Some(1),
           attributes = Map("name" -> stageIdentifier),
           jobs = jobs,
         )
@@ -280,7 +281,13 @@ class ServiceBackend(
 
     log.info(s"parallelizeAndComputeWithIndex: $token: reading results")
     val startTime = System.nanoTime()
-    val r @ (error, results) = runAllKeepFirstError(new CancellingExecutorService(executor)) {
+    val r @ (error, results) = runAll[Option, Array[Byte]](executor) {
+      /* A missing file means the job was cancelled because another job failed. Assumes that if any
+       * job was cancelled, then at least one job failed. We want to ignore the missing file
+       * exceptions and return one of the actual failure exceptions. */
+      case (opt, _: FileNotFoundException) => opt
+      case (opt, e) => opt.orElse(Some(e))
+    }(None) {
       (partIdxs, parts.indices).zipped.map { (partIdx, jobIndex) =>
         (() => readResult(root, jobIndex), partIdx)
       }

diff --git a/hail/src/main/scala/is/hail/services/BatchClient.scala b/hail/src/main/scala/is/hail/services/BatchClient.scala
@@ -14,7 +14,9 @@ import java.nio.file.Path
 
 import org.apache.http.entity.ByteArrayEntity
 import org.apache.http.entity.ContentType.APPLICATION_JSON
-import org.json4s.{CustomSerializer, DefaultFormats, Extraction, Formats, JInt, JObject, JString}
+import org.json4s.{
+  CustomSerializer, DefaultFormats, Extraction, Formats, JInt, JNull, JObject, JString,
+}
 import org.json4s.JsonAST.{JArray, JBool}
 import org.json4s.jackson.JsonMethods
 
@@ -29,6 +31,7 @@ case class JobGroupRequest(
   batch_id: Int,
   absolute_parent_id: Int,
   token: String,
+  cancel_after_n_failures: Option[Int] = None,
   attributes: Map[String, String] = Map.empty,
   jobs: IndexedSeq[JobRequest] = FastSeq(),
 )
@@ -52,9 +55,9 @@ case class JarUrl(url: String) extends JarSpec
 
 case class JobResources(
   preemptible: Boolean,
-  cpu: Option[String],
-  memory: Option[String],
-  storage: Option[String],
+  cpu: Option[String] = None,
+  memory: Option[String] = None,
+  storage: Option[String] = None,
 )
 
 case class CloudfuseConfig(
@@ -252,6 +255,9 @@ case class BatchClient private (req: Requester) extends Logging with AutoCloseab
         JObject(
           "job_group_id" -> JInt(1), // job group id relative to the update
           "absolute_parent_id" -> JInt(jobGroup.absolute_parent_id),
+          "cancel_after_n_failures" -> jobGroup.cancel_after_n_failures.map(JInt(_)).getOrElse(
+            JNull
+          ),
           "attributes" -> Extraction.decompose(jobGroup.attributes),
         )
       )),

diff --git a/hail/src/test/scala/is/hail/services/BatchClientSuite.scala b/hail/src/test/scala/is/hail/services/BatchClientSuite.scala
@@ -2,6 +2,7 @@ package is.hail.services
 
 import is.hail.HAIL_REVISION
 import is.hail.backend.service.Main
+import is.hail.services.JobGroupStates.Failure
 import is.hail.utils._
 
 import java.lang.reflect.Method
@@ -46,6 +47,38 @@ class BatchClientSuite extends TestNGSuite {
   def closeClient(): Unit =
     client.close()
 
+  @Test
+  def testCancelAfterNFailures(): Unit = {
+    val jobGroupId = client.newJobGroup(
+      req = JobGroupRequest(
+        batch_id = batchId,
+        absolute_parent_id = parentJobGroupId,
+        cancel_after_n_failures = Some(1),
+        token = tokenUrlSafe,
+        jobs = FastSeq(
+          JobRequest(
+            always_run = false,
+            process = BashJob(
+              image = "ubuntu:22.04",
+              command = Array("/bin/bash", "-c", "sleep 1d"),
+            ),
+            resources = Some(JobResources(preemptible = true)),
+          ),
+          JobRequest(
+            always_run = false,
+            process = BashJob(
+              image = "ubuntu:22.04",
+              command = Array("/bin/bash", "-c", "exit 1"),
+            ),
+          ),
+        ),
+      )
+    )
+    val result = client.waitForJobGroup(batchId, jobGroupId)
+    assert(result.state == Failure)
+    assert(result.n_cancelled == 1)
+  }
+
   @Test
   def testNewJobGroup(): Unit =
     // The query driver submits a job group per stage with one job per partition