Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

fix resuming subtasks #90

Draft
wants to merge 4 commits into
base: v0.10.5_patched
Choose a base branch
from
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
149 changes: 137 additions & 12 deletions digdag-core/src/main/java/io/digdag/core/workflow/TaskControl.java
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,9 @@
import java.util.Set;
import java.util.HashSet;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.stream.Collectors;

import com.google.common.base.Optional;
import com.google.common.base.Strings;
import com.google.common.collect.*;
Expand All @@ -15,11 +17,13 @@
import io.digdag.core.session.StoredTask;
import io.digdag.core.session.SessionStore;
import io.digdag.core.session.ArchivedTask;
import io.digdag.core.session.ImmutableResumingTask;
import io.digdag.core.session.ResumingTask;
import io.digdag.core.session.Task;
import io.digdag.core.session.TaskControlStore;
import io.digdag.core.session.TaskStateCode;
import io.digdag.core.session.TaskStateFlags;
import io.digdag.spi.TaskReport;
import io.digdag.spi.TaskResult;
import io.digdag.client.config.Config;

Expand Down Expand Up @@ -55,14 +59,12 @@ public TaskStateCode getState()

public static long addInitialTasksExceptingRootTask(
TaskControlStore store, long attemptId, long rootTaskId,
WorkflowTaskList tasks, List<ResumingTask> resumingTasks, Limits limits)
WorkflowTaskList tasks, List<ArchivedTask> archivedTasks, Limits limits)
throws TaskLimitExceededException
{
checkTaskLimit(store, attemptId, tasks, limits);
long taskId = addTasks(store, attemptId, rootTaskId,
tasks, ImmutableList.of(),
false, true, true,
resumingTasks);
long taskId = addInitialTasks(store, attemptId, rootTaskId, tasks, archivedTasks);
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a463168 (#90) のaddInitialTasksに変更。

List<ResumingTask> resumingTasks = archivedTasks.stream().map(ResumingTask::of).collect(Collectors.toList());
addResumingTasks(store, attemptId, resumingTasks);
return taskId;
}
Expand Down Expand Up @@ -185,6 +187,133 @@ private static long addTasks(TaskControlStore store,
return rootTaskId;
}

private static long addInitialTasks(TaskControlStore store, long attemptId, long rootTaskId,
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

既存のTaskControl#addTasks は別の箇所からも参照されているため、ただ単に引数のresumingTasksをArchivedTasksに変更することはできなかった。
そのため、新たにaddInitialTasksを実装した。
参照: INSERT時のparameterの不足

WorkflowTaskList workflowTasks, List<ArchivedTask> archivedTasks)
{
List<Long> indexToId = new ArrayList<>();
indexToId.add(rootTaskId);

workflowTasks.stream()
.skip(1) // skip the root task
.forEach(workflowTask -> {
long id;
long parentId = workflowTask.getParentIndex()
.transform(index -> indexToId.get(index))
.or(rootTaskId);

ArchivedTask archivedTask = archivedTasks.stream()
.filter(t -> t.getFullName().equals(workflowTask.getFullName()))
.findFirst()
.orElse(null);

if (archivedTask == null) {
Task task = Task.taskBuilder()
.parentId(Optional.of(parentId))
.fullName(workflowTask.getFullName())
.config(TaskConfig.validate(workflowTask.getConfig()))
.taskType(workflowTask.getTaskType())
.state(TaskStateCode.BLOCKED)
.stateFlags(TaskStateFlags.empty().withInitialTask())
.build();

id = store.addSubtask(attemptId, task);
} else {
TaskStateCode state;
switch(archivedTask.getState()) {
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

元々のTaskControl#addTasksでは、resumingTasksの有無によってBLOCKED or SUCCESSで分岐していたが、6aa7cb9 によって「resumeで取得するtaskがSUCCESSのみ」という前提が変わったため、stateによる分岐が必要になった。

case SUCCESS:
state = TaskStateCode.SUCCESS;
break;
case ERROR:
case CANCELED:
state = TaskStateCode.BLOCKED;
break;
default:
state = TaskStateCode.PLANNED;
break;
}

ResumingTask resumingTask = ImmutableResumingTask.builder()
.sourceTaskId(archivedTask.getId())
.fullName(archivedTask.getFullName())
.config(TaskConfig.validate(workflowTask.getConfig()))
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ここのconfigのみarchivedTaskではなく、workflowTaskのconfigをINSERTしている。
これはtaskの定義が変更された場合に新しいrevisionのconfigである必要があるため。

具体的には、この修正がないと以下のIntegration Testでのretryで落ちる。
RetryIT.java

https://github.com/treasure-data/digdag/blob/v0.10.5.1/digdag-tests/src/test/resources/acceptance/retry/retry-1.dig

+step1:
  sh>: touch ${outdir}/1-1.out
+step2:
  +a:
    sh>: touch ${outdir}/1-2a.out
  +b:
    fail>: step2b fail

https://github.com/treasure-data/digdag/blob/v0.10.5.1/digdag-tests/src/test/resources/acceptance/retry/retry-2.dig

+step1:
  sh>: touch ${outdir}/2-1.out
+step2:
  +a:
    sh>: touch ${outdir}/2-2a.out
  +b:
    sh>: touch ${outdir}/2-2b.out

.updatedAt(archivedTask.getUpdatedAt())
.subtaskConfig(archivedTask.getSubtaskConfig())
.exportParams(archivedTask.getExportParams())
.resetStoreParams(archivedTask.getResetStoreParams())
.storeParams(archivedTask.getStoreParams())
.report(archivedTask.getReport().or(TaskReport.empty()))
.error(archivedTask.getError())
.build();

id = store.addResumedSubtask(attemptId,
parentId,
workflowTask.getTaskType(),
state,
TaskStateFlags.empty().withInitialTask(),
resumingTask);
}

indexToId.add(id);

if (!workflowTask.getUpstreamIndexes().isEmpty()) {
store.addDependencies(
id,
workflowTask.getUpstreamIndexes()
.stream()
.map(index -> indexToId.get(index))
.collect(Collectors.toList())
);
}
});

Map<String, Long> taskNameAndIds = workflowTasks.stream()
.collect(Collectors.toMap(WorkflowTask::getFullName, task -> indexToId.get(workflowTasks.indexOf(task))));

archivedTasks.stream()
.filter(archivedTask -> !taskNameAndIds.keySet().contains(archivedTask.getFullName())
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

すでにINSERT済みのtaskは除外する。

&& archivedTask.getFullName().contains("^sub"))
Copy link
Author

@snagasawa snagasawa Feb 27, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

単にtask名に「^」を含む場合だと、「^failure-alert」や「^error」などの除外すべきsubtaskも含まれるため、狭義のsubtaskに限定する必要がある。

.sorted(Comparator.comparingInt((t) -> (int) t.getId()))
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

後続のtask_dependenciesテーブルへのINESRTのために、upstreamのtaskを先にtasksテーブルへのINSERTしてIDを採番する必要があるため、あらかじめretry failed前のattemptのTask IDでソートする。

.forEach(archivedSubtask -> {
String parentTaskName = archivedSubtask.getFullName().replaceAll("(\\^sub|\\+)[^\\^+]*$", "");
Long parentId = taskNameAndIds.get(parentTaskName);

TaskStateCode state;
switch(archivedSubtask.getState()) {
case SUCCESS:
state = TaskStateCode.SUCCESS;
break;
case ERROR:
case CANCELED:
state = TaskStateCode.BLOCKED;
break;
default:
state = TaskStateCode.PLANNED;
break;
}

Long id = store.addResumedSubtask(attemptId,
parentId,
archivedSubtask.getTaskType(),
state,
TaskStateFlags.empty().withInitialTask(),
ResumingTask.of(archivedSubtask));

taskNameAndIds.put(archivedSubtask.getFullName(), id);

archivedTasks.stream()
Copy link
Author

@snagasawa snagasawa Feb 27, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

TaskControl#addTasksと同じようにtask_dependenciesテーブルへのINSERTが必要なため、task名からupstreamのIDを取得しINSERTする。

.filter(t -> t.getId() == archivedSubtask.getId() - 1)
.findFirst()
.ifPresent((upstreamArchivedTask) -> {
List<Long> upstreamIds = new ArrayList<>();
Long upstreamId = taskNameAndIds.get(upstreamArchivedTask.getFullName());
upstreamIds.add(upstreamId);
store.addDependencies(id, upstreamIds);
});;
});

return rootTaskId;
}

private static void addResumingTasks(TaskControlStore store, long attemptId, List<ResumingTask> resumingTasks)
{
// store only dynamically-generated tasks
Expand All @@ -207,28 +336,24 @@ private List<ResumingTask> collectResumingTasks(long attemptId, WorkflowTaskList
return store.getResumingTasksByNamePrefix(attemptId, commonPrefix);
}

static List<ResumingTask> buildResumingTaskMap(SessionStore store, long attemptId, List<Long> resumingTaskIds)
static List<ArchivedTask> buildResumingTaskMap(SessionStore store, long attemptId, List<Long> resumingTaskIds)
throws ResourceNotFoundException
{
Set<Long> idSet = new HashSet<>(resumingTaskIds);
List<ResumingTask> resumingTasks = store
List<ArchivedTask> archivedTasks = store
.getTasksOfAttempt(attemptId)
.stream()
.filter(archived -> {
if (idSet.remove(archived.getId())) {
if (archived.getState() != TaskStateCode.SUCCESS) {
throw new IllegalResumeException("Resuming non-successful tasks is not allowed: task_id=" + archived.getId());
}
return true;
}
return false;
})
.map(archived -> ResumingTask.of(archived))
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ResumingTaskに変換せず、ArchivedTaskのまま返すように。
参照: INSERT時のparameterの不足

.collect(Collectors.toList());
if (!idSet.isEmpty()) {
throw new ResourceNotFoundException("Resuming tasks are not the members of resuming attempt: id list=" + idSet);
}
return resumingTasks;
return archivedTasks;
}

////
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
import io.digdag.core.repository.StoredRevision;
import io.digdag.core.repository.WorkflowDefinition;
import io.digdag.core.session.ResumingTask;
import io.digdag.core.session.ArchivedTask;
import io.digdag.core.session.ParameterUpdate;
import io.digdag.core.session.Session;
import io.digdag.core.session.SessionAttempt;
Expand All @@ -43,6 +44,8 @@
import io.digdag.spi.TaskConflictException;
import io.digdag.spi.TaskNotFoundException;
import io.digdag.spi.metrics.DigdagMetrics;
import io.digdag.spi.metrics.DigdagMetrics.Category;

import static io.digdag.spi.metrics.DigdagMetrics.Category;
import io.digdag.util.RetryControl;
import org.slf4j.Logger;
Expand Down Expand Up @@ -269,21 +272,15 @@ public StoredSessionAttemptWithSession submitTasks(int siteId, AttemptRequest ar

TaskConfig.validateAttempt(attempt);

List<ResumingTask> resumingTasks;
List<ArchivedTask> archivedTasks;
if (ar.getResumingAttemptId().isPresent()) {
WorkflowTask root = tasks.get(0);
resumingTasks = TaskControl.buildResumingTaskMap(
archivedTasks = TaskControl.buildResumingTaskMap(
sm.getSessionStore(siteId),
ar.getResumingAttemptId().get(),
ar.getResumingTasks());
for (ResumingTask resumingTask : resumingTasks) {
if (resumingTask.getFullName().equals(root.getFullName())) {
throw new IllegalResumeException("Resuming root task is not allowed");
}
}
}
else {
resumingTasks = ImmutableList.of();
archivedTasks = ImmutableList.of();
}

StoredSessionAttemptWithSession stored;
Expand Down Expand Up @@ -313,7 +310,7 @@ public StoredSessionAttemptWithSession submitTasks(int siteId, AttemptRequest ar
StoredSessionAttemptWithSession.of(siteId, storedSession, storedAttempt);

try {
storeTasks(store, storedAttemptWithSession, tasks, resumingTasks, ar.getSessionMonitors());
storeTasks(store, storedAttemptWithSession, tasks, archivedTasks, ar.getSessionMonitors());
}
catch (TaskLimitExceededException ex) {
throw new WorkflowTaskLimitExceededException(ex);
Expand Down Expand Up @@ -348,21 +345,21 @@ public void storeTasks(
SessionControlStore store,
StoredSessionAttemptWithSession storedAttempt,
WorkflowDefinition def,
List<ResumingTask> resumingTasks,
List<ArchivedTask> archivedTasks,
List<SessionMonitor> sessionMonitors)
throws TaskLimitExceededException
{
Workflow workflow = compiler.compile(def.getName(), def.getConfig());
WorkflowTaskList tasks = workflow.getTasks();

storeTasks(store, storedAttempt, tasks, resumingTasks, sessionMonitors);
storeTasks(store, storedAttempt, tasks, archivedTasks, sessionMonitors);
}

public void storeTasks(
SessionControlStore store,
StoredSessionAttemptWithSession storedAttempt,
WorkflowTaskList tasks,
List<ResumingTask> resumingTasks,
List<ArchivedTask> archivedTasks,
List<SessionMonitor> sessionMonitors)
throws TaskLimitExceededException
{
Expand All @@ -386,7 +383,7 @@ public void storeTasks(
store.insertRootTask(storedAttempt.getId(), rootTask, (taskStore, storedTaskId) -> {
try {
TaskControl.addInitialTasksExceptingRootTask(taskStore, storedAttempt.getId(),
storedTaskId, tasks, resumingTasks, limits);
storedTaskId, tasks, archivedTasks, limits);
}
catch (TaskLimitExceededException ex) {
throw new WorkflowTaskLimitExceededException(ex);
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@

import java.util.List;
import java.util.Set;
import java.util.Arrays;
import java.util.HashSet;
import java.util.stream.Collectors;
import javax.ws.rs.Consumes;
Expand Down Expand Up @@ -298,21 +299,24 @@ private List<Long> collectResumingTasks(RestSessionAttemptRequest.Resume resume)

private List<Long> collectResumingTasksForResumeFailedMode(long attemptId)
{
TaskStateCode[] statusArr = {TaskStateCode.SUCCESS, TaskStateCode.GROUP_ERROR, TaskStateCode.ERROR, TaskStateCode.CANCELED};
List<TaskStateCode> statuses = Arrays.asList(statusArr);

List<ArchivedTask> tasks = sm
.getSessionStore(getSiteId())
.getTasksOfAttempt(attemptId);

List<Long> successTasks = tasks.stream()
.filter(task -> task.getState() == TaskStateCode.SUCCESS)
List<Long> ids = tasks.stream()
.filter(t-> statuses.contains(t.getState()))
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

resumeするtaskをSUCCESSのみから変更。
参照: SUCCESSのみが取得対象

.map(task -> {
if (!task.getParentId().isPresent()) {
if (!task.getParentId().isPresent() && task.getState() == TaskStateCode.SUCCESS) {
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

すでに成功したattemptのresumeは禁止されているため、L310のfilterの変更に合わせて修正。
!task.getParentId().isPresent() は「parentIdがないtask = rootTask」のこと。
(rootTaskがSUCCESS = attemptもSUCCESS)

throw new IllegalArgumentException("Resuming successfully completed attempts is not supported");
}
return task.getId();
})
.collect(Collectors.toList());

return ImmutableList.copyOf(successTasks);
return ImmutableList.copyOf(ids);
}

private List<Long> collectResumingTasksForResumeFromMode(long attemptId, String fromTaskPattern)
Expand Down
Loading