End-to-end Datastore to SQL pipeline (#707)

* End-to-end Datastore to SQL pipeline

Defined InitSqlPipeline that performs end-to-end migration from
a Datastore backup to a SQL database.

Also fixed/refined multiple tests related to this migration.
This commit is contained in:
Weimin Yu
2020-07-24 09:57:43 -04:00
committed by GitHub
parent 91b7d92cf8
commit 6591e0672a
26 changed files with 1925 additions and 114 deletions
@@ -22,21 +22,35 @@ import java.lang.reflect.Method;
import java.lang.reflect.Proxy;
/**
* Sets up a placeholder {@link Environment} on a non-AppEngine platform so that Datastore Entities
* can be converted from/to Objectify entities. See {@code DatastoreEntityExtension} in test source
* for more information.
* Sets up a fake {@link Environment} so that the following operations can be performed without the
* Datastore service:
*
* <ul>
* <li>Create Objectify {@code Keys}.
* <li>Instantiate Objectify objects.
* <li>Convert Datastore {@code Entities} to their corresponding Objectify objects.
* </ul>
*
* <p>User has the option to specify their desired {@code appId} string, which forms part of an
* Objectify {@code Key} and is included in the equality check. This feature makes it easy to
* compare a migrated object in SQL with the original in Objectify.
*
* <p>Note that conversion from Objectify objects to Datastore {@code Entities} still requires the
* Datastore service.
*/
public class AppEngineEnvironment implements Closeable {
private static final Environment PLACEHOLDER_ENV = createAppEngineEnvironment();
private boolean isPlaceHolderNeeded;
public AppEngineEnvironment() {
this("PlaceholderAppId");
}
public AppEngineEnvironment(String appId) {
isPlaceHolderNeeded = ApiProxy.getCurrentEnvironment() == null;
// isPlaceHolderNeeded may be true when we are invoked in a test with AppEngineRule.
if (isPlaceHolderNeeded) {
ApiProxy.setEnvironmentForCurrentThread(PLACEHOLDER_ENV);
ApiProxy.setEnvironmentForCurrentThread(createAppEngineEnvironment(appId));
}
}
@@ -48,7 +62,7 @@ public class AppEngineEnvironment implements Closeable {
}
/** Returns a placeholder {@link Environment} that can return hardcoded AppId and Attributes. */
private static Environment createAppEngineEnvironment() {
private static Environment createAppEngineEnvironment(String appId) {
return (Environment)
Proxy.newProxyInstance(
Environment.class.getClassLoader(),
@@ -56,7 +70,7 @@ public class AppEngineEnvironment implements Closeable {
(Object proxy, Method method, Object[] args) -> {
switch (method.getName()) {
case "getAppId":
return "PlaceholderAppId";
return appId;
case "getAttributes":
return ImmutableMap.<String, Object>of();
default:
@@ -0,0 +1,75 @@
// Copyright 2020 The Nomulus Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package google.registry.beam.initsql;
import static com.google.common.base.Preconditions.checkArgument;
import static com.google.common.base.Preconditions.checkNotNull;
import com.google.appengine.api.datastore.Entity;
import java.util.Objects;
/** Helper for manipulating {@code DomainBase} when migrating from Datastore to SQL database */
final class DomainBaseUtil {
private DomainBaseUtil() {}
/**
* Removes {@link google.registry.model.billing.BillingEvent.Recurring}, {@link
* google.registry.model.poll.PollMessage PollMessages} and {@link
* google.registry.model.host.HostResource name servers} from a Datastore {@link Entity} that
* represents an Ofy {@link google.registry.model.domain.DomainBase}. This breaks the cycle of
* foreign key constraints between these entity kinds, allowing {@code DomainBases} to be inserted
* into the SQL database. See {@link InitSqlPipeline} for a use case, where the full {@code
* DomainBases} are written again during the last stage of the pipeline.
*
* <p>The returned object may be in bad state. Specifically, {@link
* google.registry.model.eppcommon.StatusValue#INACTIVE} is not added after name servers are
* removed. This only impacts tests.
*
* <p>This operation is performed on an Datastore {@link Entity} instead of Ofy Java object
* because Objectify requires access to a Datastore service when converting an Ofy object to a
* Datastore {@code Entity}. If we insist on working with Objectify objects, we face a few
* unsatisfactory options:
*
* <ul>
* <li>Connect to our production Datastore, which incurs unnecessary security and code health
* risk.
* <li>Connect to a separate real Datastore instance, which is a waster and overkill.
* <li>Use an in-memory test Datastore, which is a project health risk in that the test
* Datastore would be added to Nomulus' production binary unless we create a separate
* project for this pipeline.
* </ul>
*
* <p>Given our use case, operating on Datastore entities is the best option.
*
* @throws IllegalArgumentException if input does not represent a DomainBase
*/
static Entity removeBillingAndPollAndHosts(Entity domainBase) {
checkNotNull(domainBase, "domainBase");
checkArgument(
Objects.equals(domainBase.getKind(), "DomainBase"),
"Expecting DomainBase, got %s",
domainBase.getKind());
Entity clone = domainBase.clone();
clone.removeProperty("autorenewBillingEvent");
clone.removeProperty("autorenewPollMessage");
clone.removeProperty("deletePollMessage");
clone.removeProperty("nsHosts");
domainBase.getProperties().keySet().stream()
.filter(s -> s.startsWith("transferData."))
.forEach(s -> clone.removeProperty(s));
return clone;
}
}
@@ -0,0 +1,260 @@
// Copyright 2020 The Nomulus Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package google.registry.beam.initsql;
import static com.google.common.base.Preconditions.checkArgument;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableSet;
import com.googlecode.objectify.Key;
import google.registry.backup.AppEngineEnvironment;
import google.registry.backup.VersionedEntity;
import google.registry.beam.initsql.BeamJpaModule.JpaTransactionManagerComponent;
import google.registry.beam.initsql.Transforms.RemoveDomainBaseForeignKeys;
import google.registry.beam.initsql.Transforms.SerializableSupplier;
import google.registry.model.billing.BillingEvent;
import google.registry.model.contact.ContactResource;
import google.registry.model.domain.DomainBase;
import google.registry.model.domain.token.AllocationToken;
import google.registry.model.host.HostResource;
import google.registry.model.poll.PollMessage;
import google.registry.model.registrar.Registrar;
import google.registry.model.registrar.RegistrarContact;
import google.registry.model.registry.Registry;
import google.registry.model.reporting.HistoryEntry;
import google.registry.persistence.transaction.JpaTransactionManager;
import java.io.Serializable;
import java.util.Collection;
import java.util.Optional;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.PipelineResult;
import org.apache.beam.sdk.transforms.ParDo;
import org.apache.beam.sdk.transforms.SerializableFunction;
import org.apache.beam.sdk.transforms.Wait;
import org.apache.beam.sdk.values.PCollection;
import org.apache.beam.sdk.values.PCollectionTuple;
import org.apache.beam.sdk.values.TupleTag;
import org.joda.time.DateTime;
/**
* A BEAM pipeline that populates a SQL database with data from a Datastore backup.
*
* <p>This pipeline migrates EPP resources and related entities that cross-reference each other. To
* avoid violating foreign key constraints, writes to SQL are ordered by entity kinds. In addition,
* the {@link DomainBase} kind is written twice (see details below). The write order is presented
* below. Although some kinds can be written concurrently, e.g. {@code ContactResource} and {@code
* RegistrarContact}, we do not expect any performance benefit since the limiting resource is the
* number of JDBC connections. Google internal users may refer to <a
* href="http://go/registry-r3-init-sql">the design doc</a> for more information.
*
* <ol>
* <li>{@link Registry}: Assumes that {@code PremiumList} and {@code ReservedList} have been set
* up in the SQL database.
* <li>{@link Registrar}: Logically depends on {@code Registry}, Foreign key not modeled yet.
* <li>{@link ContactResource}: references {@code Registrar}
* <li>{@link RegistrarContact}: references {@code Registrar}.
* <li>Cleansed {@link DomainBase}: with references to {@code BillingEvent}, {@code Recurring},
* {@code Cancellation} and {@code HostResource} removed, still references {@code Registrar}
* and {@code ContactResource}. The removal breaks circular Foreign Key references.
* <li>{@link HostResource}: references {@code DomainBase}.
* <li>{@link HistoryEntry}: maps to one of three SQL entity types and may reference {@code
* Registrar}, {@code ContactResource}, {@code HostResource}, and {@code DomainBase}.
* <li>{@link AllocationToken}: references {@code HistoryEntry}.
* <li>{@link BillingEvent.Recurring}: references {@code Registrar}, {@code DomainBase} and {@code
* HistoryEntry}.
* <li>{@link BillingEvent.OneTime}: references {@code Registrar}, {@code DomainBase}, {@code
* BillingEvent.Recurring}, {@code HistoryEntry} and {@code AllocationToken}.
* <li>{@link BillingEvent.Modification}: SQL model TBD. Will reference {@code Registrar}, {@code
* DomainBase} and {@code BillingEvent.OneTime}.
* <li>{@link BillingEvent.Cancellation}: references {@code Registrar}, {@code DomainBase}, {@code
* BillingEvent.Recurring}, {@code BillingEvent.OneTime}, and {@code HistoryEntry}.
* <li>{@link PollMessage}: references {@code Registrar}, {@code DomainBase}, {@code
* ContactResource}, {@code HostResource}, and {@code HistoryEntry}.
* <li>{@link DomainBase}, original copy from Datastore.
* </ol>
*/
public class InitSqlPipeline implements Serializable {
/**
* Datastore kinds to be written to the SQL database before the cleansed version of {@link
* DomainBase}.
*/
// TODO(weiminyu): include Registry.class when it is modeled in JPA.
private static final ImmutableList<Class<?>> PHASE_ONE_ORDERED =
ImmutableList.of(Registrar.class, ContactResource.class);
/**
* Datastore kinds to be written to the SQL database after the cleansed version of {@link
* DomainBase}.
*
* <p>The following entities are missing from the list:
*
* <ul>
* <li>Those not modeled in JPA yet, e.g., {@code BillingEvent.Modification}.
* <li>Those waiting for sanitation, e.g., {@code HistoryEntry}, which would have duplicate keys
* after converting to SQL model.
* <li>Those that have foreign key constraints on the above.
* </ul>
*/
// TODO(weiminyu): add more entities when available.
private static final ImmutableList<Class<?>> PHASE_TWO_ORDERED =
ImmutableList.of(HostResource.class);
private final InitSqlPipelineOptions options;
private final Pipeline pipeline;
private final SerializableFunction<JpaTransactionManagerComponent, JpaTransactionManager>
jpaGetter;
InitSqlPipeline(InitSqlPipelineOptions options) {
this.options = options;
pipeline = Pipeline.create(options);
jpaGetter = JpaTransactionManagerComponent::cloudSqlJpaTransactionManager;
}
@VisibleForTesting
InitSqlPipeline(InitSqlPipelineOptions options, Pipeline pipeline) {
this.options = options;
this.pipeline = pipeline;
jpaGetter = JpaTransactionManagerComponent::localDbJpaTransactionManager;
}
public PipelineResult run() {
setupPipeline();
return pipeline.run();
}
@VisibleForTesting
void setupPipeline() {
PCollectionTuple datastoreSnapshot =
pipeline.apply(
"Load Datastore snapshot",
Transforms.loadDatastoreSnapshot(
options.getDatastoreExportDir(),
options.getCommitLogDir(),
DateTime.parse(options.getCommitLogStartTimestamp()),
DateTime.parse(options.getCommitLogEndTimestamp()),
ImmutableSet.<String>builder()
.add("DomainBase")
.addAll(toKindStrings(PHASE_ONE_ORDERED))
.addAll(toKindStrings(PHASE_TWO_ORDERED))
.build()));
// Set up the pipeline to write entity kinds from PHASE_ONE_ORDERED to SQL. Return a object
// that signals the completion of the phase.
PCollection<Void> blocker =
scheduleOnePhaseWrites(datastoreSnapshot, PHASE_ONE_ORDERED, Optional.empty(), null);
blocker =
writeToSql(
"DomainBase without circular foreign keys",
removeDomainBaseForeignKeys(datastoreSnapshot)
.apply("Wait on phase one", Wait.on(blocker)));
// Set up the pipeline to write entity kinds from PHASE_TWO_ORDERED to SQL. This phase won't
// start until all cleansed DomainBases have been written (started by line above).
scheduleOnePhaseWrites(
datastoreSnapshot, PHASE_TWO_ORDERED, Optional.of(blocker), "DomainBaseNoFkeys");
}
private PCollection<VersionedEntity> removeDomainBaseForeignKeys(
PCollectionTuple datastoreSnapshot) {
PCollection<VersionedEntity> domainBases =
datastoreSnapshot.get(Transforms.createTagForKind("DomainBase"));
return domainBases.apply(
"Remove circular foreign keys from DomainBase",
ParDo.of(new RemoveDomainBaseForeignKeys()));
}
/**
* Sets up the pipeline to write entities in {@code entityClasses} to SQL. Entities are written
* one kind at a time based on each kind's position in {@code entityClasses}. Concurrency exists
* within each kind.
*
* @param datastoreSnapshot the Datastore snapshot of all data to be migrated to SQL
* @param entityClasses the entity types in write order
* @param blockingPCollection the pipeline stage that blocks this phase
* @param blockingTag description of the stage (if exists) that blocks this phase. Needed for
* generating unique transform ids
* @return the output {@code PCollection} from the writing of the last entity kind. Other parts of
* the pipeline can {@link Wait} on this object
*/
private PCollection<Void> scheduleOnePhaseWrites(
PCollectionTuple datastoreSnapshot,
Collection<Class<?>> entityClasses,
Optional<PCollection<Void>> blockingPCollection,
String blockingTag) {
checkArgument(!entityClasses.isEmpty(), "Each phase must have at least one kind.");
ImmutableList<TupleTag<VersionedEntity>> tags =
toKindStrings(entityClasses).stream()
.map(Transforms::createTagForKind)
.collect(ImmutableList.toImmutableList());
PCollection<Void> prev = blockingPCollection.orElse(null);
String prevTag = blockingTag;
for (TupleTag<VersionedEntity> tag : tags) {
PCollection<VersionedEntity> curr = datastoreSnapshot.get(tag);
if (prev != null) {
curr = curr.apply("Wait on " + prevTag, Wait.on(prev));
}
prev = writeToSql(tag.getId(), curr);
prevTag = tag.getId();
}
return prev;
}
private PCollection<Void> writeToSql(String transformId, PCollection<VersionedEntity> data) {
String credentialFileUrl =
options.getSqlCredentialUrlOverride() != null
? options.getSqlCredentialUrlOverride()
: BackupPaths.getCloudSQLCredentialFilePatterns(options.getEnvironment()).get(0);
return data.apply(
"Write to sql: " + transformId,
Transforms.writeToSql(
transformId,
options.getMaxConcurrentSqlWriters(),
options.getSqlWriteBatchSize(),
new JpaSupplierFactory(credentialFileUrl, jpaGetter)));
}
private static ImmutableList<String> toKindStrings(Collection<Class<?>> entityClasses) {
try (AppEngineEnvironment env = new AppEngineEnvironment()) {
return entityClasses.stream().map(Key::getKind).collect(ImmutableList.toImmutableList());
}
}
static class JpaSupplierFactory implements SerializableSupplier<JpaTransactionManager> {
private static final long serialVersionUID = 1L;
private String credentialFileUrl;
private SerializableFunction<JpaTransactionManagerComponent, JpaTransactionManager> jpaGetter;
JpaSupplierFactory(
String credentialFileUrl,
SerializableFunction<JpaTransactionManagerComponent, JpaTransactionManager> jpaGetter) {
this.credentialFileUrl = credentialFileUrl;
this.jpaGetter = jpaGetter;
}
@Override
public JpaTransactionManager get() {
return jpaGetter.apply(
DaggerBeamJpaModule_JpaTransactionManagerComponent.builder()
.beamJpaModule(new BeamJpaModule(credentialFileUrl))
.build());
}
}
}
@@ -0,0 +1,84 @@
// Copyright 2020 The Nomulus Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package google.registry.beam.initsql;
import javax.annotation.Nullable;
import org.apache.beam.sdk.extensions.gcp.options.GcpOptions;
import org.apache.beam.sdk.options.Default;
import org.apache.beam.sdk.options.Description;
import org.apache.beam.sdk.options.Validation;
/** Pipeline options for {@link InitSqlPipeline} */
public interface InitSqlPipelineOptions extends GcpOptions {
@Description(
"Overrides the URL to the SQL credential file. " + "Required if environment is not provided.")
@Nullable
String getSqlCredentialUrlOverride();
void setSqlCredentialUrlOverride(String credentialUrlOverride);
@Description("The root directory of the export to load.")
String getDatastoreExportDir();
void setDatastoreExportDir(String datastoreExportDir);
@Description("The directory that contains all CommitLog files.")
String getCommitLogDir();
void setCommitLogDir(String commitLogDir);
@Description("The earliest CommitLogs to load, in ISO8601 format.")
@Validation.Required
String getCommitLogStartTimestamp();
void setCommitLogStartTimestamp(String commitLogStartTimestamp);
@Description("The latest CommitLogs to load, in ISO8601 format.")
@Validation.Required
String getCommitLogEndTimestamp();
void setCommitLogEndTimestamp(String commitLogEndTimestamp);
@Description(
"The deployed environment, alpha, crash, sandbox, or production. "
+ "Not required only if sqlCredentialUrlOverride is provided.")
@Nullable
String getEnvironment();
void setEnvironment(String environment);
@Description(
"The maximum JDBC connection pool size on a VM. "
+ "This value should be equal to or greater than the number of cores on the VM.")
@Default.Integer(4)
int getJdbcMaxPoolSize();
void setJdbcMaxPoolSize(int jdbcMaxPoolSize);
@Description(
"A hint to the pipeline runner of the maximum number of concurrent SQL writers to create. "
+ "Note that multiple writers may run on the same VM and share the connection pool.")
@Default.Integer(4)
int getMaxConcurrentSqlWriters();
void setMaxConcurrentSqlWriters(int maxConcurrentSqlWriters);
@Description("The number of entities to be written to the SQL database in one transaction.")
@Default.Integer(20)
int getSqlWriteBatchSize();
void setSqlWriteBatchSize(int sqlWriteBatchSize);
}
@@ -1,3 +1,17 @@
## Summary
This package contains a BEAM pipeline that populates a Cloud SQL database from a Datastore backup.
This package contains a BEAM pipeline that populates a Cloud SQL database from a
Datastore backup. The pipeline uses an unsynchronized Datastore export and
overlapping CommitLogs generated by the Nomulus server to recreate a consistent
Datastore snapshot, and writes the data to a Cloud SQL instance.
## Pipeline Visualization
The golden flow graph of the InitSqlPipeline is saved both as a text-base
[DOT file](../../../../../../test/resources/google/registry/beam/initsql/pipeline_golden.dot)
and a
[.png file](../../../../../../test/resources/google/registry/beam/initsql/pipeline_golden.png).
A test compares the flow graph of the current pipeline with the golden graph,
and will fail if changes are detected. When this happens, run the Gradle task
':core:updateInitSqlPipelineGraph' to update the golden files and review the
changes.
@@ -17,8 +17,10 @@ package google.registry.beam.initsql;
import static com.google.common.base.Preconditions.checkArgument;
import static com.google.common.base.Preconditions.checkNotNull;
import static com.google.common.base.Preconditions.checkState;
import static com.google.common.base.Throwables.throwIfUnchecked;
import static google.registry.beam.initsql.BackupPaths.getCommitLogTimestamp;
import static google.registry.beam.initsql.BackupPaths.getExportFilePatterns;
import static google.registry.persistence.JpaRetries.isFailedTxnRetriable;
import static google.registry.persistence.transaction.TransactionManagerFactory.jpaTm;
import static google.registry.persistence.transaction.TransactionManagerFactory.setJpaTm;
import static google.registry.util.DateTimeUtils.START_OF_TIME;
@@ -29,14 +31,16 @@ import static org.apache.beam.sdk.values.TypeDescriptors.kvs;
import static org.apache.beam.sdk.values.TypeDescriptors.strings;
import avro.shaded.com.google.common.collect.Iterators;
import com.google.appengine.api.datastore.Entity;
import com.google.appengine.api.datastore.EntityTranslator;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Throwables;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableMap;
import com.google.common.collect.Streams;
import google.registry.backup.AppEngineEnvironment;
import google.registry.backup.CommitLogImports;
import google.registry.backup.VersionedEntity;
import google.registry.model.domain.DomainBase;
import google.registry.model.ofy.ObjectifyService;
import google.registry.model.ofy.Ofy;
import google.registry.persistence.transaction.JpaTransactionManager;
@@ -49,7 +53,6 @@ import java.util.Optional;
import java.util.Set;
import java.util.concurrent.ThreadLocalRandom;
import java.util.function.Supplier;
import javax.persistence.OptimisticLockException;
import org.apache.beam.sdk.coders.StringUtf8Coder;
import org.apache.beam.sdk.io.Compression;
import org.apache.beam.sdk.io.FileIO;
@@ -70,7 +73,6 @@ import org.apache.beam.sdk.values.PBegin;
import org.apache.beam.sdk.values.PCollection;
import org.apache.beam.sdk.values.PCollectionList;
import org.apache.beam.sdk.values.PCollectionTuple;
import org.apache.beam.sdk.values.PDone;
import org.apache.beam.sdk.values.TupleTag;
import org.apache.beam.sdk.values.TupleTagList;
import org.apache.beam.sdk.values.TypeDescriptor;
@@ -225,7 +227,7 @@ public final class Transforms {
return new PTransform<PCollection<String>, PCollection<Metadata>>() {
@Override
public PCollection<Metadata> expand(PCollection<String> input) {
return input.apply(FileIO.matchAll().withEmptyMatchTreatment(EmptyMatchTreatment.DISALLOW));
return input.apply(FileIO.matchAll().withEmptyMatchTreatment(EmptyMatchTreatment.ALLOW));
}
};
}
@@ -263,6 +265,11 @@ public final class Transforms {
/**
* Returns a {@link PTransform} that writes a {@link PCollection} of entities to a SQL database.
* and outputs an empty {@code PCollection<Void>}. This allows other operations to {@link
* org.apache.beam.sdk.transforms.Wait wait} for the completion of this transform.
*
* <p>Errors are handled according to the pipeline runner's default policy. As part of a one-time
* job, we will not add features unless proven necessary.
*
* @param transformId a unique ID for an instance of the returned transform
* @param maxWriters the max number of concurrent writes to SQL, which also determines the max
@@ -270,22 +277,21 @@ public final class Transforms {
* @param batchSize the number of entities to write in each operation
* @param jpaSupplier supplier of a {@link JpaTransactionManager}
*/
public static PTransform<PCollection<VersionedEntity>, PDone> writeToSql(
public static PTransform<PCollection<VersionedEntity>, PCollection<Void>> writeToSql(
String transformId,
int maxWriters,
int batchSize,
SerializableSupplier<JpaTransactionManager> jpaSupplier) {
return new PTransform<PCollection<VersionedEntity>, PDone>() {
return new PTransform<PCollection<VersionedEntity>, PCollection<Void>>() {
@Override
public PDone expand(PCollection<VersionedEntity> input) {
input
public PCollection<Void> expand(PCollection<VersionedEntity> input) {
return input
.apply(
"Shard data for " + transformId,
MapElements.into(kvs(integers(), TypeDescriptor.of(VersionedEntity.class)))
.via(ve -> KV.of(ThreadLocalRandom.current().nextInt(maxWriters), ve)))
.apply("Batch output by shard " + transformId, GroupIntoBatches.ofSize(batchSize))
.apply("Write in batch for " + transformId, ParDo.of(new SqlBatchWriter(jpaSupplier)));
return PDone.in(input.getPipeline());
}
};
}
@@ -397,8 +403,10 @@ public final class Transforms {
public void setup() {
sleeper = new SystemSleeper();
ObjectifyService.initOfy();
ofy = ObjectifyService.ofy();
try (AppEngineEnvironment env = new AppEngineEnvironment()) {
ObjectifyService.initOfy();
ofy = ObjectifyService.ofy();
}
synchronized (SqlBatchWriter.class) {
if (instanceCount == 0) {
@@ -444,7 +452,10 @@ public final class Transforms {
runnable.run();
return;
} catch (Throwable throwable) {
throwIfNotCausedBy(throwable, OptimisticLockException.class);
if (!isFailedTxnRetriable(throwable)) {
throwIfUnchecked(throwable);
throw new RuntimeException(throwable);
}
int sleepMillis = (1 << attempt) * initialDelayMillis;
int jitter =
ThreadLocalRandom.current().nextInt((int) (sleepMillis * jitterRatio))
@@ -453,21 +464,28 @@ public final class Transforms {
}
}
}
}
/**
* Rethrows {@code throwable} if it is not (and does not have a cause of) {@code causeType};
* otherwise returns with no side effects.
*/
private void throwIfNotCausedBy(Throwable throwable, Class<? extends Throwable> causeType) {
Throwable t = throwable;
while (t != null) {
if (causeType.isInstance(t)) {
return;
}
t = t.getCause();
}
Throwables.throwIfUnchecked(t);
throw new RuntimeException(t);
/**
* Removes BillingEvents, {@link google.registry.model.poll.PollMessage PollMessages} and {@link
* google.registry.model.host.HostResource} from a {@link DomainBase}. These are circular foreign
* key constraints that prevent migration of {@code DomainBase} to SQL databases.
*
* <p>See {@link InitSqlPipeline} for more information.
*/
static class RemoveDomainBaseForeignKeys extends DoFn<VersionedEntity, VersionedEntity> {
@ProcessElement
public void processElement(
@Element VersionedEntity domainBase, OutputReceiver<VersionedEntity> out) {
checkArgument(
domainBase.getEntity().isPresent(), "Unexpected delete entity %s", domainBase.key());
Entity outputEntity =
DomainBaseUtil.removeBillingAndPollAndHosts(domainBase.getEntity().get());
out.output(
VersionedEntity.from(
domainBase.commitTimeMills(),
EntityTranslator.convertToPb(outputEntity).toByteArray()));
}
}
}
@@ -0,0 +1,58 @@
// Copyright 2019 The Nomulus Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package google.registry.persistence;
import com.google.common.base.Predicates;
import com.google.common.collect.ImmutableSet;
import java.sql.SQLException;
import java.util.function.Predicate;
import javax.persistence.OptimisticLockException;
/** Helpers for identifying retriable database operations. */
public final class JpaRetries {
private JpaRetries() {}
private static final ImmutableSet<String> RETRIABLE_TXN_SQL_STATE =
ImmutableSet.of(
"40001", // serialization_failure
"40P01", // deadlock_detected, PSQL-specific
"55006", // object_in_use, PSQL and DB2
"55P03" // lock_not_available, PSQL-specific
);
private static final Predicate<Throwable> RETRIABLE_TXN_PREDICATE =
Predicates.or(
OptimisticLockException.class::isInstance,
e ->
e instanceof SQLException
&& RETRIABLE_TXN_SQL_STATE.contains(((SQLException) e).getSQLState()));
public static boolean isFailedTxnRetriable(Throwable throwable) {
Throwable t = throwable;
while (t != null) {
if (RETRIABLE_TXN_PREDICATE.test(t)) {
return true;
}
t = t.getCause();
}
return false;
}
public static boolean isFailedQueryRetriable(Throwable throwable) {
// TODO(weiminyu): check for more error codes.
return isFailedTxnRetriable(throwable);
}
}