924060929 commented on code in PR #66348:
URL: https://github.com/apache/doris/pull/66348#discussion_r3733175299
##########
fe/fe-core/src/main/java/org/apache/doris/qe/Coordinator.java:
##########
@@ -2578,82 +2578,102 @@ public void
updateFragmentExecStatus(TReportExecStatusParams params) {
}
PipelineExecContext ctx =
pipelineExecContexts.get(Pair.of(params.getFragmentId(),
params.getBackendId()));
- if (ctx == null || !ctx.updatePipelineStatus(params)) {
+ boolean hasExternalCommitData = params.isSetHivePartitionUpdates()
+ || params.isSetIcebergCommitDatas() ||
params.isSetMcCommitDatas();
+ if (ctx == null) {
+ if (hasExternalCommitData) {
+ throw new IllegalStateException("Missing fragment handler for
external-file report");
+ }
+ return false;
+ }
+ if (!ctx.updatePipelineStatus(params)) {
+ if (hasExternalCommitData && !ctx.done) {
+ throw new IllegalStateException("External-file report was not
a completed fragment report");
+ }
LOG.debug("Fragment {} is not done, ignore report status: {}",
params.getFragmentId(), params.toString());
- return;
+ return ctx.done;
Review Comment:
`ctx.done` is read here outside the monitor (it is written only inside
`updatePipelineStatus`/`finishPipelineStatus`). The wait/notify happens-before
covers the wait path, but the `!params.done` early-return path reads it with no
synchronization — a data race on a plain boolean. Same for `txnId` (line 2636):
a plain long written outside any lock while other report threads read it. Low
practical risk given the BE protocol (commit data only on final reports,
single-instance top fragments), but `volatile` on both would make the reasoning
airtight.
##########
fe/fe-core/src/main/java/org/apache/doris/qe/QeProcessorImpl.java:
##########
@@ -57,6 +60,10 @@ public final class QeProcessorImpl implements QeProcessor {
private Map<TUniqueId, Integer> queryToInstancesNum;
private Map<String, AtomicInteger> userToInstancesCount;
private ExecutorService writeProfileExecutor;
+ private final Cache<String, Boolean> acceptedExternalFileReports =
CacheBuilder.newBuilder()
+ .maximumSize(1_000_000)
+ .expireAfterWrite(30, TimeUnit.MINUTES)
Review Comment:
Accept-then-reject window: once a report is accepted (commit data already
fed into the txn), this token is the only thing that keeps retries idempotent,
and it lives only 30 minutes / 1M entries. A retry that misses the cache (long
network partition, or eviction under write load) is rejected with
INTERNAL_ERROR. Per the PR description BE treats a definite rejection as
aborting the deferred uploads — so the txn keeps the already-fed commit data
while the files are deleted, and the final commit references objects that no
longer exist. The old code never had this window because it never rejected.
Suggest either aligning the TTL with the BE retry bound, or having BE keep
files until the txn terminates when the same report identity was previously
accepted. A test for the TTL-expiry case would also help (current tests cover
coordinator removal, but not token expiry).
##########
fe/fe-connector/fe-connector-iceberg/src/main/java/org/apache/doris/connector/iceberg/action/IcebergRemoveOrphanFilesAction.java:
##########
@@ -0,0 +1,370 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements. See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership. The ASF licenses this file
+// to you 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 org.apache.doris.connector.iceberg.action;
+
+import org.apache.doris.connector.spi.ConnectorColumn;
+import org.apache.doris.connector.spi.ConnectorSession;
+import org.apache.doris.connector.spi.ConnectorType;
+import org.apache.doris.connector.spi.DorisConnectorException;
+import org.apache.doris.connector.spi.pushdown.ConnectorPredicate;
+import org.apache.doris.foundation.util.ArgumentParsers;
+
+import com.google.common.collect.Lists;
+import org.apache.iceberg.DataFile;
+import org.apache.iceberg.DeleteFile;
+import org.apache.iceberg.ManifestContent;
+import org.apache.iceberg.ManifestFile;
+import org.apache.iceberg.ManifestFiles;
+import org.apache.iceberg.ManifestReader;
+import org.apache.iceberg.ReachableFileUtil;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.TableProperties;
+import org.apache.iceberg.io.FileInfo;
+import org.apache.iceberg.io.SupportsPrefixOperations;
+import org.apache.iceberg.util.PropertyUtil;
+
+import java.io.IOException;
+import java.net.URI;
+import java.time.Duration;
+import java.util.ArrayList;
+import java.util.HashSet;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Locale;
+import java.util.Map;
+import java.util.Objects;
+import java.util.Set;
+
+/** Safely lists or deletes old files that are unreachable from every retained
snapshot. */
+public class IcebergRemoveOrphanFilesAction extends BaseIcebergAction {
+ private static final long MIN_RETENTION_MS =
Duration.ofHours(24).toMillis();
+ private static final int MAX_REACHABLE_FILES = 5_000_000;
+ public static final String OLDER_THAN = "older_than";
+ public static final String LOCATION = "location";
+ public static final String DRY_RUN = "dry_run";
+ public static final String ALLOW_UNSAFE_LOCATION = "allow_unsafe_location";
+
+ public IcebergRemoveOrphanFilesAction(Map<String, String> properties,
List<String> partitionNames,
+ ConnectorPredicate whereCondition) {
+ super("remove_orphan_files", properties, partitionNames,
whereCondition);
+ }
+
+ @Override
+ protected void registerIcebergArguments() {
+ namedArguments.registerRequiredArgument(OLDER_THAN, "Creation time
cutoff in milliseconds",
+ ArgumentParsers.nonNegativeLong(OLDER_THAN));
+ namedArguments.registerOptionalArgument(LOCATION, "Prefix to scan for
orphan files",
+ null, ArgumentParsers.nonEmptyString(LOCATION));
+ namedArguments.registerOptionalArgument(DRY_RUN, "Only count orphan
files", true,
+ ArgumentParsers.booleanValue(DRY_RUN));
+ namedArguments.registerOptionalArgument(ALLOW_UNSAFE_LOCATION,
+ "Allow an explicitly supplied location whose table ownership
cannot be proved",
+ false, ArgumentParsers.booleanValue(ALLOW_UNSAFE_LOCATION));
+ }
+
+ @Override
+ protected void validateIcebergAction() {
+ validateNoPartitions();
+ validateNoWhereCondition();
+ String location = namedArguments.getString(LOCATION);
+ if (location != null) {
+ try {
+ normalizeLocation(location);
+ } catch (IllegalArgumentException e) {
+ throw new DorisConnectorException("Invalid location URI: " +
location, e);
+ }
+ }
+ }
+
+ @Override
+ protected List<String> executeAction(Table table, ConnectorSession
session) {
+ if (!(table.io() instanceof SupportsPrefixOperations)) {
+ throw new DorisConnectorException("remove_orphan_files requires
FileIO prefix listing support");
+ }
+ if (!PropertyUtil.propertyAsBoolean(table.properties(),
TableProperties.GC_ENABLED,
+ TableProperties.GC_ENABLED_DEFAULT)) {
+ // A GC-disabled table may share files with another table, so no
destructive scan is safe.
+ throw new DorisConnectorException("Cannot remove orphan files:
Iceberg GC is disabled");
+ }
+ long olderThan = namedArguments.getLong(OLDER_THAN);
+ // Reject an unsafe cutoff before opening any metadata or manifest
file.
+ if (olderThan > System.currentTimeMillis() - MIN_RETENTION_MS) {
+ throw new DorisConnectorException("older_than must retain at least
24 hours of files");
+ }
+ List<ScanScope> scanScopes = resolveScanScopes(table);
+
+ try {
+ ReachableIndex reachable = collectReachableFiles(table);
+ long orphanCount = 0;
+ long deletedCount = 0;
+ boolean dryRun = namedArguments.getBoolean(DRY_RUN);
+ for (ScanScope scope : scanScopes) {
+ // Object stores use raw prefix matching, so the separator
excludes sibling prefixes.
+ String listingPrefix = scope.root.endsWith("/") ? scope.root :
scope.root + "/";
+ for (FileInfo file : ((SupportsPrefixOperations)
table.io()).listPrefix(listingPrefix)) {
+ if (scope.owns(file.location()) && file.createdAtMillis()
< olderThan
Review Comment:
`createdAtMillis()` is the only protection against deleting in-flight files,
but the 24h cutoff collapses to zero if a FileIO implementation reports 0
(creation time unknown): `0 < olderThan` is always true, so an uncommitted
writer's file is deleted as an orphan. Suggest skipping deletion (fail closed)
when `createdAtMillis() <= 0`, and confirming every `SupportsPrefixOperations`
implementation used with this action returns a reliable creation time.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
