Ignore translog retention policy if soft-deletes enabled (#45473)

Since #45136, we use soft-deletes instead of translog in peer recovery.
There's no need to retain extra translog to increase a chance of
operation-based recoveries. This commit ignores the translog retention
policy if soft-deletes is enabled so we can discard translog more
quickly.

Backport of #45473
Relates #45136

Author: dnhatn

docs/reference/index-modules/translog.asciidoc

@@ -76,12 +76,23 @@ commit point. Defaults to `512mb`.
 
 `index.translog.retention.size`::
 
-The total size of translog files to keep. Keeping more translog files increases
-the chance of performing an operation based sync when recovering replicas. If
-the translog files are not sufficient, replica recovery will fall back to a
-file based sync. Defaults to `512mb`
+When soft deletes is disabled (enabled by default in 7.0 or later),
+`index.translog.retention.size` controls the total size of translog files to keep.
+Keeping more translog files increases the chance of performing an operation based
+sync when recovering replicas. If the translog files are not sufficient,
+replica recovery will fall back to a file based sync. Defaults to `512mb`
+
+Both `index.translog.retention.size` and `index.translog.retention.age` should not
+be specified unless soft deletes is disabled as they will be ignored.
 
 
 `index.translog.retention.age`::
 
-The maximum duration for which translog files will be kept. Defaults to `12h`.
+When soft deletes is disabled (enabled by default in 7.0 or later),
+`index.translog.retention.age` controls the maximum duration for which translog
+files to keep. Keeping more translog files increases the chance of performing an
+operation based sync when recovering replicas. If the translog files are not sufficient,
+replica recovery will fall back to a file based sync. Defaults to `12h`
+
+Both `index.translog.retention.size` and `index.translog.retention.age` should not
+be specified unless soft deletes is disabled as they will be ignored.

rest-api-spec/src/main/resources/rest-api-spec/test/indices.stats/20_translog.yml

@@ -1,14 +1,14 @@
 ---
-setup:
+"Translog retention without soft_deletes":
   - do:
       indices.create:
-          index: test
+        index: test
+        body:
+          settings:
+            soft_deletes.enabled: false
   - do:
       cluster.health:
         wait_for_no_initializing_shards: true
-
----
-"Translog retention":
   - do:
       indices.stats:
         metric: [ translog ]
@@ -64,6 +64,53 @@ setup:
   - lte: { indices.test.primaries.translog.uncommitted_size_in_bytes: $creation_size }
   - match: { indices.test.primaries.translog.uncommitted_operations: 0 }
 
+---
+"Translog retention with soft_deletes":
+  - skip:
+      version: " - 7.3.99"
+      reason:  "start ignoring translog retention policy with soft-deletes enabled in 7.4"
+  - do:
+      indices.create:
+        index: test
+        body:
+          settings:
+            soft_deletes.enabled: true
+  - do:
+      cluster.health:
+        wait_for_no_initializing_shards: true
+  - do:
+      indices.stats:
+        metric: [ translog ]
+  - set: { indices.test.primaries.translog.size_in_bytes: creation_size }
+
+  - do:
+      index:
+        index: test
+        id:    1
+        body:  { "foo": "bar" }
+
+  - do:
+      indices.stats:
+        metric: [ translog ]
+  - gt: { indices.test.primaries.translog.size_in_bytes: $creation_size }
+  - match: { indices.test.primaries.translog.operations: 1 }
+  - match: { indices.test.primaries.translog.uncommitted_operations: 1 }
+  # call flush twice to sync the global checkpoint after the last operation so that we can have the safe commit
+  - do:
+      indices.flush:
+        index: test
+  - do:
+      indices.flush:
+        index: test
+  - do:
+      indices.stats:
+        metric: [ translog ]
+  # after flushing we have one empty translog file while an empty index before flushing has two empty translog files.
+  - lt: { indices.test.primaries.translog.size_in_bytes: $creation_size }
+  - match: { indices.test.primaries.translog.operations: 0 }
+  - lt: { indices.test.primaries.translog.uncommitted_size_in_bytes: $creation_size }
+  - match: { indices.test.primaries.translog.uncommitted_operations: 0 }
+
 ---
 "Translog last modified age stats":
   - skip:
@@ -81,11 +128,20 @@ setup:
   - gte: { indices.test.primaries.translog.earliest_last_modified_age: 0 }
 
 ---
-"Translog stats on closed indices":
+"Translog stats on closed indices without soft-deletes":
   - skip:
       version: " - 7.2.99"
       reason:  "closed indices have translog stats starting version 7.3.0"
 
+  - do:
+      indices.create:
+        index: test
+        body:
+          settings:
+            soft_deletes.enabled: false
+  - do:
+      cluster.health:
+        wait_for_no_initializing_shards: true
   - do:
       index:
         index: test
@@ -123,3 +179,40 @@ setup:
         forbid_closed_indices: false
   - match: { indices.test.primaries.translog.operations: 3 }
   - match: { indices.test.primaries.translog.uncommitted_operations: 0 }
+
+---
+"Translog stats on closed indices with soft-deletes":
+  - skip:
+      version: " - 7.3.99"
+      reason:  "start ignoring translog retention policy with soft-deletes enabled in 7.4"
+  - do:
+      indices.create:
+        index: test
+        body:
+          settings:
+            soft_deletes.enabled: true
+  - do:
+      cluster.health:
+        wait_for_no_initializing_shards: true
+  - do:
+      index:
+        index: test
+        id:    1
+        body:  { "foo": "bar" }
+  - do:
+      indices.stats:
+        metric: [ translog ]
+  - match: { indices.test.primaries.translog.operations: 1 }
+  - match: { indices.test.primaries.translog.uncommitted_operations: 1 }
+  - do:
+      indices.close:
+        index: test
+        wait_for_active_shards: 1
+  - is_true: acknowledged
+  - do:
+      indices.stats:
+        metric: [ translog ]
+        expand_wildcards: all
+        forbid_closed_indices: false
+  - match: { indices.test.primaries.translog.operations: 0 }
+  - match: { indices.test.primaries.translog.uncommitted_operations: 0 }

server/src/main/java/org/elasticsearch/index/IndexSettings.java

@@ -195,24 +195,6 @@ public final class IndexSettings {
             new ByteSizeValue(Long.MAX_VALUE, ByteSizeUnit.BYTES),
             Property.Dynamic, Property.IndexScope);
 
-    /**
-     * Controls how long translog files that are no longer needed for persistence reasons
-     * will be kept around before being deleted. A longer retention policy is useful to increase
-     * the chance of ops based recoveries.
-     **/
-    public static final Setting<TimeValue> INDEX_TRANSLOG_RETENTION_AGE_SETTING =
-        Setting.timeSetting("index.translog.retention.age", TimeValue.timeValueHours(12), TimeValue.timeValueMillis(-1),
-            Property.Dynamic, Property.IndexScope);
-
-    /**
-     * Controls how many translog files that are no longer needed for persistence reasons
-     * will be kept around before being deleted. Keeping more files is useful to increase
-     * the chance of ops based recoveries.
-     **/
-    public static final Setting<ByteSizeValue> INDEX_TRANSLOG_RETENTION_SIZE_SETTING =
-        Setting.byteSizeSetting("index.translog.retention.size", new ByteSizeValue(512, ByteSizeUnit.MB), Property.Dynamic,
-            Property.IndexScope);
-
     /**
      * The maximum size of a translog generation. This is independent of the maximum size of
      * translog operations that have not been flushed.
@@ -258,6 +240,27 @@ public final class IndexSettings {
         Setting.longSetting("index.soft_deletes.retention.operations", 0, 0,
             Property.IndexScope, Property.Dynamic);
 
+    /**
+     * Controls how long translog files that are no longer needed for persistence reasons
+     * will be kept around before being deleted. Keeping more files is useful to increase
+     * the chance of ops based recoveries for indices with soft-deletes disabled.
+     * This setting will be ignored if soft-deletes is enabled.
+     **/
+    public static final Setting<TimeValue> INDEX_TRANSLOG_RETENTION_AGE_SETTING =
+        Setting.timeSetting("index.translog.retention.age",
+            settings -> INDEX_SOFT_DELETES_SETTING.get(settings) ? TimeValue.MINUS_ONE : TimeValue.timeValueHours(12), TimeValue.MINUS_ONE,
+            Property.Dynamic, Property.IndexScope);
+
+    /**
+     * Controls how many translog files that are no longer needed for persistence reasons
+     * will be kept around before being deleted. Keeping more files is useful to increase
+     * the chance of ops based recoveries for indices with soft-deletes disabled.
+     * This setting will be ignored if soft-deletes is enabled.
+     **/
+    public static final Setting<ByteSizeValue> INDEX_TRANSLOG_RETENTION_SIZE_SETTING =
+        Setting.byteSizeSetting("index.translog.retention.size", settings -> INDEX_SOFT_DELETES_SETTING.get(settings) ? "-1" : "512MB",
+            Property.Dynamic, Property.IndexScope);
+
     /**
      * Controls the maximum length of time since a retention lease is created or renewed before it is considered expired.
      */
@@ -466,8 +469,6 @@ public IndexSettings(final IndexMetaData indexMetaData, final Settings nodeSetti
         syncInterval = INDEX_TRANSLOG_SYNC_INTERVAL_SETTING.get(settings);
         refreshInterval = scopedSettings.get(INDEX_REFRESH_INTERVAL_SETTING);
         flushThresholdSize = scopedSettings.get(INDEX_TRANSLOG_FLUSH_THRESHOLD_SIZE_SETTING);
-        translogRetentionAge = scopedSettings.get(INDEX_TRANSLOG_RETENTION_AGE_SETTING);
-        translogRetentionSize = scopedSettings.get(INDEX_TRANSLOG_RETENTION_SIZE_SETTING);
         generationThresholdSize = scopedSettings.get(INDEX_TRANSLOG_GENERATION_THRESHOLD_SIZE_SETTING);
         mergeSchedulerConfig = new MergeSchedulerConfig(this);
         gcDeletesInMillis = scopedSettings.get(INDEX_GC_DELETES_SETTING).getMillis();
@@ -493,6 +494,8 @@ public IndexSettings(final IndexMetaData indexMetaData, final Settings nodeSetti
         this.indexSortConfig = new IndexSortConfig(this);
         searchIdleAfter = scopedSettings.get(INDEX_SEARCH_IDLE_AFTER);
         defaultPipeline = scopedSettings.get(DEFAULT_PIPELINE);
+        setTranslogRetentionAge(scopedSettings.get(INDEX_TRANSLOG_RETENTION_AGE_SETTING));
+        setTranslogRetentionSize(scopedSettings.get(INDEX_TRANSLOG_RETENTION_SIZE_SETTING));
 
         scopedSettings.addSettingsUpdateConsumer(MergePolicyConfig.INDEX_COMPOUND_FORMAT_SETTING, mergePolicyConfig::setNoCFSRatio);
         scopedSettings.addSettingsUpdateConsumer(MergePolicyConfig.INDEX_MERGE_POLICY_DELETES_PCT_ALLOWED_SETTING,
@@ -553,11 +556,21 @@ private void setTranslogFlushThresholdSize(ByteSizeValue byteSizeValue) {
     }
 
     private void setTranslogRetentionSize(ByteSizeValue byteSizeValue) {
-        this.translogRetentionSize = byteSizeValue;
+        if (softDeleteEnabled && byteSizeValue.getBytes() >= 0) {
+            // ignore the translog retention settings if soft-deletes enabled
+            this.translogRetentionSize = new ByteSizeValue(-1);
+        } else {
+            this.translogRetentionSize = byteSizeValue;
+        }
     }
 
     private void setTranslogRetentionAge(TimeValue age) {
-        this.translogRetentionAge = age;
+        if (softDeleteEnabled && age.millis() >= 0) {
+            // ignore the translog retention settings if soft-deletes enabled
+            this.translogRetentionAge = TimeValue.MINUS_ONE;
+        } else {
+            this.translogRetentionAge = age;
+        }
     }
 
     private void setGenerationThresholdSize(final ByteSizeValue generationThresholdSize) {
@@ -734,13 +747,19 @@ public TimeValue getRefreshInterval() {
     /**
      * Returns the transaction log retention size which controls how much of the translog is kept around to allow for ops based recoveries
      */
-    public ByteSizeValue getTranslogRetentionSize() { return translogRetentionSize; }
+    public ByteSizeValue getTranslogRetentionSize() {
+        assert softDeleteEnabled == false || translogRetentionSize.getBytes() == -1L : translogRetentionSize;
+        return translogRetentionSize;
+    }
 
     /**
      * Returns the transaction log retention age which controls the maximum age (time from creation) that translog files will be kept
      * around
      */
-    public TimeValue getTranslogRetentionAge() { return translogRetentionAge; }
+    public TimeValue getTranslogRetentionAge() {
+        assert softDeleteEnabled == false || translogRetentionAge.millis() == -1L : translogRetentionSize;
+        return translogRetentionAge;
+    }
 
     /**
      * Returns the generation threshold size. As sequence numbers can cause multiple generations to

server/src/main/java/org/elasticsearch/index/translog/TruncateTranslogAction.java

@@ -177,11 +177,19 @@ private boolean isTranslogClean(ShardPath shardPath, String translogUUID) throws
             final TranslogConfig translogConfig = new TranslogConfig(shardPath.getShardId(), translogPath,
                 indexSettings, BigArrays.NON_RECYCLING_INSTANCE);
             long primaryTerm = indexSettings.getIndexMetaData().primaryTerm(shardPath.getShardId().id());
-            final TranslogDeletionPolicy translogDeletionPolicy =
-                new TranslogDeletionPolicy(indexSettings.getTranslogRetentionSize().getBytes(),
-                    indexSettings.getTranslogRetentionAge().getMillis());
+            // We open translog to check for corruption, do not clean anything.
+            final TranslogDeletionPolicy retainAllTranslogPolicy = new TranslogDeletionPolicy(Long.MAX_VALUE, Long.MAX_VALUE) {
+                @Override
+                long minTranslogGenRequired(List<TranslogReader> readers, TranslogWriter writer) {
+                    long minGen = writer.generation;
+                    for (TranslogReader reader : readers) {
+                        minGen = Math.min(reader.generation, minGen);
+                    }
+                    return minGen;
+                }
+            };
             try (Translog translog = new Translog(translogConfig, translogUUID,
-                translogDeletionPolicy, () -> translogGlobalCheckpoint, () -> primaryTerm, seqNo -> {});
+                retainAllTranslogPolicy, () -> translogGlobalCheckpoint, () -> primaryTerm, seqNo -> {});
                  Translog.Snapshot snapshot = translog.newSnapshot()) {
                 //noinspection StatementWithEmptyBody we are just checking that we can iterate through the whole snapshot
                 while (snapshot.next() != null) {

server/src/test/java/org/elasticsearch/index/IndexServiceTests.java

@@ -396,15 +396,20 @@ public void testAsyncTranslogTrimTaskOnClosedIndex() throws Exception {
         final Path translogPath = translog.getConfig().getTranslogPath();
         final String translogUuid = translog.getTranslogUUID();
 
+        int translogOps = 0;
         final int numDocs = scaledRandomIntBetween(10, 100);
         for (int i = 0; i < numDocs; i++) {
             client().prepareIndex().setIndex(indexName).setId(String.valueOf(i)).setSource("{\"foo\": \"bar\"}", XContentType.JSON).get();
+            translogOps++;
             if (randomBoolean()) {
                 client().admin().indices().prepareFlush(indexName).get();
+                if (indexService.getIndexSettings().isSoftDeleteEnabled()) {
+                    translogOps = 0;
+                }
             }
         }
-        assertThat(translog.totalOperations(), equalTo(numDocs));
-        assertThat(translog.stats().estimatedNumberOfOperations(), equalTo(numDocs));
+        assertThat(translog.totalOperations(), equalTo(translogOps));
+        assertThat(translog.stats().estimatedNumberOfOperations(), equalTo(translogOps));
         assertAcked(client().admin().indices().prepareClose("test").setWaitForActiveShards(ActiveShardCount.DEFAULT));
 
         indexService = getInstanceFromNode(IndicesService.class).indexServiceSafe(indexService.index());