Allow to trim all ops above a certain seq# with a term lower than X by vladimirdolzhenko · Pull Request #30176 · elastic/elasticsearch

vladimirdolzhenko · 2018-04-26T13:10:39Z

Allow to trim all ops above a certain seq# with a term lower than X

Relates to #10708

elasticmachine · 2018-04-26T13:11:20Z

Pinging @elastic/es-distributed

bleskes

Thanks @vladimirdolzhenko . The basics look good. I did a sweep of the production code and left some comments. I will do the tests once these have been addressed.

bleskes · 2018-05-01T07:41:00Z

server/src/main/java/org/elasticsearch/action/resync/ResyncReplicationRequest.java

 public final class ResyncReplicationRequest extends ReplicatedWriteRequest<ResyncReplicationRequest> {

+    private long belowTermId;
+    private long trimmedAboveSeqNo;


nit: trimAboveSeqNo

bleskes · 2018-05-01T07:49:14Z

server/src/main/java/org/elasticsearch/action/resync/ResyncReplicationRequest.java

 */
 public final class ResyncReplicationRequest extends ReplicatedWriteRequest<ResyncReplicationRequest> {

+    private long belowTermId;


I think this can be removed. When we execute a replication operation, we are guaranteed that the replica is on the same term as the primary that generated the request. The trim semantics can be trimOperationOfPreviousPrimaryTerms(long minSeqNoToTrim)

bleskes · 2018-05-01T07:49:53Z

server/src/main/java/org/elasticsearch/action/resync/ResyncReplicationRequest.java

    public void readFrom(final StreamInput in) throws IOException {
        assert Version.CURRENT.major <= 7;
-        if (in.getVersion().equals(Version.V_6_0_0)) {
+        if (in.getVersion().onOrBefore(Version.V_6_0_0)) {


why did you change this?

bleskes · 2018-05-01T07:51:17Z

server/src/main/java/org/elasticsearch/action/resync/ResyncReplicationRequest.java

            throw new IllegalStateException("resync replication request serialization is broken in 6.0.0");
        }
        super.readFrom(in);
+        if (in.getVersion().onOrAfter(Version.V_6_4_0)) {


you need to start with 7.0.0. You can only push this down to 6.4 once this has been backported.

It's a bit unclear how it has to be done. Shall it be merged into master first with V_7_0_0, after backported to 6.x branch and after adjusted in master to V_6_X ?

There is a way to do it. First you build it with 7.0.0. Before backporting to 6.x you disable bwc tests on master, push your 6.x code and then renable bwc tests on master. I'll show you how to do it once we're ready.

bleskes · 2018-05-01T07:52:28Z

server/src/main/java/org/elasticsearch/action/resync/ResyncReplicationRequest.java

        super.readFrom(in);
+        if (in.getVersion().onOrAfter(Version.V_6_4_0)) {
+            belowTermId = in.readVLong();
+            trimmedAboveSeqNo = in.readVLong();


UNASSIGNED_SEQ_NO is -2 and NO_OPS_PERFORMED is -1. Since readVLong can't take negative numbers we use readZLong

bleskes · 2018-05-01T08:30:29Z

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


+    /**
+     * Trims translog (effectively moves max visible seq# {@link Checkpoint#trimmedAboveSeqNo}) for terms below <code>belowTerm</code>
+     * and seq# above <code>aboveSeqNo</code>


can you expend a bit that these operations will not be exposed in snapshots?

bleskes · 2018-05-01T08:31:15Z

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

+            return;
+        }
+
+        final ChannelFactory channelFactory = FileChannel::open;


nit: how much does this buy us? I think we can just inline it and have better readability?

bleskes · 2018-05-01T08:35:48Z

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

+                    IOUtils.fsync(checkpointFile, false);
+                    IOUtils.fsync(checkpointFile.getParent(), true);
+
+                    newReader = reader.withNewCheckpoint(newCheckpoint);


this is super dangerous. If we have an exception here (or in pervious iterations) we will leak a file channel, as no one closes newReader. I think you need testing for this, see TranslogTests#testWithRandomException to see how this can be tested.

agreed, going to fix it

bleskes · 2018-05-01T08:39:36Z

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

+     * Create a new reader with new checkoint that shares resources with current one
+     */
+    TranslogReader withNewCheckpoint(final Checkpoint newCheckpoint){
+        return new TranslogReader(newCheckpoint, channel, path, header,  closed);


the are a couple of problems with this:

A closed translog should never be trimmed.

There should be a clear transfer of owner ship of the underlying channel between the old reader and the new - i.e. the old reader should be closed and the new reader should be open.

You can see how to do this kind of ownership transfer by looking at TranslogWriter.closeIntoReader .

Instead of splitting the trimming logic of a reader in two places (Translog.trim and here), I think we should move it all here. Call the method closeIntoTrimmedReader and do all the checkpoint + state juggling here.

agreed, going to fix it

bleskes · 2018-05-01T08:40:29Z

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

-            return null;
+        while (readOperations < totalOperations) {
+            final Translog.Operation operation = readOperation();
+            if (operation.seqNo() <= checkpoint.trimmedAboveSeqNo || checkpoint.trimmedAboveSeqNo < 0) {


can hard check timmerdAboveSeqNo to UNASSINGED_SEQ_NO? I don't like the implicit lower than 0. It's hard to grep for usages like that.

bleskes · 2018-05-01T08:42:12Z

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

    }

+    /**
+     * Trims translog (effectively moves max visible seq# {@link Checkpoint#trimmedAboveSeqNo}) for terms below <code>belowTerm</code>


also we should be clear we're talking about primary terms of the files not the operations.

Relates to elastic#10708

… fixes on PR#2 Relates to elastic#10708

… fixes on PR elastic#3 Relates to elastic#10708

… fixes on PR elastic#4 Relates to elastic#10708

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

+        /**
+         * The number of operations have been skipped (overridden or trimmed) in the snapshot so far.
+         */
+        default int skippedOperations() {


… fixes on PR elastic#5 Relates to elastic#10708

… fixes for the case of no entities Relates to elastic#10708

bleskes

Thanks @vladimirdolzhenko . Good iteration. I left some more comments.

bleskes · 2018-05-09T10:41:07Z

server/src/main/java/org/elasticsearch/action/resync/TransportResyncReplicationAction.java

            }
        }
+        if (request.getTrimAboveSeqNo() != SequenceNumbers.UNASSIGNED_SEQ_NO) {
+            replica.trimOperationOfPreviousPrimaryTerms(replica.getPrimaryTerm(), request.getTrimAboveSeqNo());


there is no need to pass the primary term to the replica - it already knows it.

server/src/main/java/org/elasticsearch/index/engine/Engine.java

    public abstract boolean isThrottled();

+    /**
+     * Trims translog for terms below <code>belowTerm</code> and seq# above <code>aboveSeqNo</code>


bleskes · 2018-05-09T10:44:30Z

server/src/main/java/org/elasticsearch/index/engine/Engine.java

+     * Trims translog for terms below <code>belowTerm</code> and seq# above <code>aboveSeqNo</code>
+     */
+    public void trimTranslog(long belowTerm, long aboveSeqNo) throws IOException {
+        getTranslog().trim(belowTerm, aboveSeqNo);


wrap with exceptions handling to fail the engine if this fails. Also, I think you want to hold the read lock, just for sanity. See InternalEngine#rollTranslogGeneration for an example

bleskes · 2018-05-09T10:45:09Z

server/src/main/java/org/elasticsearch/index/shard/IndexShard.java

        assert currentEngineReference.get() == null;
    }

+    public void trimOperationOfPreviousPrimaryTerms(long belowTerm, long aboveSeqNo) throws IOException {


See comment about remove the belowTerm parameter and using the shard's term

bleskes · 2018-05-09T10:48:19Z

server/src/main/java/org/elasticsearch/index/shard/PrimaryReplicaSyncer.java

            }

-            if (!operations.isEmpty()) {
+            final long trimmedAboveSeqNo = firstMessage.get() && maxSeqNo > 0 ? maxSeqNo : SequenceNumbers.UNASSIGNED_SEQ_NO;


I think you want maxSeqNo >= 0 (0 is a valid sequence number). Semantically, I think it's nice to say maxSeqNo != SequenceNumbers.NO_OPS_PERFORMED. Also, double check your tests that they cover this case (only one operation on the primary and two ops on replica. It will be good to understand why this wasn't catched.

bleskes · 2018-05-09T11:36:23Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        }
+
+        failableTLog.rollGeneration();
+        fail.failAlways();


I don't think you need to fail always but rather fail randomly. This is important to test failures that can come in any step, not just the first.

bleskes · 2018-05-09T11:36:48Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+            failableTLog.add(operation);
+        }
+
+        failableTLog.rollGeneration();


as in the previous test - you need way more evilness - more roll generations, more terms, move variance.

bleskes · 2018-05-09T11:37:01Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+            operations.add(operation);
+        }
+        // shuffle a bit - move several first items to the end
+        for(int i = 0, len = randomIntBetween(5, 10); i < len; i++){


see previous comments about shuffle.

bleskes · 2018-05-09T11:37:17Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        }
+
+        try {
+            failableTLog.newSnapshot();


you can use expectThrows

bleskes · 2018-05-09T11:38:11Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        } catch (MockDirectoryWrapper.FakeIOException ex) {
+            // all is fine
+        }
+


please make sure all files closed and no file is leaked.

do we really need to manually check that all files are closed while that check is provided by ESTestCase ?

If that's the case, we don't need - that's making sure :D - where do you see it's done in ESTestCase? (I didn't check myself)

TestRuleTemporaryFilesCleanup takes care of it (in case I don't do close of file handlers):

java.lang.RuntimeException: file handle leaks: [FileChannel(/private/var/folders/dz/750jvg8j31j21s52xyhzcj4c0000gn/T/org.elasticsearch.index.translog.TranslogTests_FBDFE9B843A061D2-001/tempDir-003/translog-1.tlog), FileChannel(/private/var/folders/dz/750jvg8j31j21s52xyhzcj4c0000gn/T/org.elasticsearch.index.translog.TranslogTests_FBDFE9B843A061D2-001/tempDir-003/translog-2.tlog), FileChannel(/private/var/folders/dz/750jvg8j31j21s52xyhzcj4c0000gn/T/org.elasticsearch.index.translog.TranslogTests_FBDFE9B843A061D2-001/tempDir-003/translog-3.tlog)] at __randomizedtesting.SeedInfo.seed([FBDFE9B843A061D2]:0) at org.apache.lucene.mockfile.LeakFS.onClose(LeakFS.java:63) at org.apache.lucene.mockfile.FilterFileSystem.close(FilterFileSystem.java:77) at org.apache.lucene.mockfile.FilterFileSystem.close(FilterFileSystem.java:78) at org.apache.lucene.util.TestRuleTemporaryFilesCleanup.afterAlways(TestRuleTemporaryFilesCleanup.java:228) at com.carrotsearch.randomizedtesting.rules.TestRuleAdapter$1.afterAlways(TestRuleAdapter.java:31) at com.carrotsearch.randomizedtesting.rules.StatementAdapter.evaluate(StatementAdapter.java:43) at com.carrotsearch.randomizedtesting.rules.StatementAdapter.evaluate(StatementAdapter.java:36) at org.apache.lucene.util.TestRuleAssertionsRequired$1.evaluate(TestRuleAssertionsRequired.java:53) at org.apache.lucene.util.TestRuleMarkFailure$1.evaluate(TestRuleMarkFailure.java:47) at org.apache.lucene.util.TestRuleIgnoreAfterMaxFailures$1.evaluate(TestRuleIgnoreAfterMaxFailures.java:64) at org.apache.lucene.util.TestRuleIgnoreTestSuites$1.evaluate(TestRuleIgnoreTestSuites.java:54) at com.carrotsearch.randomizedtesting.rules.StatementAdapter.evaluate(StatementAdapter.java:36) at com.carrotsearch.randomizedtesting.ThreadLeakControl$StatementRunner.run(ThreadLeakControl.java:368) at java.base/java.lang.Thread.run(Thread.java:844) Caused by: java.lang.Exception at org.apache.lucene.mockfile.LeakFS.onOpen(LeakFS.java:46) at org.apache.lucene.mockfile.HandleTrackingFS.callOpenHook(HandleTrackingFS.java:81) at org.apache.lucene.mockfile.HandleTrackingFS.newFileChannel(HandleTrackingFS.java:197) at org.apache.lucene.mockfile.HandleTrackingFS.newFileChannel(HandleTrackingFS.java:166) at java.base/java.nio.channels.FileChannel.open(FileChannel.java:292) at java.base/java.nio.channels.FileChannel.open(FileChannel.java:340) at org.elasticsearch.index.translog.Translog.openReader(Translog.java:279) at org.elasticsearch.index.translog.Translog.recoverFromFiles(Translog.java:225) at org.elasticsearch.index.translog.Translog.<init>(Translog.java:177) at org.elasticsearch.index.translog.TranslogTests$4.<init>(TranslogTests.java:2128) at org.elasticsearch.index.translog.TranslogTests.getFailableTranslog(TranslogTests.java:2128) at org.elasticsearch.index.translog.TranslogTests.testExceptionOnTrimAboveSeqNo(TranslogTests.java:1552)

sorry - that rule is defined in LuceneTestCase rather than in ESTestCase

… fixes for PR#6 Relates to elastic#10708

bleskes

Thanks @vladimirdolzhenko . I did another round. Getting close.

bleskes · 2018-05-10T12:59:42Z

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

                IOUtils.closeWhileHandlingException(newReaders);
-                IOUtils.closeWhileHandlingException(current);
-                IOUtils.closeWhileHandlingException(readers);
+                close();


please wrap with try catch and add any exception as an suppressed exception to e

bleskes · 2018-05-10T13:03:02Z

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

        final long globalCheckpoint = SequenceNumbers.UNASSIGNED_SEQ_NO;
-        final long minTranslogGeneration = -1L;
-        return new Checkpoint(in.readLong(), in.readInt(), in.readLong(), minSeqNo, maxSeqNo, globalCheckpoint, minTranslogGeneration);
+        final long minTranslogGeneration = SequenceNumbers.UNASSIGNED_SEQ_NO;


?? this is not a sequence number??

bleskes · 2018-05-10T13:08:27Z

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

+     */
+    TranslogReader closeIntoTrimmedReader(long belowTerm, long aboveSeqNo, ChannelFactory channelFactory) throws IOException {
+        ensureOpen();
+        if (getPrimaryTerm() < belowTerm && aboveSeqNo < checkpoint.maxSeqNo) {


See #30176 (comment) , which I think you missed. This method is called close and I think it should always close. term based checking can be done on the translog layer

nope, the check is in Translog - agree that ensureOpen check in TranslogReader can go.

server/src/test/java/org/elasticsearch/index/replication/RecoveryDuringReplicationTests.java


    @TestLogging("org.elasticsearch.index.shard:TRACE,org.elasticsearch.action.resync:TRACE")
    public void testResyncAfterPrimaryPromotion() throws Exception {
-        // TODO: check translog trimming functionality once it's implemented


server/src/test/java/org/elasticsearch/index/replication/RecoveryDuringReplicationTests.java

                assertThat(task.getResyncedOperations(), greaterThanOrEqualTo(extraDocs));
            }
-            shards.assertAllEqual(initialDocs + extraDocs);
+            List<IndexShard> replicas = shards.getReplicas();


bleskes · 2018-05-10T13:29:06Z

server/src/test/java/org/elasticsearch/index/replication/RecoveryDuringReplicationTests.java

+                Translog.Operation next;
+                while ((next = snapshot.next()) != null) {
+                    translogOperations++;
+                    assertTrue("unexpected op: " + next, next.seqNo() < initialDocs + extraDocs);


can we strength this and look at doc ids also confirm they have the right seq# + term as on the primary ? (now you can have a seq# that happens to be fine but with the wrong doc)

bleskes · 2018-05-10T13:29:56Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        }
+
+        primaryTerm.incrementAndGet();
+        translog.rollGeneration();


see #30176 (comment)

sorry, i thought it relates to testRandomExceptionsOnTrimOperations

bleskes · 2018-05-10T13:31:12Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        final Translog failableTLog =
+            getFailableTranslog(fail, config, randomBoolean(), false, null, createTranslogDeletionPolicy(), fileChannels);
+
+        expectThrows(IOException.class,


This enforce that the exception is thrown, but the current code doesn't always throw one.

bleskes · 2018-05-10T13:31:47Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+        expectThrows(IOException.class,
+            () -> {
+                int translogOperations = 0;
+                int maxAttempts = randomIntBetween(5, 10);


just fix a number, I don't think randomizing this adds much.

bleskes · 2018-05-10T13:58:14Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+            fileChannels.stream().filter(f -> f.isOpen()).findFirst().isPresent(), is(false));
+
+        final AlreadyClosedException alreadyClosedException = expectThrows(AlreadyClosedException.class, () -> failableTLog.newSnapshot());
+        assertThat(alreadyClosedException.getMessage(),


can you check that you can reopen the translog and that you can at least read all operations that weren't trimmed? this should not result in a translog corruption.

… Relates to elastic#10708 added TODO note for test regarding Lucene rollback

… Relates to elastic#10708 dropped extra debug info of replicas seq# stats

… Relates to elastic#10708 check source of a doc after trimming

… Relates to elastic#10708 minTranslogGeneration has NOT to refer to SeqNumbers

vladimirdolzhenko · 2018-06-01T17:39:49Z

@elasticmachine test this please

bleskes

LGTM. I left one suggestion that I don't feel needs another cycle. Thanks @vladimirdolzhenko for all the iterations.

bleskes · 2018-06-05T08:16:41Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

+            operationsList
+                .stream()
+                // handle all expect last one - it has `current` ops
+                .limit(operationsList.size() - 1)


"Current" ops should never be trimmed due to our assumptions, that said I think it's good not to model that here but rather make sure that the way we generate ops conforms to those assumptions?

vladimirdolzhenko · 2018-06-05T17:34:36Z

thanks @bleskes for comments - indeed it's better to keep current operations and other operations apart, could you please have a look ?

vladimirdolzhenko · 2018-06-05T23:51:05Z

@elasticmachine test this please

bleskes · 2018-06-06T14:58:08Z

@vladimirdolzhenko I think there's some misunderstanding (I'm happy to discuss). What I meant in my comment is that we shouldn't make any distinction between current and the rest in the competing mock translog. That one should be just a dumb array of operations where trim simply.

vladimirdolzhenko · 2018-06-06T16:21:46Z

@bleskes thank you 👍 indeed - it should be no any assumption what to trim - all or (all expect current) - current has not to contain any op with primaryTerm >= belowTerm

vladimirdolzhenko · 2018-06-06T21:13:12Z

@elasticmachine test this please

bleskes · 2018-06-07T11:28:53Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

-            return IntStream.range(0, size + 1)
-                // current + reverse traverse of operations
-                .mapToObj(i -> i == 0 ? currentOperations : operationsList.get(size - i))
+            return IntStream.range(0, size)


why just use operationList.stream() ?

I'm sorry. got confused. nevermind

That said, instead of being fancy here, we can do this on add and existing ops if they have the same seq# but a lower term. That will improve the test.

vladimirdolzhenko · 2018-06-07T20:46:20Z

@bleskes thanks, it makes it way simpler

bleskes · 2018-06-08T11:47:03Z

server/src/test/java/org/elasticsearch/index/translog/TranslogTests.java

-        void rollGeneration() {
-            operationsList.add(new LinkedList<>());
+            final Translog.Operation old = operations.put(operation.seqNo(), operation);
+            assert old == null || old.primaryTerm() <= operation.primaryTerm();


vladimirdolzhenko · 2018-06-08T16:14:28Z

thanks a lot @bleskes for review and comments

* master: Move default location of dependencies report (#31228) Remove dependencies report task dependencies (#31227) Add recognition of MPL 2.0 (#31226) Fix unknown licenses (#31223) Remove version from license file name for GCS SDK (#31221) Fully encapsulate LocalCheckpointTracker inside of the engine (#31213) [DOCS] Added 'fail_on_unsupported_field' param to MLT. Closes #28008 (#31160) Add licenses for transport-nio (#31218) Remove DocumentFieldMappers#simpleMatchToFullName. (#31041) Allow to trim all ops above a certain seq# with a term lower than X, post backport fix (#31211) Compliant SAML Response destination check (#31175) Remove DocumentFieldMappers#smartNameFieldMapper, as it is no longer needed. (#31018) Remove extraneous references to 'tokenized' in the mapper code. (#31010) Allow to trim all ops above a certain seq# with a term lower than X (#30176) SQL: Make a single JDBC driver jar (#31012) Enhance license detection for various licenses (#31198) [DOCS] Add note about long-lived idle connections (#30990) Move number of language analyzers to analysis-common module (#31143) Default max concurrent search req. numNodes * 5 (#31171) flush job to ensure all results have been written (#31187)

…ecker * elastic/master: (309 commits) [test] add fix for rare virtualbox error (elastic#31212) Move default location of dependencies report (elastic#31228) Remove dependencies report task dependencies (elastic#31227) Add recognition of MPL 2.0 (elastic#31226) Fix unknown licenses (elastic#31223) Remove version from license file name for GCS SDK (elastic#31221) Fully encapsulate LocalCheckpointTracker inside of the engine (elastic#31213) [DOCS] Added 'fail_on_unsupported_field' param to MLT. Closes elastic#28008 (elastic#31160) Add licenses for transport-nio (elastic#31218) Remove DocumentFieldMappers#simpleMatchToFullName. (elastic#31041) Allow to trim all ops above a certain seq# with a term lower than X, post backport fix (elastic#31211) Compliant SAML Response destination check (elastic#31175) Remove DocumentFieldMappers#smartNameFieldMapper, as it is no longer needed. (elastic#31018) Remove extraneous references to 'tokenized' in the mapper code. (elastic#31010) Allow to trim all ops above a certain seq# with a term lower than X (elastic#30176) SQL: Make a single JDBC driver jar (elastic#31012) Enhance license detection for various licenses (elastic#31198) [DOCS] Add note about long-lived idle connections (elastic#30990) Move number of language analyzers to analysis-common module (elastic#31143) Default max concurrent search req. numNodes * 5 (elastic#31171) ...

vladimirdolzhenko requested a review from bleskes April 26, 2018 13:10

vladimirdolzhenko added :Distributed/Recovery Anything around constructing a new shard, either from a local or a remote source. backport pending :Distributed/Engine Anything around managing Lucene and the Translog in an open shard. labels Apr 26, 2018

vladimirdolzhenko force-pushed the trim_translog branch 2 times, most recently from 38446d1 to 594dff0 Compare April 27, 2018 09:41

Allow to trim all ops above a certain seq# with a term lower than X

d3d210a

Relates to elastic#10708

vladimirdolzhenko force-pushed the trim_translog branch from 594dff0 to d3d210a Compare April 30, 2018 06:52

bleskes suggested changes May 1, 2018

View reviewed changes

bleskes reviewed May 1, 2018

View reviewed changes

Vladimir Dolzhenko added 5 commits May 4, 2018 10:33

Allow to trim all ops above a certain seq# with a term lower than X

e622794

Relates to elastic#10708

Allow to trim all ops above a certain seq# with a term lower than X -…

586908c

… fixes on PR#2 Relates to elastic#10708

Allow to trim all ops above a certain seq# with a term lower than X -…

92addd1

… fixes on PR#2 Relates to elastic#10708

Allow to trim all ops above a certain seq# with a term lower than X -…

2a9c58c

… fixes on PR elastic#3 Relates to elastic#10708

Allow to trim all ops above a certain seq# with a term lower than X -…

1e67a8a

… fixes on PR elastic#4 Relates to elastic#10708

vladimirdolzhenko commented May 4, 2018

View reviewed changes

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

/**

* The number of operations have been skipped (overridden or trimmed) in the snapshot so far.

*/

default int skippedOperations() {

This comment was marked as resolved.

Sign in to view

Vladimir Dolzhenko added 3 commits May 4, 2018 16:17

Allow to trim all ops above a certain seq# with a term lower than X -…

d454a30

… fixes on PR elastic#5 Relates to elastic#10708

merged from master

92d2d68

Allow to trim all ops above a certain seq# with a term lower than X -…

13ae045

… fixes for the case of no entities Relates to elastic#10708

vladimirdolzhenko requested a review from bleskes May 8, 2018 08:06

bleskes suggested changes May 9, 2018

View reviewed changes

Vladimir Dolzhenko added 2 commits May 10, 2018 11:36

Allow to trim all ops above a certain seq# with a term lower than X -…

762f298

… fixes for PR#6 Relates to elastic#10708

Allow to trim all ops above a certain seq# with a term lower than X -…

3896280

… fixes for PR#6 Relates to elastic#10708

vladimirdolzhenko requested a review from bleskes May 10, 2018 10:15

bleskes suggested changes May 10, 2018

View reviewed changes

Vladimir Dolzhenko added 4 commits May 11, 2018 13:16

Allow to trim all ops above a certain seq# with a term lower than X -…

2c13980

… Relates to elastic#10708 added TODO note for test regarding Lucene rollback

Allow to trim all ops above a certain seq# with a term lower than X -…

49167c8

… Relates to elastic#10708 dropped extra debug info of replicas seq# stats

Allow to trim all ops above a certain seq# with a term lower than X -…

eaef2e4

… Relates to elastic#10708 check source of a doc after trimming

Allow to trim all ops above a certain seq# with a term lower than X -…

ef7811d

… Relates to elastic#10708 minTranslogGeneration has NOT to refer to SeqNumbers

Merge remote-tracking branch 'remotes/origin/master' into trim_translog

ba339d9

bleskes approved these changes Jun 5, 2018

View reviewed changes

keep operationsList and currentOperations apart

0f9b206

drop assumption on not trimming current in InMemoryTranslog

9d74767

Merge remote-tracking branch 'remotes/origin/master' into trim_translog

546c890

bleskes reviewed Jun 7, 2018

View reviewed changes

Vladimir Dolzhenko added 2 commits June 7, 2018 10:09

Merge remote-tracking branch 'remotes/origin/master' into trim_translog

db4105f

InMemoryTranslog simplification

88f702b

bleskes reviewed Jun 8, 2018

View reviewed changes

vladimirdolzhenko added v7.0.0 v6.4.0 labels Jun 8, 2018

vladimirdolzhenko merged commit a86c0f8 into elastic:master Jun 8, 2018

vladimirdolzhenko removed the backport pending label Jun 8, 2018

vladimirdolzhenko mentioned this pull request Jun 22, 2018

IndexShard should not return null stats #31528

Merged

vladimirdolzhenko deleted the trim_translog branch June 22, 2018 15:49

bleskes mentioned this pull request Aug 14, 2018

Add Sequence Numbers to write operations #10708

Closed

64 tasks

lcawl added the >enhancement label Aug 20, 2018

jpountz removed the :Distributed/Recovery Anything around constructing a new shard, either from a local or a remote source. label Jan 29, 2019

colings86 added v7.0.0-beta1 and removed v7.0.0 labels Feb 7, 2019

Conversation

vladimirdolzhenko commented Apr 26, 2018

Uh oh!

elasticmachine commented Apr 26, 2018

Uh oh!

bleskes left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment was marked as resolved.

Uh oh!

bleskes left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment was marked as resolved.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bleskes left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!