If a shard was serving as a replica when another shard was promoted to
primary, then its Lucene index was reset to the global checkpoint.
However, if the new primary fails before the primary/replica resync
completes and we are now being promoted, we have to restore the reverted
operations by replaying the translog to avoid losing acknowledged writes.
Relates #33473
Relates #32867
It's possible for the set "seqNos" to contain only the "unFinishedSeq"
in the testConcurrentReplica test. If this is the case, the call
`randomValueOtherThan` won't make any progress because the predicate
will never be false.
This commit removes this expectation because it's incorrect and it's no
longer needed as we have a dedicated test to verify the contains method.
Relates #33871
Drops `Settings` from some of the methods to lookup loggers and
deprecates another logger lookup that takes `Settings` because
`Settings` is no longer required to build a logger.
* ingest: support simulate with verbose for pipeline processor
This change better supports the use of simulate?verbose with the
pipeline processor. Prior to this change any pipeline processors
executed with simulate?verbose would not show all intermediate
processors for the inner pipelines.
This changes also moves the PipelineProcess and TrackingResultProcessor
classes to enable instance checks and to avoid overly public classes.
As well this updates the error message for when cycles are detected
in pipelines calling other pipelines.
Today all searches happen on the search threadpool which is the correct
behavior in almost any case. Yet, there are exceptions where for instance
searches searches should be passed through a single-thread
thread-pool to reduce impact on a node. This change adds a index-private setting that allows to mark an index as throttled for searches and forks off all non-stats searcher access to this thread-pool for indices that are marked as `index.search.throttled`
Transient settings override persistent settings, but in fact all of the tests
that run as part of `:server:test` and `:server:integTest` will pass if the
precedence is changed to be the other way round. This change adds a test that
verifies the precedence is as documented.
With this commit we clear the fielddata cache per field as it is
supposed to be. Previously we retrieved the proper field from the cache
but then cleared the entire cache anyway.
Closes#33798
Relates #33807
This change adds "contains" method to LocalCheckpointTracker.
One of the use cases is to check if a given operation has been processed
in an engine or not by looking up its seq_no in LocalCheckpointTracker.
Relates #33656
Changes the default of the `node.name` setting to the hostname of the
machine on which Elasticsearch is running. Previously it was the first 8
characters of the node id. This had the advantage of producing a unique
name even when the node name isn't configured but the disadvantage of
being unrecognizable and not being available until fairly late in the
startup process. Of particular interest is that it isn't available until
after logging is configured. This forces us to use a volatile read
whenever we add the node name to the log.
Using the hostname is available immediately on startup and is generally
recognizable but has the disadvantage of not being unique when run on
machines that don't set their hostname or when multiple elasticsearch
processes are run on the same host. I believe that, taken together, it
is better to default to the hostname.
1. Running multiple copies of Elasticsearch on the same node is a fairly
advanced feature. We do it all the as part of the elasticsearch build
for testing but we make sure to set the node name then.
2. That the node.name defaults to some flavor of "localhost" on an
unconfigured box feels like it isn't going to come up too much in
production. I expect most production deployments to at least set the
hostname.
As a bonus, production deployments need no longer set the node name in
most cases. At least in my experience most folks set it to the hostname
anyway.
By moving CompletionStats into the engine we can easily cache the stats for
read-only engines if necessary. It also moves the responsibiltiy out of IndexShard
which has quiet some complexity already.
Relates to #33835
Today if we fetch common stats from a shard we might get a partial response
if the shard is closed while we fetch the stats. This causes hard to track and
reproduce NPEs. This change streamlines null checking to ensure we only render
stats we actually received.
Wraps all lines in our test framework at 140 characters because that is
our standard line length and removes all of the checkstyle suppressions
for the test framework.
Drops most of `ModuleTestCase` because it isn't used and we're moving
away from using guice in the way that it wants to test anyway. Also
switches a few classes that extend it but don't use it to extend
`ESTestCase` instead.
We currently special-case SynonymFilterFactory and SynonymGraphFilterFactory, which need to
know their predecessors in the analysis chain in order to correctly analyze their synonym lists. This
special-casing doesn't work with Referring filter factories, such as the Multiplexer or Conditional
filters. We also have a number of filters (eg the Multiplexer) that will break synonyms when they
appear before them in a chain, because they produce multiple tokens at the same position.
This commit adds two methods to the TokenFilterFactory interface.
* `getChainAwareTokenFilterFactory()` allows a filter factory to rewrite itself against its preceding
filter chain, or to resolve references to other filters. It replaces `ReferringFilterFactory` and
`CustomAnalyzerProvider.checkAndApplySynonymFilter`, and by default returns `this`.
* `getSynonymFilter()` defines whether or not a filter should be applied when building a synonym
list `Analyzer`. By default it returns `true`.
Fixes#33609
This test occasionally fails in `testCollectSearchShards` waiting on what seems
to be a search request to a remote cluster for one second. Given that the test
fails here very rarely I suspect maybe one second is very rarely not enough so
we could fix it by increasing the max wait time slightly.
Closes#33852
By moving DocStats into the engine we can easily cache the stats for
read-only engines if necessary. It also moves the responsibility out of IndexShard
which has quiet some complexity already.
The fix in #33757 introduces some workaround since FilterCodecReader didn't
support unwrapping. This cuts over to a more elegant fix to access the readers
segment infos.
This commit changes the random_score function to use the global docID of the document
rather than the segment docID to generate random scores. As a result documents that have
the same segment docID within the shard will generate different scores.
Add minimal sanity checks to custom/scripted similarities.
Lucene 8 introduced more constraints on similarities, in particular:
- scores must not be negative,
- scores must not decrease when term freq increases,
- scores must not increase when norm (interpreted as an unsigned long)
increases.
We can't check every single case, but could at least run some sanity checks.
Relates #33309
* Profiler: Don’t profile NEXTDOC for ConstantScoreQuery.
A ConstantScore query will return the iterator of its inner query.
However, when profiling, the constant score query is wrapped separately
from its inner query, which distorts the times emitted by the profiler.
Return the iterator directly in such a case.
Closes#23430
The change in #27500 introduces this regression that causes `_get` and `_term_vector`
actions to run on the network thread if the realtime flag is set.
This fixes the issue by delegating to the super method forking on the corresponding threadpool.
We use similar / same concepts in SerachTransportService and HandledTransportAction but both
duplicate the efforts with slightly different implementation details. This streamlines
sending responses / exceptions back to a channel in an ActionListener with appropriate logging.
In #33241 we moved the file-based discovery functionality to core
Elasticsearch, but preserved the `discovery-file` plugin, and support for the
existing location of the `unicast_hosts.txt` file, for BWC reasons. This commit
completes the removal of this plugin.
New plugin for annotated_text field type.
Largely a copy of `text` field type but adds ability to include markdown-like syntax in the text.
The “AnnotatedText” class parses text+markup and converts into plain text and AnnotationTokens.
The annotation token values are injected unchanged alongside the regular text tokens to provide a
form of additional indexed overlay useful in positional searches and highlighting.
Annotated_text fields do not support fielddata as we want to phase this out.
Also includes a new "annotated" highlighter type that retains annotations and merges in search
hits as additional annotation markup.
Closes#29467
* MINOR: Drop Redundant Ctx. Check in ScriptService
* This check is completely redundant, the expression script
engine will throw anyway (and with a similar message) for
those contexts that it cannot compile. Moreover, the update context
is not the only context that is not suported by the expression engine
at this point so handling the update context separately here makes
no sense.
* Same fix idea as in #10666a4 to prevent background
threads trying to reconnect after the tests are done from
throwing `ExecutionCancelledException` and breaking the test
* Closes#30714
Allows to skip shard balancing when the cluster_concurrent_rebalance threshold is already reached, which cuts down the time spent in the rebalance method of BalancedShardsAllocator.
Currently `IndexMetadata#getCustomData(...)` wraps the custom metadata
in an unmodifiable map, but in case there is no entry for the specified
key then a NPE is thrown by Collections.unmodifiableMap(...). This is not
ideal in case callers like to throw an exception with a specific message.
(like in the case for ccr to indicate that the follow index was not created
by the create_and_follow api and therefor incompatible as follow index)
I think making `DiffableStringMap` itself immutable is better then just wrapping
custom metadata with `Collections.unmodifiableMap(...)` in all methods that access it.
Also removed the `equals()`, `hashcode()` and to `toString()` methods of
`DiffableStringMap`, because `AbstractMap` already implements these methods.
This commit switches the joda time backcompat in scripting to use
augmentation over ZonedDateTime. The augmentation methods provide
compatibility with the missing methods between joda's DateTime and
java's ZonedDateTime. Due to getDayOfWeek returning an enum in the java
API, ZonedDateTime is wrapped so that the method can return int like the
joda time does. The java time api version is renamed to
getDayOfWeekEnum, which will be kept through 7.x for compatibility while
users switch back to getDayOfWeek once joda compatibility is removed.
This commit adds a guard around the rare case that no documents in the
10 iterations actually have any values, thus making the warning check
incorrect.
closes#32779
This commit is a cleanup of the assertions in global checkpoint
listeners, simplifying them and adding some messages to them in case the
assertions trip.
When we add a global checkpoint listener, it is also carries along with
it a value that it thinks is the current global checkpoint. This value
can be above the actual global checkpoint on a shard if the listener
knows the global checkpoint from another shard copy (e.g., the primary),
and the current shard copy is lagging behind. Today we notify the
listener whenever the global checkpoint advances, regardless if it goes
above the current global checkpoint known to the listener. This commit
reworks this implementation. Rather than thinking of the value
associated with the listener as the current global checkpoint known to
the listener, we think of it as the value that the listener is waiting
for the global checkpoint to advance to (inclusive). Now instead of
notifying all waiting listeners when the global checkpoint advances, we
only notify those that are waiting for a value not larger than the
actual global checkpoint that we advanced to.
This is something that we were already doing when sorting by field, which is
now also done when sorting by score. As-is this change will speed up top-k
`term` queries. This could work for `match_all` queries as well when we
implement the `setMinCompetitiveScore` API on their Scorer.
The existing approach used date formatters when a format based string
like `date_time||epoch_millis` was used, instead of the custom code.
In order to properly solve this, a new interface called
`DateFormatter` has been added, which now can be implemented for custom
formatters. Currently there are two implementations, one using java time
and one doing the epoch_millis formatter, which simply parses a number
and then converts it to a date in UTC timezone.
The DateFormatter interface now also has a method to retrieve the name
of the formatter pattern, which is needed for mapping changes anyway.
The existing `CompoundDateTimeFormatter` class has been removed, the
name was not really nice anyway.
One more minor change is the fact, that the new java time using
FormatDateFormatter does not try to parse the date with its printer
implementation first (which might be a strict one and fail), but a
printer can now be specified in addition. This saves one potential
failure/exception when parsing less strict dates.
If only a printer is specified, the printer will also be used as a
parser.
This race can occur if the latch from the listener notifies the test
thread and the test thread races ahead before the scheduler thread has a
chance to emit the log message. This commit fixes this test by not
counting down the latch until after the log message we are going to
assert on has been emitted.
We used TimeoutException here but that's not serializable. This commit
switches to a serializable exception so that we can test for the
exception type on the remote side.
This change fixes a bug introduced in 6.3 that prevents fields with an explicit
similarity to be updated. It also adds a test that checks this case for similarities
but also for analyzers since they could suffer from the same problem.
Closes#33611
Today we use a special unicast hosts provider, the `MockUncasedHostsProvider`,
in many integration tests, to deal with the dynamic nature of the allocation of
ports to nodes. However #33241 allows us to use file-based discovery to achieve
the same goal, so the special test-only `MockUncasedHostsProvider` is no longer
required.
This change removes `MockUncasedHostProvider` and replaces it with file-based
discovery in tests based on `EsIntegTestCase`.
We fail to notify the resync listener if the resync replication hits a
shard unavailable exception. Moreover, we no longer need to swallow
these unavailable exceptions.
Relates #28571Closes#33613
This field does not need to be volatile because all accesses are done
under a lock. This commit removes the unnecessary volatile modifier from
this field.
The remote cluster settings search.remote.* have been renamed to
cluster.remote.* and are automatically upgraded in the cluster state on
gateway recovery, and on put. This commit adds a note to the migration
docs for these changes.
This change adds a `_source` only snapshot repository that allows to wrap
any existing repository as a _backend_ to snapshot only the `_source` part
including live docs markers. Snapshots taken with the `source` repository
won't include any indices, doc-values or points. The snapshot will be reduced in size and
functionality such that it requires full re-indexing after it's successfully restored.
The restore process will copy the `_source` data locally starts a special shard and engine
to allow `match_all` scrolls and searches. Any other query, or get call will fail with and unsupported operation exception. The restored index is also marked as read-only.
This feature aims mainly for disaster recovery use-cases where snapshot size is
a concern or where time to restore is less of an issue.
**NOTE**: The snapshot produced by this repository is still a valid lucene index. This change doesn't allow for any longer retention policies which is out of scope for this change.
In cross-cluster replication, we will use global checkpoint listeners to
long poll for updates to a shard. However, we do not want these polls to
wait indefinitely as it could be difficult to discern if the listener is
still waiting for updates versus something has gone horribly wrong and
cross-cluster replication is stuck. Instead, we want these listeners to
timeout after some period (for example, one minute) so that they are
notified and we can update status on the following side that
cross-cluster replication is still active. After this, we will
immediately enter back into a poll mode.
To do this, we need the ability to associate a timeout with a global
checkpoint listener. This commit adds this capability.
This change clarifies the documentation of the context completion suggester
regarding filtering and boosting with contexts.
Unlike the suggester v1, filtering on multiple contexts
works as a disjunction, a suggestion matches if it contains at least one of the provided
context values and boosting selects the maximum score among the matching contexts.
This commit also adapts an old test that was written for the v1 suggester and commented out
for version 2 because the behavior changed.
This commit adds settings upgraders for the search.remote.* settings
that can be in the cluster state to automatically upgrade these settings
to cluster.remote.*. Because of the infrastructure that we have here,
these settings can be upgraded when recovering the cluster state, but
also when a user tries to make a dynamic update for these settings.
This commit adds test coverage for two cases not previously covered by
the existing testing. Namely, we add coverage ensuring that the executor
is used to notify listeners being added that are immediately notified
because the shard is closed or because the global checkpoint is already
beyond what the listener knows.
When a replica starts following a newly promoted primary, it may have
some operations which don't exist on the new primary. Thus we need to
throw those operations to align a replica with the new primary. This can
be done by first resetting an engine from the safe commit, then replaying
the local translog up to the global checkpoint.
Relates #32867
* Improves doc values format deprecation message
This changes the deprecation message when doc values fields do not
supply a format form logging a deprecation warning for each offending
field individually to logging a single message which lists all
offending fields
Closes#33572
* Updates YAML test with new deprecation message
Also adds a test to ensure multiple deprecation warnings are collated
into one message
* Condenses collection of fields without format check
Moves the collection of fields that don't have a format to a separate
loop and moves the logging of the deprecation warning to be next to it
at the expesnse of looping through the field list twice
* fixes typo
* Fixes test
Currently we keep track of how many bytes are currently being written to disk
in an AtomicLong within InternalEngine, updating it on refresh. The IndexWriter
has its own accounting for this, and exposes it via a getFlushingBytes method
in the latest lucene 8 snapshot. This commit removes the InternalEngine tracking
in favour of just using the IndexWriter method.
Upgrading list settings is broken because of the conversion that we do
to strings, and then when we try to put back the upgraded value we do
not know that it is a representation of a list. This commit addresses
this by adding special handling for list settings.
This change adds an engine implementation that opens a reader on an
existing index but doesn't permit any refreshes or modifications
to the index.
Relates to #32867
Relates to #32844
When we see a settings value, it could be a list. Yet this should only
happen if the underlying setting type is a list setting type. This
commit adds validation that when we get a setting value that is a list,
that the setting that we are getting is a list setting. And similarly,
if we get a value for a list setting, the underlying value should be a
list.
This change copies and validates the soft-deletes setting during resize.
If the source enables soft-deletes, the target must also enable it.
Closes#33321
* LeafCollector.setScorer() now takes a Scorable
* Scorers may not have null Weights
* IndexWriter.getFlushingBytes() reports how much memory is being used by IW threads writing to disk
Today the FilterRoutingTests take the belt-and-braces approach of excluding
some node attribute values and including some others. This means that we don't
really test that both inclusion and exclusion work correctly: as long as one of
them works as expected then the test will pass. This change improves these
tests by only using one approach at once, demonstrating that both do indeed
work, and adds tests for various other scenarios too.
In some cases we want to deprecate a setting, and then automatically
upgrade uses of that setting to a replacement setting. This commit adds
infrastructure for this so that we can upgrade settings when recovering
the cluster state, as well as when such settings are dynamically applied
on cluster update settings requests. This commit only focuses on cluster
settings, index settings can build on this infrastructure in a
follow-up.
This commit ensures that we bootstrap a new history_uuid when force
allocating a stale primary. A stale primary should never be the source
of an operation-based recovery to another shard which exists before the
forced-allocation.
Closes#26712
* CompoundProcessor is in the ingest package now
-> resolved
* Java generics don't offer type checking so nothing
can be done here -> remvoed TODO and test
* #16019 was closed and not acted on
-> todo can go away
Today when checking settings dependencies, we do not check if fallback
settings are present. This means, for example, that if
cluster.remote.*.seeds falls back to search.remote.*.seeds, and
cluster.remote.*.skip_unavailable and search.remote.*.skip_unavailable
depend on cluster.remote.*.seeds, and we have set search.remote.*.seeds
and search.remote.*.skip_unavailable, then validation will fail because
it is expected that cluster.ermote.*.seeds is set here. This commit
addresses this by also checking fallback settings when validating
dependencies. To do this, we adjust the settings exist method to also
check for fallback settings, a case that it was not handling previously.
Change the logging infrastructure to handle when the node name isn't
available in `elasticsearch.yml`. In that case the node name is not
available until long after logging is configured. The biggest change is
that the node name logging no longer fixed at pattern build time.
Instead it is read from a `SetOnce` on every print. If it is unset it is
printed as `unknown` so we have something that fits in the pattern.
On normal startup we don't log anything until the node name is available
so we never see the `unknown`s.
We invoke force merge twice in the test to verify that recovery sources
are pruned when the global checkpoint advanced. However, if the global
checkpoint equals to the local checkpoint in the first force-merge, the
second force-merge will be a noop because all deleted docs are expunged
in the first merge already. We need to flush a new segment to make merge
happen so we can verify that all recovery sources are pruned.
Instead of passing DirectoryService which causes yet another dependency
on Store we can just pass in a Directory since we will just call
`DirectoryService#newDirectory()` on it anyway.
This change collapses all metrics aggregations classes into a single package `org.elasticsearch.aggregations.metrics`.
It also restricts the visibility of some classes (aggregators and factories) that should not be used outside of the package.
Relates #22868
When we rollover and index we write the conditions of the rollover that
the old index met into the old index. Loading this index metadata
requires a working `NamedXContentRegistry` that has been populated with
parsers from the rollover infrastructure. We had a few loads that didn't
use a working `NamedXContentRegistry` and so would fail if they ever
encountered an index that had been rolled over. Here are the locations
of the loads and how I fixed them:
* IndexFolderUpgrader - removed entirely. It existed to support opening
indices made in Elasticsearch 2.x. Since we only need this change as far
back as 6.4.1 which will supports reading from indices created as far
back as 5.0.0 we should be good here.
* TransportNodesListGatewayStartedShards - wired the
`NamedXContentRegistry` into place.
* TransportNodesListShardStoreMetaData - wired the
`NamedXContentRegistry` into place.
* OldIndexUtils - removed entirely. It existed to support the zip based
index backwards compatibility tests which we've since replaced with code
that actually runs old versions of Elasticsearch.
In addition to fixing the actual problem I added full cluster restart
integration tests for rollover which would have caught this problem and
I added an extra assertion to IndexMetaData's deserialization code which
will trip if we try to deserialize and index's metadata without a fully
formed `NamedXContentRegistry`. It won't catch if use the *wrong*
`NamedXContentRegistry` but it is better than nothing.
Closes#33316
This commit allows us to use different TranslogRecoveryRunner when
recovering an engine from its local translog. This change is a
prerequisite for the commit-based rollback PR.
Relates #32867
The main benefit of the upgrade for users is the search optimization for top scored documents when the total hit count is not needed. However this optimization is not activated in this change, there is another issue opened to discuss how it should be integrated smoothly.
Some comments about the change:
* Tests that can produce negative scores have been adapted but we need to forbid them completely: #33309Closes#32899
When index sorting is enabled, toXContent tried to serialize an
SortField object, resulting in an exception, when using the _segments
endpoint.
Relates #29120
With this commit we use the classic parent circuit breaker which does
not account for real memory usage. In those tests we want to have
reproducible results and hence it makes sense to disable the real memory
circuit breaker there.
Christoph Büscher
Nhat Nguyen
Alan Woodward
Yannick Welsch
Nik Everett
Jake Landis
Simon Willnauer
David Turner
Tim Vernum
Daniel Mitterdorfer
Vladimir Dolzhenko
Jim Ferenczi
Adrien Grand
Ignacio Vera
Zachary Tong
Armin Braun
markharwood
Or Bin
Bukhtawar
Martijn van Groningen
Ryan Ernst
Jason Tedor
Alexander Reelsen
Igor Motov
lipsill
Colin Goodheart-Smithe
Tanguy Leroux
S.Y. Wang
Jim Ferenczi
Daniel Mitterdorfer