Exceptions thrown by the cluster applier service's settings and cluster appliers are bubbled up, and
block the state from being applied instead of silently being ignored. In combination with the cluster
state publishing lag detector, this will throw a node out of the cluster that can't properly apply
cluster state updates.
As acking can fail for any reason (unrelated node being too slow, node disconnecting), it should not
be required for acking to succeed in order for index requests with dynamic mapping updates to
successfully complete.
Relates to #30672 and Closes#30844
* Remove Custom Listeners from SnapshotsService
Motivations:
* Shorten the code some more
* Use ActionListener#wrap to get easy to reason about behavior in failure scenarios
* Remove duplication in the logic of handling snapshot completion listeners (listeners removing themselves and comparing snapshots to their targets)
* Also here, move all listener handling into `SnapshotsService` and remove custom listener class by putting listeners in a map
Today we support a smooth rolling upgrade from Zen1 to Zen2 by automatically
bootstrapping the cluster once all the Zen1 nodes have left, as long as the
`minimum_master_nodes` count is satisfied. However this means that Zen2 nodes
also require the `minimum_master_nodes` setting for this one specific and
transient situation.
Since nodes only perform this automatic bootstrapping if they previously
belonged to a Zen1 cluster, they can keep track of the `minimum_master_nodes`
setting from the previous master instead of requiring it to be set on the Zen2
node.
- Add deprecation warning to RestGetFieldMappingAction
- Add two new java HRLC classes GetFieldMappingsRequest and
GetFieldMappingsResponse. These classes use new typeless forms
of a request and response, and differ in that from the server
versions.
Relates to #35190
Currently we have the ability to listen for setting changes to two group
affix settings. However, it is possible that we might have the need to
listen to more than two. This commit adds a method that allows consumer
to listen to a list of affix settings for changes.
In some cases we only have a string collection instead of a string list
that we want to serialize out. We have a convenience method for writing
a list of strings, but no such method for writing a collection of
strings. Yet, a list of strings is a collection of strings, so we can
simply liberalize StreamOutput#writeStringList to be more generous in
the collections that it accepts and write out collections of strings
too. On the other side, we do not have a convenience method for reading
a list of strings. This commit addresses both of these issues.
* Add PersistentTasksClusterService::unassignPersistentTask method
* adding cancellation test
* Adding integration test for unallocating tasks from a node
* Addressing review comments
* adressing minor PR comments
Due to https://issues.apache.org/jira/browse/LUCENE-8634 this test
may fail if a really tiny polygon is generated. This commit checks for
tiny polygons and skips the final check, which is expected to fail
until the lucene bug is fixed and new version of lucene is released.
* This test compleletly overrode the `reroute` method and hence did nothing put test the override itself
* Removed the test since it tests nothing and simplified `reroute` accordingly
This commit moves the collectSearchShards method out of RemoteClusterService into TransportSearchAction that currently calls it. RemoteClusterService used to be used only for cross-cluster search but is now also used in cross-cluster replication where different API are called through the RemoteClusterAwareClient. There is no reason for the collectSearchShards and fetchShards methods to be respectively in RemoteClusterService and RemoteClusterConnection. The search shards API can be called through the RemoteClusterAwareClient too, the only missing bit is a way to handle failures based on the skip_unavailable setting for each cluster (currently only supported in RemoteClusterConnection#fetchShards) which is achieved by adding a isSkipUnavailable(String clusterAlias) method to RemoteClusterService.
This change is useful for #32125 as we will very soon need to also call the search API against remote clusters, which will be done through RemoteClusterAwareClient. In that case we will also need to support skip_unavailable when calling the search API so we need some way to handle the skip_unavailable setting like we currently do for the search_shards call.
Relates to #32125
Today we have several implementations of executing SearchOperationListener
in SearchService. While all of them seem to be safe at least on, the one that
executes scroll searches can cause illegal execution of SearchOperationListener
that can then in-turn trigger assertions in ShardSearchStats. This change
adds a SearchOperationListenerExecutor that uses try-with blocks to ensure
listeners are called in a safe way.
Relates to #37185
This commit moves the aggregation and mapping code from joda time to
java time. This includes field mappers, root object mappers, aggregations with date
histograms, query builders and a lot of changes within tests.
The cut-over to java time is a requirement so that we can support nanoseconds
properly in a future field mapper.
Relates #27330
Users may require the sequence number and primary terms to perform optimistic concurrency control operations. Currently, you can get the sequence number via the `docvalues_fields` API but the primary term is not accessible because it is maintained by the `SeqNoFieldMapper` and the infrastructure can't find it.
This commit adds a dedicated sub fetch phase to return both numbers that is connected to a new `seq_no_primary_term` parameter.
1. testSimpleOnlyMasterNodeElection - requires cluster bootstrap when
the first master node is started.
2. testElectOnlyBetweenMasterNodes - requires cluster bootstrap when
the first master node is started and requires adding voting exclusion
before shutting down the first master node.
3. testAliasFilterValidation - requires cluster bootstrap when the
first master node is started.
It's not safe to continue writing state using MetaDataStateFormat
after dirty WriteStateException occurred if it's not recovered by
successful subsequent state write.
We've encountered test failure of testFailRandomlyAndReadAnyState.
The test breaks in the following way. There are 3 state paths. And what
happens next
Successful write at the beginning of the test yields 0 0 0 state
files in the directories.
1st write in the loop is unsuccessful, but not dirty - 0 0 0.
2nd write in the loop is not successful and dirty (failure during
fsync), however before removing new files we have 1 1 1. But now during
deletion, the first deletion fails and we get - 1 0 0.
3rd write in the loop is unsuccessful, but not dirty - so we want to
keep old generation, which happens to be the 1st generation, so now we
have 1 x x in state folders. Now we assert that we either load 0 or 1
state from the state folders and select only 2rd and 3th folder to
emulate disk failures - this results in NPE because there is nothing in
these folders.
Fortunately, this won’t be a problem in real life, because if there is
a dirty exception, we shut down the node and make sure we perform a
successful write on the node startup.
This adds a set of helper classes to determine if an agg "has a value".
This is needed because InternalAggs represent "empty" in different
manners according to convention. Some use `NaN`, `+/- Inf`, `0.0`, etc.
A user can pass the Internal agg type to one of these helper methods
and it will report if the agg contains a value or not, which allows the
user to differentiate "empty" from a real `NaN`.
These helpers are best-effort in some cases. For example, several
pipeline aggs share a single return class but use different conventions
to mark "empty", so the helper uses the loosest definition that applies
to all the aggs that use the class.
Sums in particular are unreliable. The InternalSum simply returns 0.0
if the agg is empty (which is correct, no values == sum of zero). But this
also means the helper cannot differentiate from "empty" and `+1 + -1`.
This commit removes the warn-date from warning headers. Previously we
were stamping every warning header with when the request
occurred. However, this has a severe performance penalty when
deprecation logging is called frequently, as obtaining the current time
and formatting it properly is expensive. A previous change moved to
using the startup time as the time to stamp on every warning header, but
this was only to prove that the timestamping was expensive. Since the
warn-date is optional, we elect to remove it from the warning
header. Prior to this commit, we worked in Kibana to make the warn-date
treated as optional there so that we can follow-up in Elasticsearch and
remove the warn-date. This commit does that.
Prefer publishing to master-eligible nodes first, so that cluster state updates are committed more
quickly, and master-eligible nodes also turned more quickly into followers after a leader election.
PersistentTasksClusterService decides if a task should be reassigned by
checking there is a node in the cluster with the same Id. If a node is
restarted PersistentTasksClusterService may not observe the change and
decide the task still has a valid assignment because the node's
ephemeral Id is not used in that decision. This change un-assigns tasks
as the nodes in the cluster change.
* Fail start of non-data node if node has data
Check that nodes started with node.data=false cannot start if they have
shard data to avoid (old) indexes being resurrected into the cluster in red status.
Issue #27073
When publications were cancelled because a node turned to follower or candidate, it would still
show as time out, which can be confusing in the logs. This change adapts the improper call of
onTimeout by generalizing it to a cancel method.
The method calls "enabled" in addition to what the super.index() does, but this
seems to be done explicitely now in the TypeParsers `parse` method. The removed
method has been deprecated since at least 6.0. Also making some of the Builders
methods and ctos private since they are only used internally in this class.
Today when bootstrapping a Zen2 cluster we wait for every node in the
`initial_master_nodes` setting to be discovered, so that we can map the
node names or addresses in the `initial_master_nodes` list to their IDs for
inclusion in the initial voting configuration. This means that if any of
the expected master-eligible nodes fails to start then bootstrapping will
not occur and the cluster will not form. This is not ideal, and we would
prefer the cluster to bootstrap even if some of the master-eligible nodes
do not start.
Safe bootstrapping requires that all pairs of quorums of all initial
configurations overlap, and this is particularly troublesome to ensure
given that nodes may be concurrently and independently attempting to
bootstrap the cluster. The solution is to bootstrap using an initial
configuration whose size matches the size of the expected set of
master-eligible nodes, but with the unknown IDs replaced by "placeholder"
IDs that can never belong to any node. Any quorum of received votes in any
of these placeholder-laden initial configurations is also a quorum of the
"true" initial set of master-eligible nodes, giving the guarantee that it
intersects all other quorums as required.
Note that this change means that the initial configuration is not
necessarily robust to any node failures. Normally the cluster will form and
then auto-reconfigure to a more robust configuration in which the
placeholder IDs are replaced by the IDs of genuine nodes as they join the
cluster; however if a node fails between bootstrapping and this
auto-reconfiguration then the cluster may become unavailable. This we feel
to be less likely than a node failing to start at all.
This commit also enormously simplifies the cluster bootstrapping process.
Today, the cluster bootstrapping process involves two (local) transport actions
in order to support a flexible bootstrapping API and to make it easily
accessible to plugins. However this flexibility is not required for the current
design so it is adding a good deal of unnecessary complexity. Here we remove
this complexity in favour of a much simpler ClusterBootstrapService
implementation that does all the work itself.
In order to be able to parse epoch seconds and epoch milli seconds own
java time fields had been introduced. These fields are however not
compatible with the way that java time allows one to configure default
fields (when a part of a timestamp cannot be read then a default value
is added), which is used for the formatters that are rounding up to the
next value.
This commit allows java date formatters to configure its round up parsing
by setting default values via a consumer. By default all formats are setting
JavaDateFormatter.ROUND_UP_BASE_FIELDS for rounding up. The epoch
however parsers both need to set different fields. The merged date
formatters do not set any fields, they just append all the round up formatters.
Also the formatter now properly copies the locale and the timezone,
fractional parsing has been set to nano seconds with proper width.
Since version 6.7.0 the Close Index API guarantees that all translog
operations have been correctly flushed before the index is closed. If
the index is reopened as a Frozen index (which uses a ReadOnlyEngine)
we can verify that the maximum sequence number from the last Lucene
commit is indeed equal to the last known global checkpoint and refuses
to open the read only engine if it's not the case. In this PR the check is
only done for indices created on or after 6.7.0 as they are guaranteed
to be closed using the new Close Index API.
Related #33888
This commit introduces a NetworkMessage class. This class has two
subclasses - InboundMessage and OutboundMessage. These messages can
be serialized and deserialized independent of the transport. This allows
more granular testing. Additionally, the serialization mechanism is now
a simple Supplier. This builds the framework to eventually move the
serialization of transport messages to the network thread. This is the
one serialization component that is not currently performed on the
network thread (transport deserialization and http serialization and
deserialization are all on the network thread).
Currently we create dedicated network threads for both the http and
transport implementations. Since these these threads should never
perform blocking operations, these threads could be shared. This commit
modifies the nio-transport to have 0 http workers be default. If the
default configs are used, this will cause the http transport to be run
on the transport worker threads. The http worker setting will still exist
in case the user would like to configure dedicated workers. Additionally,
this commmit deletes dedicated acceptor threads. We have never had these
for the netty transport and they can be added back if a need is
determined in the future.
* The repo id was determined wrong when the delete picked up on an in progress snapshot
* NOTE: This solution is still a best-effort fix and there's a slight chance of running into concurrency issues here
when multiple create and delete requests for the same snapshot name are happening concurrently, but these require a sequence
of multiple cluster state updates between the changed method reading the genId and submitting its cluster state update task
* Added test reproduced the issue reliably in about 50% of runs
* Closes#37581
This will be used in cross-cluster search when reduction will be
performed locally on each cluster. The CCS coordinating node will send
one search request per remote cluster involved and will get one search
response back from each one of them. Such responses contain all the info
to be able to perform an additional reduction and return results back
to the user.
Relates to #32125
This commit optimizes some of the performance issues from using
deprecation logging:
- we optimize encoding the deprecation value
- we optimize formatting the deprecation string
- we optimize away getting the current time (by using cached startup
time)
To make further refactoring of GeoGrid aggregations
easier (related: #30320), splitting out these inner
class dependencies into their own files makes it
easier to map the relationship between classes
From #29453 and #37285, the `include_type_name` parameter was already present and defaulted to false. This PR makes the following updates:
- Add deprecation warnings to `RestPutMappingAction`, plus tests in `RestPutMappingActionTests`.
- Add a typeless 'put mappings' method to the Java HLRC, and deprecate the old typed version. To do this cleanly, I opted to create a new `PutMappingRequest` object that differs from the existing server one.
This adds deprecation to _type in the script contexts for ingest and update.
This adds a DeprecationMap that wraps the ctx Map containing _type for these
specific contexts.
Some tests (e.g. testRestoreIndexWithShardsMissingInLocalGateway) were split-braining since
being switched to Zen2 because the bootstrap setting was left around when nodes got restarted
with data folders wiped.
The test in question here was starting one node (which autobootstrapped to that single node), then
another node. The first node was then shut down (after excluding it from the voting configuration),
its data folder wiped, and restarted. After restart, the node had an empty data folder yet
initial_master_nodes set to itself (i.e. same name). This made the node sometimes form a cluster of
its own, and not rejoin the existing cluster with the other node.
Currently when adding a response header, we do some de-duplication, and
maybe drop the header on the floor if we have reached capacity. Yet, we
still update the thread local tracking the response headers. This is
really expensive because under the hood there is a shared reference that
we synchronize on. In the case of a request processed across many shards
in a tight loop, this contention can be detrimental to performance. We
can avoid updating the thread local in these cases though, when the
response header is duplicate of one that we have already seen, or when
it's dropped on the floor. This commit addresses these performance
issues by avoiding the unnecessary set.
As of today the Close Index API does its best to close indices,
but closing an index with ongoing recoveries might or might not
be acknowledged depending of the values of the max seq number
and global checkpoint at the time the
TransportVerifyShardBeforeClose action is executed.
These tests failed because they always expect that the index is
correctly closed on the first try, which is not always the case.
Instead we need to retry the closing until it succeed.
Closes#37571
* Fixes `testTwoNodeFirstNodeCleared` by manipulating voting config exclusions.
* Removes `testRecoveryDifferentNodeOrderStartup` since state recovery is now
handled entirely on the elected master, so the order in which the data nodes
start is irrelevant.
This test was actually passing, for the wrong reason: it asserts a
`MasterNotDiscoveredException` is thrown, expecting this to be due to a failure
to perform state recovery, but in fact it's thrown because the node is not
correctly bootstrapped.
This change adds deprecation warning to the indices.get_mapping API in case the
"inlcude_type_name" parameter is set to "true" and changes the parsing code in
GetMappingsResponse to parse the type-less response instead of the one
containing types. As a consequence the HLRC client doesn't need to force
"include_type_name=true" any more and the GetMappingsResponseTests can be
adapted to the new format as well. Also removing some "include_type_name"
parameters in yaml test and docs where not necessary.
delete and close index actions threw IllegalArgumentExceptions
when attempting to run against an index that has a snapshot
in progress.
This change introduces a dedicated SnapshotInProgressException
for these scenarios. This is done to explicitly signal to clients that
this is the reason the action failed, and it is a retryable error.
relates to #37541.
This is a continuation of #28667 and has as goal to convert all executors to propagate errors to the
uncaught exception handler. Notable missing ones were the direct executor and the scheduler. This
commit also makes it the property of the executor, not the runnable, to ensure this property. A big
part of this commit also consists of vastly improving the test coverage in this area.
This change adds a way to customize how phrase prefix queries should be created
on field types. The match phrase prefix query is exposed in field types in order
to allow optimizations based on the options set on the field.
For instance the text field uses the configured prefix field (if available) to
build a span near that mixes the original field and the prefix field on the last
position.
This change also contains a small refactoring of the match/multi_match query that
simplifies the interactions between the builders.
Closes#31921
This test failed because the refresh at the end of the test is not guaranteed to run before the
indexing is completed, and therefore there's no guarantee that the refresh will free all operations.
This triggers an assertion failure in the test clean-up, which asserts that there are no more pending
operations.
Some systems default to a nofile ulimit of 65535. To reduce the pain of
deploying Elasticsearch to such systems, this commit lowers the required
limit from 65536 to 65535.
We flush quite often in testAddNewReplicas to create the safe index
commit with gaps in sequence numbers. This test is failing recently
because CI is too slow to complete 5 small flushes in 10 seconds.
This commit increases timeout for this test and also ensures to always
terminate the background indexing. The latter is to eliminate unrelated
failures if this test fails again.
Closes#37183
All tests except testRestorePersistentSettings (renamed to
testExceptionWhenRestoringPersistentSettings) worked fine.
testExceptionWhenRestoringPersistentSettings re-written to use a custom
setting, because "minimum master node" setting is no longer available
in Zen2. It turns out there is no good replacement for "minimum master
node" setting for this test, that's why the custom setting is
introduced.
Unfortunately, there is #37485 bug and currently
RestoreService does not perform setting validation. That's why the
test is annotated with @AwaitsFix, the idea is to merge this commit and
then fix the issue and enable the test. (The test passes with a simple
fix, that adds a single line to RestoreService).
Currently all proxied actions are denied for the `SystemPrivilege`.
Unfortunately, there are use cases (CCR) where we would like to proxy
actions to a remote node that are normally performed by the
system context. This commit allows the system context to perform
proxy actions if they are actions that the system context is normally
allowed to execute.
Currently it takes a type, but this isn't really needed now that indices can
have at most one type. The only downside is that we might return a different
error when trying to index into a type that doesnt't exist yet.
This commit removes some leniency from REST handling where we move to
reject all requests that have a body where the body is not used during
the course of handling the request. For example,
DELETE /index
{
"query" : {
"term" : {
"field" : "value"
}
}
}
is now rejected.
* The internal create request is absolutely redundant, the only difference to the transport request is that we resolved the snapshot
name when moving from the transport to the internal version
* Removed it and passed the transport request into the snapshot service instead
* nicer way of resolve snapshot name in callback
The AbstracLifecycleComponent used to extend AbstractComponent, so it had to pass settings to the constractor of its supper class.
It no longer extends the AbstractComponent so there is no need for this constructor
There is also no need for AbstracLifecycleComponent subclasses to have Settings in their constructors if they were only passing it over to super constructor.
This is part 1. which will be backported to 6.x with a migration guide/deprecation log.
part 2 will have this constructor removed in 7
relates #35560
relates #34488
This commit adds one more underlying implementation of MockPersistedState.
Previously only InMemoryPersistentState was used, not GatewayMetaState
is used rarely.
When adding GatewayMetaState support the main question was: do we want to
emulate exceptions as we do today in MockPersistedState before
delegating to GatewayMetaState or do we want these exceptions to
propagate from the lower level, i.e. file system exceptions?
On the one hand, lower level exception propagation is already tested in
GatewayMetaStateTests, so this won't improve the coverage.
On the other hand, the benefit of low-level exceptions is to see how all these
components work in conjunction. Finally, we abandoned the idea of low-level
exceptions because we don't have a way to deal with IOError today in
CoordinatorTests, but hacking GatewayMetaState not to throw
IOError seems unnatural.
So MockPersistedState rarely throws an exception before delegating to
GatewayMetaState, which is not supposed to throw the exception.
This commit required two changes:
Move GatewayMetaStateUT to upper-level from
GatewayMetaStatePersistedStateTests, because otherwise, it's not easy
to construct GatewayMetaState instance in CoordinatorTests.
Move addition of STATE_NOT_RECOVERED_BLOCK from GatewayMetaState
constructor to GatewayMetaState.applyClusterUpdaters, because
CoordinatorTests class assumes that there is no such block and most of
them fail.
The completion suggester ignores the original weight of the suggestion when duplicates are removed. This change fixes this bug and keeps the best weighted suggestion among the duplicates. It also removes the custom implementation of the top docs suggest collector now that https://issues.apache.org/jira/browse/LUCENE-8529 is committed in Lucene.
Closes#35836
This commit prepares the required infra to make send a translog snapshot
of the recovery source non-blocking. I'll make a follow-up to make the send
snapshot method non-blocking.
Relates #37291
There were 5 tests in MinimumMasterNodesIT. 2 of them removed, 3 of
them changed and renamed.
1) testSimpleMinimumMasterNodes -> testTwoNodesNoMasterBlock. The
flow of this test is left intact but in order to make it work on
Zen2, additional work for the cluster bootstrapping and voting
exclusions is needed.
2) testDynamicUpdateMinimumMasterNodes -> removed, there is nothing
that corresponds to the dynamic change of the minimum master nodes
setting.
3) testCanNotBringClusterDown -> removed, it also plays with changing
minimum master nodes dynamically.
4) testMultipleNodesShutdownNonMasterNodes ->
testThreeNodesNoMasterBlock. Previously this test was checking that
there would be no master block, if min_master_nodes=3 and 4 nodes are
started, then 2 nodes are brought down. Zen2 dynamically accommodates
to the number of nodes in the cluster, so it's possible that there
still will be a master in 2 nodes cluster. For Zen2, we start up 3
nodes. And shut down 2 of them (w/o voting exclusions), which results
in no master block.
5) testCanNotPublishWithoutMinMastNodes ->
testCanNotCommitStateThreeNodes. Test flow is not changed. But
previously there was no check that nodes in the bigger part of
network partition will elect the master, before healing the network
partition. For Zen2 it does not work, because persistent setting
addition is accepted on the old master and if it's elected new master
again, this setting will appear in the cluster state.
Also, I have a feeling that we need to remove this class, but could not
come up with a good name.
Adds the node's current term and the term and version of the the last-accepted
cluster state to the message reported by the `ClusterFormationFailureHelper`,
since these values may be of importance when tracking down a cluster formation
failure.
The test that remote clusters used by ML datafeeds have
a license that allows ML was not accounting for the
possibility that the remote cluster name could be
wildcarded. This change fixes that omission.
Fixes#36228
The test testSendSnapshotSendsOps is currently using a mock instance of
RecoveryTargetHandler which will be hard to modify when we make the
RecoveryTargetHandler non-blocking. This commit prepares for the
incoming changes by replacing the mock instance with a stub.
The Inject Annotation was removed from IndicesClusterStateService as
part of reformatting in e11a32e, but this causes CreationException on
cluster startup.
This commit adds a simple method for executing a runnable against a
shard under a primary permit. Today there is only a single caller for
this method, but this there are two upcoming use-cases for which having
this method will help keep the code simpler.
This commit reformats some classes in the index universe with the
purpose of breaking some long method definitions and invocations into a
line per parameter. This has the advantage that for an upcoming change
to these definitions and invocations, the diff for that change will be a
single line per definition or invocation. That makes these sorts of
changes easier to read.
This commit is a simple introduction of the serialization of retention
leases, which will be needed when they are sent across the wire while
synchronizing retention leases to replicas.
* Default include_type_name to false for get and put mappings.
* Default include_type_name to false for get field mappings.
* Add a constant for the default include_type_name value.
* Default include_type_name to false for get and put index templates.
* Default include_type_name to false for create index.
* Update create index calls in REST documentation to use include_type_name=true.
* Some minor clean-ups around the get index API.
* In REST tests, use include_type_name=true by default for index creation.
* Make sure to use 'expression == false'.
* Clarify the different IndexTemplateMetaData toXContent methods.
* Fix FullClusterRestartIT#testSnapshotRestore.
* Fix the ml_anomalies_default_mappings test.
* Fix GetFieldMappingsResponseTests and GetIndexTemplateResponseTests.
We make sure to specify include_type_name=true during xContent parsing,
so we continue to test the legacy typed responses. XContent generation
for the typeless responses is currently only covered by REST tests,
but we will be adding unit test coverage for these as we implement
each typeless API in the Java HLRC.
This commit also refactors GetMappingsResponse to follow the same appraoch
as the other mappings-related responses, where we read include_type_name
out of the xContent params, instead of creating a second toXContent method.
This gives better consistency in the response parsing code.
* Fix more REST tests.
* Improve some wording in the create index documentation.
* Add a note about types removal in the create index docs.
* Fix SmokeTestMonitoringWithSecurityIT#testHTTPExporterWithSSL.
* Make sure to mention include_type_name in the REST docs for affected APIs.
* Make sure to use 'expression == false' in FullClusterRestartIT.
* Mention include_type_name in the REST templates docs.
Today file-chunks are sent sequentially one by one in peer-recovery. This is a
correct choice since the implementation is straightforward and recovery is
network bound in most of the time. However, if the connection is encrypted, we
might not be able to saturate the network pipe because encrypting/decrypting
are cpu bound rather than network-bound.
With this commit, a source node can send multiple (default to 2) file-chunks
without waiting for the acknowledgments from the target.
Below are the benchmark results for PMC and NYC_taxis.
- PMC (20.2 GB)
| Transport | Baseline | chunks=1 | chunks=2 | chunks=3 | chunks=4 |
| ----------| ---------| -------- | -------- | -------- | -------- |
| Plain | 184s | 137s | 106s | 105s | 106s |
| TLS | 346s | 294s | 176s | 153s | 117s |
| Compress | 1556s | 1407s | 1193s | 1183s | 1211s |
- NYC_Taxis (38.6GB)
| Transport | Baseline | chunks=1 | chunks=2 | chunks=3 | chunks=4 |
| ----------| ---------| ---------| ---------| ---------| -------- |
| Plain | 321s | 249s | 191s | * | * |
| TLS | 618s | 539s | 323s | 290s | 213s |
| Compress | 2622s | 2421s | 2018s | 2029s | n/a |
Relates #33844
This is related to #35975. It implements a file based restore in the
CcrRepository. The restore transfers files from the leader cluster
to the follower cluster. It does not implement any advanced resiliency
features at the moment. Any request failure will end the restore.
Without pulling out the supplier function to the enclosing class, Eclipse 4.8
complains with the following error "No enclosing instance of type
CoordinatorTests.Cluster is available due to some intermediate constructor
invocation"
DeprecationLogger has warning de-duplication logic but it is expensive to run as it involves parsing existing warning headers. This PR changes the upstream bulk indexing code to do its own "event thinning" rather than relying on DeprecationLogger's trimming.
Closes#37411
The constructors in PutPipelineRequest and SimulatePipelineRequest that guess
the xContent type from the provided source are deprecated since 6.0 and each
have a counterpart that takes the xContent type as an explicit argument.
Removing these ctors together with the builders and methods in
ClusterAdminClient that don't have the xContent type as argument.
Today the SyncedFlushService flow is written with multiple nested
callbacks which are hard to read. This commit replaces them with
sequential step listeners.
We recently migrated suggestions to `Writeable`. That allows us to also
clean up empty constructors and methods that called them as they are no
longer needed. They are replaced by constructors that accept a
`StreamInput` instance.
Today a peer-recovery may run into a deadlock if the value of
node_concurrent_recoveries is too high. This happens because the
peer-recovery is executed in a blocking fashion. This commit attempts
to make the recovery source partially non-blocking. I will make three
follow-ups to make it fully non-blocking: (1) send translog operations,
(2) primary relocation, (3) send commit files.
Relates #36195
* Fix PrimaryAllocationIT Race Condition
* Forcing a stale primary allocation on a green index was tripping the assertion that was removed
* Added a test that this case still errors out correctly
* Made the ability to wipe stopped datanode's data public on the internal test cluster and used it to ensure correct behaviour on the fixed test
* Previously it simply passed because the test finished before the index went green and would NPE when the index was green at the time of the shard store status request, that would then come up empty
* Closes#37345
This commit introduces StepListener which provides a simple way to write
a flow consisting of multiple asynchronous steps without having nested
callbacks.
Relates #37291
With the `include_type_name` available now for indices.get on 6.x after the
backport, the corresponsing yaml test can include anything from 6.7 on.
Also changing the RestGetIndicesActionTests base test class.
* Due to a race between retrying the snapshot creation and the failed snapshot create trying to delete the snapshot there is no guarantee that the snapshot is eventually created by retries
* Adjusted the assertion accordingly
* Closes#36779
This commit moves DisruptableMockTransport to use a more accurate representation of connection
management, which allows to use the full connection manager and does not require mocking out
any behavior. With this, we can implement restarting nodes in CoordinatorTests.
Updates perform realtime get, perform the requested update and then index the document again
using optimistic concurrency control. This PR changes the logic to use sequence numbers instead
of versioning.
Note that the current versioning logic isn't suffering from the same problem as external OCC
requests because the get and indexing is always done on the same primary.
Relates #36148
Relates #10708
* Add benchmark
* Use java time API instead of exception handling
when several formatters are used, the existing way of parsing those is
to throw an exception catch it, and try the next one. This is is
considerably slower than the approach taken in joda time, so that
indexing is reduced when a date format like `x||y` is used and y is the
date format being used.
This commit now uses the java API to parse the date by appending the
date time formatters to each other and does not rely on exception
handling.
* fix benchmark
* fix tests by changing formatter, also expose printer
* restore optional printing logic to fix tests
* fix tests
* incorporate review comments
This commit makes the use of empty retention lease suppliers to always
be an empty list as opposed to in some cases an empty set. This commit
is solely for consistency reasons, there is no functional change here.
This commit adds some simple validation that the values input to the
retention lease constructor our valid values. We will later rely on
these values being within the validated range.
Added warnings checks to existing tests
Added “defaultTypeIfNull” to DocWriteRequest interface so that Bulk requests can override a null choice of document type with any global custom choice.
Related to #35190
* Add include_type_name to the get field mappings API.
* Make sure the API specification lists include_type_name as a boolean.
* Add include_type_name to the get index templates API.
* Add include_type_name to the put index templates API.
When the deprecation log is written to within scripting support code
like ScriptDocValues, it runs under the reduces privileges of scripts.
Sometimes this can trigger log rolling, which then causes uncaught
security errors, as was handled in #28485. While doing individual
deprecation handling within each deprecation scripting location is
possible, there are a growing number of deprecations in scripts.
This commit wraps the logging call within the deprecation logger use a
doPrivileged block, just was we would within individual logging call
sites for scripting utilities.
* Get indices shard store status before enqueuing the reallocation state update task to prevent
tasks that would fail because a node does not hold a stale copy of the shard on a best effort basis
* Closes#37098
Adds join validation to Zen2, which prevents a node from joining a cluster when the node does not
have the right ES version or does not satisfy any other of the join validation constraints.
This SearchType was deprecated since at least 6.0 and according to the
documentation is only kept around for pre-5.3 requests. Removing and leaving a
comment as placeholder so we don't reuse the byte value associated with it
without further consideration.
* Java Time: Fix timezone parsing
An independent test uncovered an issue when parsing a timezone
containing a colon like `01:00` - some formats did not properly support
this.
This commit adds test for all formats in the dueling tests and fixes a
few issues with existing date formatters.
* fix tests, so they run under java8
* Tests: Add ElasticsearchAssertions.awaitLatch method
Some tests are using assertTrue(latch.await(...)) in their code. This
leads to an assertion error without any error message. This adds a
method which has a nicer error message and can be used in tests.
* fix forbidden apis
* fix spaces
Today we still wrap recovery source readers on merge even if we
keep all documents recovery source. This basically disables bulk
merging for stored fields. This change skips wrapping if all docs
sources are kept anyway.
This change fixes an unreleased bug that trips an assertion because a static instance
shared among threads is modified during the search. This commit copies the static
instance in order to ensure that each thread can modify the value without modifying
the other instances.
Closes#37179Closes#37266
* ingest: compile mustache template only if field includes '{{''
Prior to this change, any field in an ingest node processor that supports
script templates would be compiled as mustache template regardless if they
contain a template or not. Compiling normal text as mustache templates is
harmless. However, each compilation counts against the script compilation
circuit breaker. A large number of processors without any templates or scripts
could un-intuitively trip the too many script compilations circuit breaker.
This change simple checks for '{{' in the text before it attempts to compile.
fixes#37120
Update BucketSortPipelineAggregator to use a List and Collections.sort() for sorting instead of a priority queue. This preserves the order for equal values. Closes#36322.
* TESTS: Real Coordinator in SnapshotServiceTests
* Introduce real coordinator in SnapshotServiceTests to be able to test network disruptions realistically
* Make adjustments to cluster applier service so that we can pass a mocked single threaded executor for tests
This change adds support for the 'include_type_name' parameter for the
indices.get API. This parameter, which defaults to `false` starting in 7.0,
changes the response to not include the indices type names any longer.
If the parameter is set in the request, we additionally emit a deprecation
warning since using the parameter should be only temporarily necessary while
adapting to the new response format and we will remove it with the next major
version.
This change turns an assertion into an IllegalStateException in SearchPhaseController#getTotalHits.
The goal is to help identify the cause of the failures in https://github.com/elastic/elasticsearch/issues/37179
which seems to fail only in CI.
The assertion will be restored when the issue is solved (NORELEASE).
The test intercepts TransportVerifyShardBeforeCloseAction shard
requests, so it needs a minimum of 2 primary shards on 2 different
nodes to correctly intercepts requests.
These tests failed on CI multiple times in the past weeks because they use a
test cluster with a SUITE scope that recreates nodes between tests. With such
a scope, nodes can be recreated in between test executions and can inherit a
node id from a previous test execution, while they are assigned a random data
path. With the successive node recreations it is possible that a newly recreated
node shares the same node id (but different data path) as a non recreated node.
This commit changes the cluster scope of the CorruptedFileIT and FlushIT
tests which often fail.
The failure is reproducable with :
./gradlew :server:integTest -Dtests.seed=EF3A50C225CF377
-Dtests.class=org.elasticsearch.index.store.CorruptedFileIT
-Dtests.security.manager=true -Dtests.locale=th-TH-u-nu-thai-x-lvariant-TH -Dtests.timezone=America/Rio_Branco
-Dcompiler.java=11 -Druntime.java=8
* [Analysis] Deprecate Standard Html Strip Analyzer
Deprecate only Standard Html Strip Analyzer
If user create index with the analyzer since 7.0, es throws an exception.
If an index was created before 7.0, es issue deprecation log
We will remove it in 8.0
Related #4704
Today we create a global instance of RecoveryResponse then mutate it
when executing each recovery step. This is okay for the current
sequential recovery flow but not suitable for an asynchronous recovery
which we are targeting. With this commit, we return the result of each
step separately, then construct a RecoveryResponse at the end.
Relates #37174
Traditionally remote clusters can be configured dynamically. However,
the compress and ping settings are not currently set to be configured
dynamically. This commit changes that.
This commit implements a straightforward approach to retention lease
expiration. Namely, we inspect which leases are expired when obtaining
the current leases through the replication tracker. At that moment, we
clean the map that persists the retention leases in memory.
This commit converts the epoch time parsing implementation which uses
the java time api to create DateTimeFormatters instead of DateFormatter
implementations. This will allow multi formats for java time to be
implemented in a single DateTimeFormatter in a future change.
Remove several unused helper methods. Most of them are one-liners and should
be easier to be used from the corresponding primitive wrapper classes.
The bytes array conversion methods are unused as well, it should be easy to
re-create them if needed.
This commit fixes an issue with a settings builder method that allows
setting a duration by time unit. In particular, this method can suffer
from a loss of precision. For example, if the input duration is 1500
microseconds then internally we are converting this to "1ms",
demonstrating the loss of precision. Instead, we should internally
convert this to a TimeValue that correctly represents the input
duration, and then convert this to a string using a method that does not
lose the unit. That is what this commit does.
Version needs to be updated after backporting #36997 & #37142
where we added support for providing and serializing localClusterAlias
as well ass absoluteStartMillis.
Relates to #36997 & #37142
Today, a setting can declare that its validity depends on the values of other
related settings. However, the validity of a setting is not always checked
against the correct values of its dependent settings because those settings'
correct values may not be available when the validator runs.
This commit separates the validation of a settings updates into two phases,
with separate methods on the `Setting.Validator` interface. In the first phase
the setting's validity is checked in isolation, and in the second phase it is
checked again against the values of its related settings. Most settings only
use the first phase, and only the few settings with dependencies make use of
the second phase.
The `cluster.unsafe_initial_master_node_count` setting was introduced as a
temporary measure while the design of `cluster.initial_master_nodes` was being
finalised. This commit removes this temporary setting, replacing it with usages
of `cluster.initial_master_nodes` where appropriate.
This commit adds a unique id to cluster blocks, so that they can be uniquely
identified if needed. This is important for the Close Index API where multiple
concurrent closing requests can be executed at the same time. By adding a
UUID to the cluster block, we can generate unique "closing block" that can
later be verified on shards and then checked again from the cluster state
before closing the index. When the verification on shard is done, the closing
block is replaced by the regular INDEX_CLOSED_BLOCK instance.
If something goes wrong, calling the Open Index API will remove the block.
Related to #33888
This commit is the first in a series which will culminate with
fully-functional shard history retention leases.
Shard history retention leases are aimed at preventing shard history
consumers from having to fallback to expensive file copy operations if
shard history is not available from a certain point. These consumers
include following indices in cross-cluster replication, and local shard
recoveries. A future consumer will be the changes API.
Further, index lifecycle management requires coordinating with some of
these consumers otherwise it could remove the source before all
consumers have finished reading all operations. The notion of shard
history retention leases that we are introducing here will also be used
to address this problem.
Shard history retention leases are a property of the replication group
managed under the authority of the primary. A shard history retention
lease is a combination of an identifier, a retaining sequence number, a
timestamp indicating when the lease was acquired or renewed, and a
string indicating the source of the lease. Being leases they have a
limited lifespan that will expire if not renewed. The idea of these
leases is that all operations above the minimum of all retaining
sequence numbers will be retained during merges (which would otherwise
clear away operations that are soft deleted). These leases will be
periodically persisted to Lucene and restored during recovery, and
broadcast to replicas under certain circumstances.
This commit is merely putting the basics in place. This first commit
only introduces the concept and integrates their use with the soft
delete retention policy. We add some tests to demonstrate the basic
management is correct, and that the soft delete policy is correctly
influenced by the existence of any retention leases. We make no effort
in this commit to implement any of the following:
- timestamps
- expiration
- persistence to and recovery from Lucene
- handoff during primary relocation
- sharing retention leases with replicas
- exposing leases in shard-level statistics
- integration with cross-cluster replication
These will occur individually in follow-up commits.
This commit addresses an issue when setting a byte size value setting
using a value that has a fractional component when converted to its
string representation. For example, trying to set a byte size value
setting to a value of 1536 bytes is problematic because internally this
is converted to the string "1.5k". When we go to get this setting, we
try to parse "1.5k" back to a byte size value, which does not support
fractional values. The problem is that internally we are relying on a
method which loses the unit when doing the string conversion. Instead,
we are going to use a method that does not lose the unit and therefore
we can roundtrip from the byte size value to the string and back to the
byte size value.
* Randomly doing non-atomic writes causes rare 0 byte reads from `index-N` files in tests
* Removing this randomness fixes these random failures and is valid because it does not reproduce a real-world failure-mode:
* Cloud-based Blob stores (S3, GCS, and Azure) do not have inconsistent partial reads of a blob, either you read a complete blob or nothing on them
* For file system based blob stores the atomic move we do (to atomically write a file) by setting `java.nio.file.StandardCopyOption#ATOMIC_MOVE` would throw if the file system does not provide for atomic moves
* Closes#37005
I started referring to this parameter name from various places in #37149 so I
think it's a good idea to simplify things by referring to a common constant.
We have recently added support for providing a local cluster alias to a
SearchRequest through a package protected constructor. When executing
cross-cluster search requests with local reduction on each cluster, the
CCS coordinating node will have to provide such cluster alias to each
remote cluster, as well as the absolute start time of the search action
in milliseconds from the time epoch, to be used when evaluating date
math expressions both while executing queries / scripts as well as when
resolving index names.
This commit adds support for providing the start time together with the
cluster alias. It is a final member in the search request, which will
only be set when using cross-cluster search with local reduction (also
known as alternate execution mode). When not provided, the coordinating
node will determine the current time and pass it through (by calling
`System.currentTimeMillis`).
Relates to #32125
This pull request changes the Freeze Index and Close Index actions so
that these actions always requires a Task. The task's id is then propagated
from the Freeze action to the Close action, and then to the Verify shard action.
This way it is possible to track which Freeze task initiates the closing of an index,
and which consecutive verifiy shard are executed for the index closing.
As suggested in #36775, this pull request renames the following methods:
ClusterBlocks.hasGlobalBlock(int)
ClusterBlocks.hasGlobalBlock(RestStatus)
ClusterBlocks.hasGlobalBlock(ClusterBlockLevel)
to something that better reflects the property of the ClusterBlock that is searched for:
ClusterBlocks.hasGlobalBlockWithId(int)
ClusterBlocks.hasGlobalBlockWithStatus(RestStatus)
ClusterBlocks.hasGlobalBlockWithLevel(ClusterBlockLevel)
This commit addresses an issue when setting a time value setting using a
value that has a fractional component when converted to its string
representation. For example, trying to set a time value setting to a
value of 1500ms is problematic because internally this is converted to
the string "1.5s". When we go to get this setting, we try to parse
"1.5s" back to a time value, which does not support fractional
values. The problem is that internally we are relying on a method which
loses the unit when doing the string conversion. Instead, we are going
to use a method that does not lose the unit and therefore we can
roundtrip from the time value to the string and back to the time value.
* The retries on the failing master can lead to concurrently trying to create and delete a snapshot, catch this for now to fix this test
* closes#36779
We don't want two FSDirectories manage pending deletes separately
and optimize file listing. This confuses IndexWriter and causes exceptions
when files are deleted twice but are pending for deletion. This change
move to using a NIOFS subclass that only delegates to MMAP for opening files
all metadata and pending deletes are managed on top.
Closes#37111
Relates to #36668
The query object was incorrectly added to the remote object in the
xcontent. This fix moves the query back into the source, if it was
passed in as part of the RemoteInfo. It also adds a IPv6 test for
reindex from remote such that we can properly validate this.
In Lucene 8 searches can skip non-competitive hits if the total hit count is not requested.
It is also possible to track the number of hits up to a certain threshold. This is a trade off to speed up searches while still being able to know a lower bound of the total hit count. This change adds the ability to set this threshold directly in the track_total_hits search option. A boolean value (true, false) indicates whether the total hit count should be tracked in the response. When set as an integer this option allows to compute a lower bound of the total hits while preserving the ability to skip non-competitive hits when enough matches have been collected.
Relates #33028
Today we block using the generic thread-pool on the target side
until the source side has fully executed the recovery. We still
block on the source side executing the recovery in a blocking fashion
but there is no reason to block on the target side. This will
release generic threads early if there are many concurrent recoveries
happen.
Relates to #36195
Today it's very difficult to see which indices are frozen or rather
throttled via the commonly used monitoring APIs. This change adds
a cell to the `_cat/indices` API to render if an index is `search.throttled`
Relates to #34352
With #36997 we added support for providing a local cluster alias with a
`SearchRequest`. We intended to make sure that when provided as part of
a search request, the cluster alias would never be used for connection
lookups. Yet due to a bug we would still end up looking up the
connection from the remote ones.
This commit adds a test to make sure that whenever we set the cluster
alias to the `SearchRequest` (which can only be done at transport), such
alias is used as index prefix in the returned hits. No errors are thrown
despite no remote clusters are configured indicating that such alias is
never used for connection look-ups.
Also, we add explicit support for the empty cluster alias when printing
out index names through `RemoteClusterAware#buildRemoteIndexName`.
In fact we don't want to print out `:index` when the cluster alias is
set to empty string, but rather `index`. Yet, the semantic of empty
string is different compared to `null` as it will still disable final
reduction. This will be used in CCS when searching against remote
clusters as well as the local one, the local one will have empty prefix
yet it will need to disable final reduction so that its results will be
properly merged with the ones coming from the remote clusters.
Today when electing a master in Zen2 we use the cluster state version to
determine whether a node has a fresh-enough cluster state to become master.
However the cluster state version is not a reliable measure of freshness in the
Zen1 world; furthermore in 6.x the cluster state version is not persisted. This
means that when upgrading from 6.x via a full cluster restart a cluster state
update may be lost if a stale master wins the initial election.
This change fixes this by using the metadata version as a measure of freshness
when in term 0, since this is persisted in 6.x and does more reliably indicate
the freshness of nodes.
It also makes changes parallel to elastic/elasticsearch-formal-models#40 to
support situations in which nodes accept cluster state versions in term 0: this
does not happen in a pure Zen2 cluster, but can happen in mixed clusters and
during upgrades.
Now that we unwrap mappings in DocumentMapperParser#extractMappings, it is not
necessary for the mapping definition to always be nested under the type. This
leniency around the mapping format was added in 2341825358.
* The static threadpool leaks a lot of memory in these tests because it prevents things like the connect listeners from `org.elasticsearch.transport.TcpTransport#initiateConnection`
to be GCed between tests (since they keep being referenced by the threadpool) which in turn reference channels and their underlying buffers
* I could not find any slowdown in executing these tests from this change, if anything they are slightly faster now on my machine
* Relates #36906 (which may be caused by slowness from leaking memory and also becomes testable in a loop by this change)
Types can be used both in the source and dest section of the body which will
be translated to search and index requests respectively. Adding a deprecation warning
for those cases and removing examples using more than one type in reindex since
support for this is going to be removed.
The `composite` aggregation uses a TreeMap to keep track of the best buckets.
This ensures a log(n) time cost to insert new buckets but also to retrieve buckets
that are already present in the map. In order to speed up the retrieval of buckets
this change replaces the TreeMap with a priority queue and a HashMap. The insertion
cost is still log(n) but the retrieval of buckets through the HashMap is now done in constant
time. This optimization can bring significant improvement since each document needs
to check if its associated buckets are already present in the current best buckets.
With this commit we rename `node.store.allow_mmapfs` to
`node.store.allow_mmap`. Previously this setting has controlled whether
`mmapfs` could be used as a store type. With the introduction of
`hybridfs` which also relies on memory-mapping,
`node.store.allow_mmapfs` also applies to `hybridfs` and thus we rename
it in order to convey that it is actually used to allow memory-mapping
but not a specific store type.
Relates #36668
Relates #37070
The QueryStringQueryBuilder does not currently delegate to the field mapper's prefixQuery
method, so does not use indexed prefixes. This commit corrects this.
It also fixes a bug where a query a* would not match the word a if indexed prefixes were used with
a minchar setting of 2.
When executing terms aggregations we set the shard_size, meaning the
number of buckets to collect on each shard, to a value that's higher than
the number of requested buckets, to guarantee some basic level of
precision. We have an optimization in place so that we leave shard_size
set to size whenever we are searching against a single shard, in which
case maximum precision is guaranteed by definition.
Such optimization requires us access to the total number of shards that
the search is executing against. In the context of cross-cluster search,
once we will introduce multiple reduction steps (one per cluster) each
cluster will only know the number of local shards, which is problematic
as we should only optimize if we are searching against a single shard in a
single cluster. It could be that we are searching against one shard per cluster
in which case the current code would optimize number of terms causing
a loss of precision.
While discussing how to address the CCS scenario, we decided that we do
not want to introduce further complexity caused by this single shard
optimization, as it benefits only a minority of cases, especially when
the benefits are not so great.
This commit removes the single shard optimization, meaning that we will
always have heuristic enabled on how many number of buckets to collect
on the shards, even when searching against a single shard.
This will cause more buckets to be collected when searching against a single
shard compared to before. If that becomes a problem for some users, they
can work around that by setting the shard_size equal to the size.
Relates to #32125
With #36221 we introduced shards counting to address a rare failure.
This caused a worse problem in this test when replicas were allocated
and shards failures were randomly returned. The latch has to take into
account additional attempts caused by the shard failures, which means
that in order for run to be called, performPhaseOnShard will be called
(numShards + numFailures) times.
To address this, we need to decide upfront which shard is going to fail,
making sure that at least one shards is successful otherwise the whole
request fails.
Closes#37074
With this commit we introduce a new store type `hybridfs` that is a
hybrid between `mmapfs` and `niofs`. This store type chooses different
strategies to read Lucene files based on the read access pattern (random
or linear) in order to optimize performance.
This store type has been available in earlier versions of Elasticsearch
as `default_fs`. We have chosen a different name now in order to convey
the intent of the store type instead of tying it to the fact whether it
is the default choice.
Relates #36668
SearchAsyncActionTests may fail with RejectedExecutionException as InitialSearchPhase may try to execute a runnable after the test has successfully completed, and the corresponding executor was already shut down. The latch was located in getNextPhase that is almost correct, but does not cover for the last finishAndRunNext round that gets executed after onShardResult is invoked.
This commit moves the latch to count the number of shards, and allowing the test to count down later, after finishAndRunNext has been potentially forked. This way nothing else will be executed once the executor is shut down at the end of the tests.
Closes#36221Closes#33699
* Fixes the issue reproduced in the added tests:
* When having open index requests on a shard that are waiting for a refresh, relocating that shard
becomes blocked until that refresh happens (which could be never as in the test scenario).
With #36997 we added the ability to provide a cluster alias with a
SearchRequest.
The next step is to disable the final reduction whenever a cluster alias
is provided with the SearchRequest. A cluster alias will be provided
when executing a cross-cluster search request with alternate execution
mode, where each cluster does its own reduction locally. In order for
the CCS node to be able to later perform an additional reduction of the
results, we need to make sure that all the needed info stays available.
This means that terms aggregations can be reduced but not pruned, and
pipeline aggs should not be executed. The final reduction will happen
later in the CCS coordinating node.
Relates to #36997 & #32125
With the upcoming cross-cluster search alternate execution mode, the CCS
node will be able to split a CCS request into multiple search requests,
one per remote cluster involved. In order to do that, the CCS node has
to be able to signal to each remote cluster that such sub-requests are
part of a CCS request. Each cluster does not know about the other
clusters involved, and does not know either what alias it is given in
the CCS node, hence the CCS coordinating node needs to be able to provide
the alias as part of the search request so that it is used as index prefix
in the returned search hits.
The cluster alias is a notion that's already supported in the search shards
iterator and search shard target, but it is currently used in CCS as both
index prefix and connection lookup key when fanning out to all the shards.
With CCS alternate execution mode the provided cluster alias needs to be
used only as index prefix, as shards are local to each cluster hence no
cluster alias should be used for connection lookups.
The local cluster alias can be set to the SearchRequest at the transport layer
only, and its constructor/getter methods are package private.
Relates to #32125
Improves on #36449 which did not cover the situation where a node had bumped its term during
the election, and not when receiving the first follower check. This was uncovered while refactoring
NodeJoinTests so that they don't need to access to an internal field of Coordinator anymore (which
can now be made private).
* This speeds up the test from an average 25s down to 7s runtime
* There is no need for artificially slowing down the snapshot to reproduce the issue of an out of sync routing table in practice.
Over hundreds of test runs the test's snapshot shard service still runs in the index not found exception every time reproducing this issue.
* Relates #36294
Keys are compared in BucketSortPipelineAggregation so making key type (ArrayMap) implement Comparable. Maps are compared using the entry set's iterator so ordered maps order is maintain. For each entry first comparing key then value. Assuming all keys are strings. When comparing entries' values if type is not identical and\or type not implementing Comparable, throwing exception. Not implementing equals() and hashCode() functions as parent's ones are sufficient. Tests included.
Today the routing of a SourceToParse is assigned in a separate step
after the object is created. We can easily forget to set the routing.
With this commit, the routing must be provided in the constructor of
SourceToParse.
Relates #36921
We introduce a typeless API in #35790 where we translate the default
docType "_doc" to the user-defined docType. However, we do not rewrite
the SourceToParse with the resolved docType. This leads to a situation
where we have two translog operations for the same document with
different types:
- prvOp [Index{id='9LCpwGcBkJN7eZxaB54L', type='_doc', seqNo=1,
primaryTerm=1, version=1, autoGeneratedIdTimestamp=1545125562123}]
- newOp [Index{id='9LCpwGcBkJN7eZxaB54L', type='not_doc', seqNo=1,
primaryTerm=1, version=1, autoGeneratedIdTimestamp=-1}]
Closes#36769
In this test, we verify that the LocalCheckpointTracker is initialized
with the operations of the safe commit. And the test fails because
Engine#Index does not implement the equals method (should not
implement as it consists of a mutable ParsedDocument).
Closes#36470
This is a follow-up to some discussions around #36399. Currently we have
relatively confusing compression behavior where compression can be
configured for requests based on transport.compress or a specific
setting for a remote cluster. However, we can only compress responses
based on transport.compress as we do not know where a request is
coming from (currently).
This commit modifies the behavior to NEVER compress responses based on
settings. Instead, a response will only be compressed if the request was
compressed. This commit also updates the documentation to more clearly
described transport level compression.
When the script contexts were created in 6, the use of params.ctx was
deprecated. This commit cleans up that code and ensures that params.ctx
is null in both watcher script contexts.
Relates: #34059
This commit adds a RemoteClusterAwareRequest interface that allows a
request to specify which remote node it should be routed to. The remote
cluster aware client will attempt to route the request directly to this
node. Otherwise it will send it as a proxy action to eventually end up
on the requested node.
It implements the ccr clean_session action with this client.
* Adds term number and greppable phrase 'coordinator becoming' to Coordinator
mode changes
* Adds term and version to messages from the ClusterApplier about master
changes
* Reduces some LeaderChecker messages to TRACE level
This commit modifies ESSingleNodeTestCase and ESIntegTestCase and
several concrete test classes to use node names when bootstrapping the
cluster.
Today ClusterBootstrapService.INITIAL_MASTER_NODE_COUNT_SETTING
setting is used to bootstrap clusters in tests. Instead, we want to use
ClusterBootrstapService.INITIAL_MASTER_NODES_SETTING and get rid of
the former setting eventually.
There were two main problems when refactoring InternalTestCluster:
1. Nodes are created one-by-one in buildNode method. And node.name
is created in this method as well. It's not suitable for bootstrapping,
because we need to have the names of all master eligible nodes in
advance, before creating the node with bootstrapping configuration set.
We address this issue by separating buildNode into two methods:
getNodeSettings and buildNode. We first iterate over all nodes to
get nodes settings, then change the setting for the bootstrapping node
and then proceed with building the node.
2. If autoManageMinMasterNodes = false, there is no way for the test to
set the list of bootstrapping nodes because node names are not known in
advance. This problem is solved by adding updateNodesSettings method
to NodeConfigurationSource and ESIntegTestCase (which could be
overridden by concrete integration test class). Once we have the list
of settings for all nodes, the integration test class is allowed to
update it. In our case, we update the
ClusterBootrstapService.INITIAL_MASTER_NODES_SETTING setting.
... MlDistributedFailureIT.testLoseDedicatedMasterNode.
An intermittent failure has been observed in
`MlDistributedFailureIT. testLoseDedicatedMasterNode`.
The test launches a cluster comprised by a dedicated master node
and a data and ML node. It creates a job and datafeed and starts them.
It then shuts down and restarts the master node. Finally, the test asserts
that the two tasks have been reassigned within 10s.
The intermittent failure is due to the assertions that the tasks have been
reassigned failing. Investigating the failure revealed that the `assertBusy`
that performs that assertion times out. Furthermore, it appears that the
job task is not reassigned because the memory tracking info is stale.
Memory tracking info is refreshed asynchronously when a job is attempted
to be reassigned. Tasks are attempted to be reassigned either due to a relevant
cluster state change or periodically. The periodic interval is controlled by a cluster
setting called `cluster.persistent_tasks.allocation.recheck_interval` and defaults to 30s.
What seems to be happening in this test is that if all cluster state changes after the
master node is restarted come through before the async memory info refresh completes,
then the job might take up to 30s until it is attempted to reassigned. Thus the `assertBusy`
times out.
This commit changes the test to reduce the periodic check that reassigns persistent
tasks to `200ms`. If the above theory is correct, this should eradicate those failures.
Closes#36760
Negative timestamps are currently supported in joda time. These are
dates before epoch. However, it doesn't really make sense to have a
negative timestamp, since this is a modern format. Any dates before
epoch can be represented with normal date formats, like ISO8601.
Additionally, implementing negative epoch timestamp parsing in java time
has an edge case which would more than double the code required. This
commit deprecates use of negative epoch timestamps.
Currently the ByteBufferReference does not duplicate the buffer.
This means that any changes to the buffer's limit or position will
impact the reference. This can lead to unexpected behavior. This commit
uses the ByteBuffer#slice method to ensure that the reference retains
its own ByteBuffer.
This commit partially reverts #36447 by using the ability of Joda time's
DateTimeFormatterBuilder to append multiple parsers instead of using the
MergedDateFormatter. The MergedDateFormatter will be removed in a future
change, as it is not as performant due to creating potentially many
exceptions during heavy date parsing. This change is a stop-gap until
that followup is ready.
closes#36602
The test testClusterInfoServiceInformationClearOnError relies on timing behavior. It sets
InternalClusterInfoService.INTERNAL_CLUSTER_INFO_TIMEOUT_SETTING to 1s and relies on the
fact that the stats request completes within that timeframe (which our ever-so-slow CI seems to
violate at times). Unfortunately the logging has been misimplemented in InternalClusterInfoService,
so the corresponding log messages showing that the requests have timed out are missing for this.
The issue can be locally reproduced by reducing the timeout to something lower.
Closes#36554
If the master sends both its follower checks just before disconnection then
neither will receive a response, meaning it must wait for the checks to time
out and send another in order to detect the disconnection and stand down.
Closes#36788
Single-node discovery is not persisting cluster states, which was caused by a recent 7.0-only
refactoring. This commit ensures that the cluster state is properly persisted when using single-node
discovery and adds a corresponding test.
* Replace 0L with an UNASSIGNED_PRIMARY_TERM constant
0 is an illegal value for a primary term that is often used to indicate
the primary term isn't assigned or is irrelevant. This PR replaces the
usage of 0 with a constant, to improve readability and so it can be
tracked and if needed, replaced.
* feedback
The default index_prefix settings will index prefixes of between 2 and 5 characters in length.
Currently, if a prefix search falls outside of this range at either end we fall back to a standard prefix
expansion, which is still very expensive for single character prefixes. However, we have an option
here to use a wildcard expansion rather than a prefix expansion, so that a query of a* gets remapped
to a? against the _index_prefix field - likely to be a very small set of terms, and certain to be much
smaller than a* against the whole index.
This commit adds this extra level of mapping for any prefix term whose length is one less than
the min_chars parameter of the index_prefixes field.
Today we might corrupt a fully deleted segment which is then pruned
once a snapshot is taken. This causes random test failures in CorruptedFileIT.
This change hardens the selection of files to corrupt and removes some fragile
code preventing fully deleted segments to be taken into account.
Closes#36526
Commit #36786 updated docs and strings to reference transport.port instead of
transport.tcp.port. However, this breaks backwards compatibility tests
as the tests rely on string configurations and transport.port does not
exist prior to 6.6. This commit reverts the places were we reference
transport.tcp.port for tests. This work will need to be reintroduced in
a backwards compatible way.
Lucene 7.6 uses a smaller encoding for LatLonShape. This commit forks the LatLonShape classes to Elasticsearch's local lucene package. These classes will be removed on the release of Lucene 7.6.
This is related to #36652. In 7.0 we plan to deprecate a number of
settings that make reference to the concept of a tcp transport. We
mostly just have a single transport type now (based on tcp). Settings
should only reference tcp if they are referring to socket options. This
commit updates the settings in the docs. And removes string usages of
the old settings. Additionally it adds a missing remote compress setting
to the docs.
TransportWriteAction.WriteReplicaResult is not properly synchronized, which can lead to a data race
between the thread that calls respond and the AsyncAfterWriteAction that calls either onSuccess or
onFailure. This data race results in the response listener not being called, which ultimately results in
a stuck replication task on the replica.
This commit is related to #36127. It adds a CcrRestoreSourceService to
track Engine.IndexCommitRef need for in-process file restores. When a
follower starts restoring a shard through the CcrRepository it opens a
session with the leader through the PutCcrRestoreSessionAction. The
leader responds to the request by telling the follower what files it
needs to fetch for a restore. This is not yet implemented.
Once, the restore is complete, the follower closes the session with the
DeleteCcrRestoreSessionAction action.
This commit removes the originalSettings member from Node. It was only
needed to allows test clusters to recreate the node in certain
situations. Instead, the test cluster now keeps track of these settings.
The joda epoch parsing code currently supports passing epoch time as a
number in scientific notation. However, no systems appear to exist which
output timestamps in scientific notation. In java time, it is
particularly complex to implement scientific notation timestamp parsing
within a DateTimeFormatter. This commit adds a deprecation warning when
the epoch time parsers in joda parse scientific notation, so that it can
be removed when switching to java time.
joda are passed a time in scientific notation.
* [Geo] Expose BKDBackedGeoShapes as new VECTOR strategy
This commit exposes lucene's LatLonShape field as a new
strategy in GeoShapeFieldMapper. To use the new indexing
approach, strategy should be set to "vector" in the
geo_shape field mapper. If the tree parameter is set
the mapper will throw an IAE. Note the following:
When using vector strategy:
* geo_shape query does not support querying by POINT,
MULTIPOINT, or GEOMETRYCOLLECTION.
* LINESTRING and MULTILINESTRING queries do not support
WITHIN relation.
* CONTAINS relation is not supported.
* The tree, precision, tree_levels, distance_error_pct,
and points_only parameters will not throw an exception
but they have no effect and will be marked as
deprecated..
All other features are supported.
* revert change to PercolatorFieldMapper
* fix ExistsQuery for geo_shape vector strategy
* add deprecation logging for tree, precision, tree_levels, distance_error_pct, and points_only
* initial update to geoshape docs, including mapping migration updates
* initial support for GeoCollection queries
* fix docs and javadoc errors
* clean up geocollection queries
* set deprecated mapping tests to NOTCONSOLE
* fix geo-shape mapper asciidoc mapping and test warnings
* add support for point queries using LatLonShapeBoundingBoxQuery
* update GeoShapeQueryBuilderTests to include POINT queries for VECTOR strategy. Other comment cleanups
* add lucene geometry build testing to ShapeBuilder tests
* remove deprecated prefix tree mapping from geo-shape.asciidoc
* refactor GeoShapeFieldMapper into LegacyGeoShapeFieldMapper and GeoShapeFieldMapper
Both classes derive from BaseGeoShapeFieldMapper that provides shared parameters:
coerce, ignoreMalformed, ignore_z_value, orientation.
* update docs to remove vector strategy
* fix GeometryCollectionBuilder#buildLucene to return the object created by the shape builder
* fix LineLength failure in GeoJsonShapeParserTests
* ShapeMapper refactor changes from PR feedback
* fix typo in geo-shape.asciidoc
* ignore circle test in docs
* update indexing-approach ref to geoshape-indexing-approach
* add warnings check for LegacyGeoShapeFieldMapper to AbstractBuilderTestCase
* fix deprecatedParameters setup
* update indexing approach
* fixing unexpected warnings failures
* move orientation back to field type
* remove if in LegacyGeoShapeFieldMapper#doXContent. Fix GeoShapeFieldMapper to work with double array as a point
* fix indexing-approach link in circle section of geoshape docs
* add strategy to deprecation warnings check
* fix test failures
* fix typo in QueryStringQueryBuilderTests
* fix total hits to totalHits().value
* fix version number
* add version check to BaseGeoShapeFieldMapper
* fix line length!
* revert version check in BaseGeoShapeFieldMapper
* Fix serialization of mappings of legacy shapes.
Yannick Welsch
Simon Willnauer
David Roberts
Armin Braun
David Turner
Mayya Sharipova
Tim Brooks
Jason Tedor
Benjamin Trent
Igor Motov
Julie Tibshirani
David Kyle
Nhat Nguyen
Luca Cavanna
Tanguy Leroux
Alexander Reelsen
Boaz Leskes
Andrey Ershov
Zachary Tong
Henning Andersen
Christoph Büscher
Adrien Grand
Alpar Torok
Tal Levy
Jack Conradson
James Baiera
Jim Ferenczi
Martijn van Groningen
Torgeir Thoresen
Przemyslaw Gomulka
Dimitrios Liappis
Fabricio Archanjo Fonseca
Tim Vernum
markharwood
Daniel Mitterdorfer
Ignacio Vera
Ryan Ernst
Michael Basnight
Jim Ferenczi
Jake Landis
Evangelos Chatzikalymnios
Jun Ohtani
Christophe Bismuth
Nick Knize
David Findley
Ke Li
Alan Woodward
Josh Soref
Alexis Wilke
Itamar Benjamin
Dimitris Athanasiou
David Causse