Today when we get a metadata snapshot directly from a store directory,
we acquire a metadata lock, then acquire an IndexWriter lock. However,
we create a CheckIndex in IndexShard without acquiring the metadata lock
first. This causes a recovery failed because the IndexWriter lock can be
still held by method snapshotStoreMetadata. This commit makes sure to
create a CheckIndex under the metadata lock.
Closes#24481Closes#27731
Relates #24787
The last operation executed in IndicesClientDocumentationIT.testCreate()
is an asynchronous index creation. Because nothing waits for its
completion, on slow machines the index can sometimes be created after
the testCreate() test is finished, and it can fail the following test.
Closes#27754
When the first parameter of `ESTestCase#randomValueOtherThan` is `null`
then run the supplier until it returns non-`null`. Previously,
`randomValueOtherThan` just ran the supplier one time which was
confusing.
Unexpectedly, it looks like not tests rely on the original `null`
handling.
Closes#27821
We currently have a complicated port assignment scheme to make sure that the nodes span off by the internal test cluster will be assigned fixed port ranges that will also not collide between clusters. The port ranges need to be fixed in advance so that the nodes will be able to find each other via `UnicastZenPing`.
This approach worked well for the last few years but we are now at a point that our testing has grown beyond it and we exceed the 5 reusable ranges per JVM. This means that nodes are not always assigned the first 5 ports in their range which causes cluster formation issues. On top of that, most of the clusters that are span up don't even rely on `UnicastZenPing` but rather `MockZenPings` that uses in memory maps for discovery (with the down side that they are not influenced by network disruption simulations).
This PR changes `InternalTestCluster` to use port 0 as a fixed assignment. This will allow the OS to manage ports and will ensure we don't have collisions. For tests that need to simulate network disruptions (and thus can't use `MockZenPings`), a new `UnicastHostProvider` is introduced that is based on the current state of the test cluster. Since that is only resolved at run time, it is aware of the port assignments of the OS.
Closes#27818Closes#27762
This commit moves GlobalCheckpointTracker from the engine to IndexShard, where it better fits logically: Tracking the global checkpoint based on the local checkpoints of all shards in the replication group is not a property of the engine, but rather a property fulfilled by the current primary shard. The LocalCheckpointTracker on the other hand is driven by the contents of the local translog. By moving GlobalCheckpointTracker to IndexShard, it makes little sense to keep the SequenceNumbersService class around - it would only wrap the LocalCheckpointTracker. This commit therefore removes the class and replaces occurrences of SequenceNumbersService in the engine directly by LocalCheckpointTracker.
AnalysisFactoryTestCase checks that the ES custom token filter multi-term
awareness matches the underlying lucene factory. For the trim filter this
won't be the case until LUCENE-8093 is released in 7.3, so we add a
temporary exclusion
Closes#27310
This commit fixes the version tests for release tests. The problem here
is that during release tests all version should be treated as released
so the assertions must be modified accordingly.
Relates #27815
When snapshotting the primary we capture a lucene commit at an arbitrary moment from a sequence number perspective. This means that it is possible that the commit misses operations and that there is a gap between the local checkpoint in the commit and the maximum sequence number.
When we restore, this will create a primary that "misses" operations and currently will mean that the sequence number system is stuck (i.e., the local checkpoint will be stuck). To fix this we should fill in gaps when we restore, in a similar fashion to normal store recovery.
Currently randomNonNegativeLong() returns 0 half as often as any positive long,
but random number generators are typically expected to return
uniformly-distributed values unless otherwise specified. This fixes this issue
by mapping Long.MIN_VALUE directly onto 0 rather than resampling.
This commit is related to #27260. It adds a base NioGroup for use in
different transports. This class creates and starts the underlying
selectors. Different protocols or transports are established by passing
the ChannelFactory to the bindServerChannel or openChannel
methods. This allows a TcpChannelFactory to be passed which will
create and register channels that support the elasticsearch tcp binary
protocol or a channel factory that will create http channels (or other).
When an ESSelector is created an underlying nio selector is opened. This
selector is closed by the event loop after close has been signalled by
another thread.
However, there is a possibility that an ESSelector is created and some
exception in the startup process prevents it from ever being started
(however, close will still be called). The allows the selector to leak.
This commit addresses this issue by having the signalling thread close
the selector if the event loop is not running when close is signalled.
Allowing `_doc` as a type will enable users to make the transition to 7.0
smoother since the index APIs will be `PUT index/_doc/id` and `POST index/_doc`.
This also moves most of the documentation to `_doc` as a type name.
Closes#27750Closes#27751
We need to keep index commits and translog operations up to the current
global checkpoint to allow us to throw away unsafe operations and
increase the operation-based recovery chance. This is achieved by a new
index deletion policy.
Relates #10708
This commit attempts to continue unifying the logic between different
transport implementations. As transports call a `TcpTransport` callback
when a new channel is accepted, there is no need to internally track
channels accepted. Instead there is a set of accepted channels in
`TcpTransport`. This set is used for metrics and shutting down channels.
This is related to #27563. This commit modifies the
InboundChannelBuffer to support releasable byte pages. These byte
pages are provided by the PageCacheRecycler. The PageCacheRecycler
must be passed to the Transport with this change.
This is a follow up to #27695. This commit adds a test checking that
across multiple writes using multiple buffers, a write operation
properly keeps track of which buffers still need to be written.
This is a followup to #27551. That commit introduced a bug where the
incorrect byte buffers would be returned when we attempted a write. This
commit fixes the logic.
This is related to #27563. In order to interface with java nio, we must
have buffers that are compatible with ByteBuffer. This commit introduces
a basic ByteBufferReference to easily allow transferring bytes off the
wire to usage in the application.
Additionally it introduces an InboundChannelBuffer. This is a buffer
that can internally expand as more space is needed. It is designed to
be integrated with a page recycler so that it can internally reuse pages.
The final piece is moving all of the index work for writing bytes to a
channel into the WriteOperation.
This commit adds a new dynamic cluster setting named `search.max_buckets` that can be used to limit the number of buckets created per shard or by the reduce phase. Each multi bucket aggregator can consume buckets during the final build of the aggregation at the shard level or during the reduce phase (final or not) in the coordinating node. When an aggregator consumes a bucket, a global count for the request is incremented and if this number is greater than the limit an exception is thrown (TooManyBuckets exception).
This change adds the ability for multi bucket aggregator to "consume" buckets in the global limit, the default is 10,000. It's an opt-in consumer so each multi-bucket aggregator must explicitly call the consumer when a bucket is added in the response.
Closes#27452#26012
Add support for filtering fields returned as part of mappings in get index, get mappings, get field mappings and field capabilities API.
Plugins can plug in their own function, which receives the index as argument, and return a predicate which controls whether each field is included or not in the returned output.
This commit adds the node name to the names of thread pool executors so
that the node name is visible in rejected execution exception messages.
Relates #27663
Today we exclude internal refreshes in the refresh stats. Yet, it's very much
confusing to not take these into account. This change includes internal refreshes
into the stats until we have a dedicated stats for this.
* Add accounting circuit breaker and track segment memory usage
This commit adds a new circuit breaker "accounting" that is used for tracking
the memory usage of non-request-tied memory users. It also adds tracking for the
amount of Lucene segment memory used by a shard as a user of the new circuit
breaker.
The Lucene segment memory is updated when the shard refreshes, and removed when
the shard relocates away from a node or is deleted. It should also be noted that
all tracking for segment memory uses `addWithoutBreaking` so as not to fail the
shard if a limit is reached.
The `accounting` breaker has a default limit of 100% and will contribute to the
parent breaker limit.
Resolves#27044
This potential issue was exposed when I saw this PR #27542. Essentially
we currently execute the write listeners all over the place without
consistently catching and handling exceptions. Some of these exceptions
will be logged in different ways (including as low as `debug`).
This commit adds a single location where these listeners are executed.
If the listener throws an execption, the exception is caught and logged
at the `warn` level.
Pull request #20220 added a change where the store files
that have the same name but are different from the ones in the
snapshot are deleted first before the snapshot is restored.
This logic was based on the `Store.RecoveryDiff.different`
set of files which works by computing a diff between an
existing store and a snapshot.
This works well when the files on the filesystem form valid
shard store, ie there's a `segments` file and store files
are not corrupted. Otherwise, the existing store's snapshot
metadata cannot be read (using Store#snapshotStoreMetadata())
and an exception is thrown
(CorruptIndexException, IndexFormatTooOldException etc) which
is later caught as the begining of the restore process
(see RestoreContext#restore()) and is translated into
an empty store metadata (Store.MetadataSnapshot.EMPTY).
This will make the deletion of different files introduced
in #20220 useless as the set of files will always be empty
even when store files exist on the filesystem. And if some
files are present within the store directory, then restoring
a snapshot with files with same names will fail with a
FileAlreadyExistException.
This is part of the #26865 issue.
There are various cases were some files could exist in the
store directory before a snapshot is restored. One that
Igor identified is a restore attempt that failed on a node
and only first files were restored, then the shard is allocated
again to the same node and the restore starts again (but fails
because of existing files). Another one is when some files
of a closed index are corrupted / deleted and the index is
restored.
This commit adds a test that uses the infrastructure provided
by IndexShardTestCase in order to test that restoring a shard
succeed even when files with same names exist on filesystem.
Related to #26865
This is related to #27260. Currently, basic nio constructs (nio
channels, the channel factories, selector event handlers, etc) implement
logic that is specific to the tcp transport. For example, NioChannel
implements the TcpChannel interface. These nio constructs at some point
will also need to support other protocols (ex: http).
This commit separates the TcpTransport logic from the nio building
blocks.
This change removes the module named aggs-composite and adds the `composite` aggs
as a core aggregation. This allows other plugins to use this new aggregation
and simplifies the integration in the HL rest client.
This is related to #27260. Currently every nio channel has a profile
field. Profile is a concept that only relates to the tcp transport. Http
channels will not have profiles. This commit moves the profile from the
nio channel to the read context. The context is the level that protocol
specific features and logic should live.
Currently we use ActionListener<TcpChannel> for connect, close, and send
message listeners in TcpTransport. However, all of the listeners have to
capture a reference to a channel in the case of the exception api being
called. This commit changes these listeners to be type <Void> as passing
the channel to onResponse is not necessary. Additionally, this change
makes it easier to integrate with low level transports (which use
different implementations of TcpChannel).
This commit removes the ability to use ${prompt.secret} and
${prompt.text} as valid config settings. Secure settings has obsoleted
the need for this, and it cleans up some of the code in Bootstrap.
Projects the depend on the CLI currently depend on core. This should not
always be the case. The EnvironmentAwareCommand will remain in :core,
but the rest of the CLI components have been moved into their own
subproject of :core, :core:cli.
This is related to #27260. Currently, every ESSelector keeps track of
all channels that are registered with it. ESSelector is just an
abstraction over a raw java nio selector. The java nio selector already
tracks its own selection keys. This commit removes our tracking and
relies on the java nio selector tracking.
It leads to harder-to-parse logs that look like this:
```
1> [2017-11-16T20:46:21,804][INFO ][o.e.t.r.y.ClientYamlTestClient] Adding header Content-Type
1> with value application/json
1> [2017-11-16T20:46:21,812][INFO ][o.e.t.r.y.ClientYamlTestClient] Adding header Content-Type
1> with value application/json
1> [2017-11-16T20:46:21,820][INFO ][o.e.t.r.y.ClientYamlTestClient] Adding header Content-Type
1> with value application/json
1> [2017-11-16T20:46:21,966][INFO ][o.e.t.r.y.ClientYamlTestClient] Adding header Content-Type
1> with value application/json
```
This is related to #27260. In the nio transport work we do not catch or
handle `Throwable`. There are a few places where we have exception
handlers that accept `Throwable`. This commit removes those cases.
This commit is a follow up to the work completed in #27132. Essentially
it transitions two more methods (sendMessage and getLocalAddress) from
Transport to TcpChannel. With this change, there is no longer a need for
TcpTransport to be aware of the specific type of channel a transport
returns. So that class is no longer parameterized by channel type.
This is a follow up to #27132. As that PR greatly simplified the
connection logic inside a low level transport implementation, much of
the functionality provided by the NioClient class is no longer
necessary. This commit removes that class.
* This change adds a module called `aggs-composite` that defines a new aggregation named `composite`.
The `composite` aggregation is a multi-buckets aggregation that creates composite buckets made of multiple sources.
The sources for each bucket can be defined as:
* A `terms` source, values are extracted from a field or a script.
* A `date_histogram` source, values are extracted from a date field and rounded to the provided interval.
This aggregation can be used to retrieve all buckets of a deeply nested aggregation by flattening the nested aggregation in composite buckets.
A composite buckets is composed of one value per source and is built for each document as the combinations of values in the provided sources.
For instance the following aggregation:
````
"test_agg": {
"terms": {
"field": "field1"
},
"aggs": {
"nested_test_agg":
"terms": {
"field": "field2"
}
}
}
````
... which retrieves the top N terms for `field1` and for each top term in `field1` the top N terms for `field2`, can be replaced by a `composite` aggregation in order to retrieve **all** the combinations of `field1`, `field2` in the matching documents:
````
"composite_agg": {
"composite": {
"sources": [
{
"field1": {
"terms": {
"field": "field1"
}
}
},
{
"field2": {
"terms": {
"field": "field2"
}
}
},
}
}
````
The response of the aggregation looks like this:
````
"aggregations": {
"composite_agg": {
"buckets": [
{
"key": {
"field1": "alabama",
"field2": "almanach"
},
"doc_count": 100
},
{
"key": {
"field1": "alabama",
"field2": "calendar"
},
"doc_count": 1
},
{
"key": {
"field1": "arizona",
"field2": "calendar"
},
"doc_count": 1
}
]
}
}
````
By default this aggregation returns 10 buckets sorted in ascending order of the composite key.
Pagination can be achieved by providing `after` values, the values of the composite key to aggregate after.
For instance the following aggregation will aggregate all composite keys that sorts after `arizona, calendar`:
````
"composite_agg": {
"composite": {
"after": {"field1": "alabama", "field2": "calendar"},
"size": 100,
"sources": [
{
"field1": {
"terms": {
"field": "field1"
}
}
},
{
"field2": {
"terms": {
"field": "field2"
}
}
}
}
}
````
This aggregation is optimized for indices that set an index sorting that match the composite source definition.
For instance the aggregation above could run faster on indices that defines an index sorting like this:
````
"settings": {
"index.sort.field": ["field1", "field2"]
}
````
In this case the `composite` aggregation can early terminate on each segment.
This aggregation also accepts multi-valued field but disables early termination for these fields even if index sorting matches the sources definition.
This is mandatory because index sorting picks only one value per document to perform the sort.
Right now our different transport implementations must duplicate
functionality in order to stay compliant with the requirements of
TcpTransport. They must all implement common logic to open channels,
close channels, keep track of channels for eventual shutdown, etc.
Additionally, there is a weird and complicated relationship between
Transport and TransportService. We eventually want to start merging
some of the functionality between these classes.
This commit starts moving towards a world where TransportService retains
all the application logic and channel state. Transport implementations
in this world will only be tasked with returning a channel when one is
requested, calling transport service when a channel is accepted from
a server, and starting / stopping itself.
Specifically this commit changes how channels are opened and closed. All
Transport implementations now return a channel type that must comply with
the new TcpChannel interface. This interface has the methods necessary
for TcpTransport to completely manage the lifecycle of a channel. This
includes setting the channel up, waiting for connection, adding close
listeners, and eventually closing.
We use affix settings to group settings / values under a certain namespace.
In some cases like login information for instance a setting is only valid if
one or more other settings are present. For instance `x.test.user` is only valid
if there is an `x.test.passwd` present and vice versa. This change allows to specify
such a dependency to prevent settings updates that leave settings in an inconsistent
state.
We cut over to internal and external IndexReader/IndexSearcher in #26972 which uses
two independent searcher managers. This has the downside that refreshes of the external
reader will never clear the internal version map which in-turn will trigger additional
and potentially unnecessary segment flushes since memory must be freed. Under heavy
indexing load with low refresh intervals this can cause excessive segment creation which
causes high GC activity and significantly increases the required segment merges.
This change adds a dedicated external reference manager that delegates refreshes to the
internal reference manager that then `steals` the refreshed reader from the internal
reference manager for external usage. This ensures that external and internal readers
are consistent on an external refresh. As a sideeffect this also releases old segments
referenced by the internal reference manager which can potentially hold on to already merged
away segments until it is refreshed due to a flush or indexing activity.
* Decouple `ChannelFactory` from Tcp classes
This is related to #27260. Currently `ChannelFactory` is tightly coupled
to classes related to the elasticsearch Tcp binary protocol. This commit
modifies the factory to be able to construct http or other protocol
channels.
If an out of memory error is thrown while merging, today we quietly
rewrap it into a merge exception and the out of memory error is
lost. Instead, we need to rethrow out of memory errors, and in fact any
fatal error here, and let those go uncaught so that the node is torn
down. This commit causes this to be the case.
Relates #27265
The warnings headers have a fairly limited set of valid characters
(cf. quoted-text in RFC 7230). While we have assertions that we adhere
to this set of valid characters ensuring that our warning messages do
not violate the specificaion, we were neglecting the possibility that
arbitrary user input would trickle into these warning headers. Thus,
missing here was tests for these situations and encoding of characters
that appear outside the set of valid characters. This commit addresses
this by encoding any characters in a deprecation message that are not
from the set of valid characters.
Relates #27269
This change adds a new `_split` API that allows to split indices into a new
index with a power of two more shards that the source index. This API works
alongside the `_shrink` API but doesn't require any shard relocation before
indices can be split.
The split operation is conceptually an inverse `_shrink` operation since we
initialize the index with a _syntetic_ number of routing shards that are used
for the consistent hashing at index time. Compared to indices created with
earlier versions this might produce slightly different shard distributions but
has no impact on the per-index backwards compatibility. For now, the user is
required to prepare an index to be splittable by setting the
`index.number_of_routing_shards` at index creation time. The setting allows the
user to prepare the index to be splittable in factors of
`index.number_of_routing_shards` ie. if the index is created with
`index.number_of_routing_shards: 16` and `index.number_of_shards: 2` it can be
split into `4, 8, 16` shards. This is an intermediate step until we can make
this the default. This also allows us to safely backport this change to 6.x.
The `_split` operation is implemented internally as a DeleteByQuery on the
lucene level that is executed while the primary shards execute their initial
recovery. Subsequent merges that are triggered due to this operation will not be
executed immediately. All merges will be deferred unti the shards are started
and will then be throttled accordingly.
This change is intended for the 6.1 feature release but will not support pre-6.1
indices to be split unless these indices have been shrunk before. In that case
these indices can be split backwards into their original number of shards.
While it's not possible to upgrade the Jackson dependencies
to their latest versions yet (see #27032 (comment) for more)
it's still possible to upgrade to the latest 2.8.x version.
We have an hidden setting called `index.queries.cache.term_queries` that disables caching of term queries in the query cache.
Though term queries are not cached in the Lucene UsageTrackingQueryCachingPolicy since version 6.5.
This makes the es policy useless but also makes it impossible to re-enable caching for term queries.
This change appeared in Lucene 6.5 so this setting is no-op since version 5.4 of Elasticsearch
The change in this PR removes the setting and the custom policy.
Only tests should use the single argument Environment constructor. To
enforce this the single arg Environment constructor has been replaced with
a test framework factory method.
Production code (beyond initial Bootstrap) should always use the same
Environment object that Node.getEnvironment() returns. This Environment
is also available via dependency injection.
For FsBlobStore and HdfsBlobStore, if the repository is read only, the blob store should be aware of the readonly setting and do not create directories if they don't exist.
Closes#21495
When partitioning version constants into released and unreleased
versions, today we have a bug in finding the last unreleased
version. Namely, consider the following version constants on the 6.x
branch: ..., 5.6.3, 5.6.4, 6.0.0-alpha1, ..., 6.0.0-rc1, 6.0.0-rc2,
6.0.0, 6.1.0. In this case, our convention dictates that: 5.6.4, 6.0.0,
and 6.1.0 are unreleased. Today we correctly detect that 6.0.0 and 6.1.0
are unreleased, and then we say the previous patch version is unreleased
too. The problem is the logic to remove that previous patch version is
broken, it does not skip alphas/betas/RCs which have been released. This
commit fixes this by skipping backwards over pre-release versions when
finding the previous patch version to remove.
Relates #27206
* Enhances exists queries to reduce need for `_field_names`
Before this change we wrote the name all the fields in a document to a `_field_names` field and then implemented exists queries as a term query on this field. The problem with this approach is that it bloats the index and also affects indexing performance.
This change adds a new method `existsQuery()` to `MappedFieldType` which is implemented by each sub-class. For most field types if doc values are available a `DocValuesFieldExistsQuery` is used, falling back to using `_field_names` if doc values are disabled. Note that only fields where no doc values are available are written to `_field_names`.
Closes#26770
* Addresses review comments
* Addresses more review comments
* implements existsQuery explicitly on every mapper
* Reinstates ability to perform term query on `_field_names`
* Added bwc depending on index created version
* Review Comments
* Skips tests that are not supported in 6.1.0
These values will need to be changed after backporting this PR to 6.x
It is required in order to work correctly with bulk scorer implementations
that change the scorer during the collection process. Otherwise sub collectors
might call `Scorer.score()` on the wrong scorer.
Closes#27131
This commit is a minor refactoring of internal engine to move hooks for
generating sequence numbers into the engine itself. As such, we refactor
tests that relied on this hook to use the new hook, and remove the hook
from the sequence number service itself.
Relates #27082
The headers passed to reindex were skipped except for the last one. This
commit fixes the copying of the headers, as well as adds a base test
case for rest client builders to access the headers within the built
rest client.
relates #22976
Till now the yaml test runner was verifying that the provided path parts and parameters are supported.
With this PR, yaml test runner also checks that all required path parts and parameters are provided.
Introduce minimal thread scheduler as a base class for `ThreadPool`. Such a class can be used from the `BulkProcessor` to schedule retries and the flush task. This allows to remove the `ThreadPool` dependency from `BulkProcessor`, which requires to provide settings that contain `node.name` and also needed log4j for logging. Instead, it needs now a `Scheduler` that is much lighter and gets automatically created and shut down on close.
Closes#26028
Right now we are attempting to set SO_LINGER to 0 on server channels
when we are stopping the tcp transport. This is not a supported socket
option and throws an exception. This also prevents the channels from
being closed.
This commit 1. doesn't set SO_LINGER for server channges, 2. checks
that it is a supported option in nio, and 3. changes the log message
to warn for server channel close exceptions.
While opening a connection to a node, a channel can subsequently
close. If this happens, a future callback whose purpose is to close all
other channels and disconnect from the node will fire. However, this
future will not be ready to close all the channels because the
connection will not be exposed to the future callback yet. Since this
callback is run once, we will never try to disconnect from this node
again and we will be left with a closed channel. This commit adds a
check that all channels are open before exposing the channel and throws
a general connection exception. In this case, the usual connection retry
logic will take over.
Relates #26932
Today we return a `String[]` that requires copying values for every
access. Yet, we already store the setting as a list so we can also directly
return the unmodifiable list directly. This makes list / array access in settings
a much cheaper operation especially if lists are large.
The shard preference _primary, _replica and its variants were useful
for the asynchronous replication. However, with the current impl, they
are no longer useful and should be removed.
Closes#26335
* Add additional low-level logging handler
We have the trace handler which is useful for recording sent messages
but there are times where it would be useful to have more low-level
logging about the events occurring on a channel. This commit adds a
logging handler that can be enabled by setting a certain log level
(org.elasticsearch.transport.netty4.ESLoggingHandler) to trace that
provides trace logging on low-level channel events and includes some
information about the request/response read/write events on the channel
as well.
* Remove imports
* License header
* Remove redundant
* Add test
* More assertions
Today we represent each value of a list setting with it's own dedicated key
that ends with the index of the value in the list. Aside of the obvious
weirdness this has several issues especially if lists are massive since it
causes massive runtime penalties when validating settings. Like a list of 100k
words will literally cause a create index call to timeout and in-turn massive
slowdown on all subsequent validations runs.
With this change we use a simple string list to represent the list. This change
also forbids to add a settings that ends with a .0 which was internally used to
detect a list setting. Once this has been rolled out for an entire major
version all the internal .0 handling can be removed since all settings will be
converted.
Relates to #26723
Since `#getAsMap` exposes internal representation we are trying to remove it
step by step. This commit is cleaning up some xcontent writing as well as
usage in tests
This commit fixes a #26855. Right now we set SO_LINGER to 0 if we are
stopping the transport. This can throw a ChannelClosedException if the
raw channel is already closed. We have a number of scenarios where it is
possible this could be called with a channel that is already closed.
This commit fixes the issue be checking that the channel is not closed
before attempting to set the socket option.
Currently we only log generic messages about errors in logs from the
nio event handler. This means that we do not know which channel had
issues connection, reading, writing, etc.
This commit changes the logs to include the local and remote addresses
and profile for a channel.
We use group settings historically instead of using a prefix setting which is more restrictive and type safe. The majority of the usecases needs to access a key, value map based on the _leave node_ of the setting ie. the setting `index.tag.*` might be used to tag an index with `index.tag.test=42` and `index.tag.staging=12` which then would be turned into a `{"test": 42, "staging": 12}` map. The group settings would always use `Settings#getAsMap` which is loosing type information and uses internal representation of the settings. Using prefix settings allows now to access such a method type-safe and natively.
Currently we only log generic messages about errors in logs from the
nio event handler. This means that we do not know which channel had
issues connection, reading, writing, etc.
This commit changes the logs to include the local and remote addresses
and profile for a channel.
This change adds a fromXContent method to Settings that allows to read
the xcontent that is produced by toXContent. It also replaces the entire settings
loader infrastructure and removes the structured map representation. Future PRs will
also tackle the `getAsMap` that exposes the internal represenation of settings for
better encapsulation.
It is the exciting return of the global checkpoint background
sync. Long, long ago, in snapshot version far, far away we had and only
had a global checkpoint background sync. This sync would fire
periodically and send the global checkpoint from the primary shard to
the replicas so that they could update their local knowledge of the
global checkpoint. Later in time, as we sped ahead towards finalizing
the initial version of sequence IDs, we realized that we need the global
checkpoint updates to be inline. This means that on a replication
operation, the primary shard would piggy back the global checkpoint with
the replication operation to the replicas. The replicas would update
their local knowledge of the global checkpoint and reply with their
local checkpoint. However, this could allow the global checkpoint on the
primary to advance again and the replicas would fall behind in their
local knowledge of the global checkpoint. If another replication
operation never fired, then the replicas would be permanently behind. To
account for this, we added one more sync that would fire when the
primary shard fell idle. However, this has problems:
- the shard idle timer defaults to five minutes, a long time to wait
for the replicas to learn of the new global checkpoint
- if a replica missed the sync, there was no follow-up sync to catch
them up
- there is an inherent race condition where the primary shard could
fall idle mid-operation (after having sent the replication request to
the replicas); in this case, there would never be a background sync
after the operation completes
- tying the global checkpoint sync to the idle timer was never natural
To fix this, we add two additional changes for the global checkpoint to
be synced to the replicas. The first is that we add a post-operation
sync that only fires if there are no operations in flight and there is a
lagging replica. This gives us a chance to sync the global checkpoint to
the replicas immediately after an operation so that they are always kept
up to date. The second is that we add back a global checkpoint
background sync that fires on a timer. This timer fires every thirty
seconds, and is not configurable (for simplicity). This background sync
is smarter than what we had previously in the sense that it only sends a
sync if the global checkpoint on at least one replica is lagging that of
the primary. When the timer fires, we can compare the global checkpoint
on the primary to its knowledge of the global checkpoint on the replicas
and only send a sync if there is a shard behind.
Relates #26591
Removing several occurrences of this typo in the docs and javadocs, seems to be
a common mistake. Corrections turn up once in a while in PRs, better to correct
some of this in one sweep.
Today we have all non-plugin mappers in core. I'd like to start moving those
that neither map to json datatypes nor are very frequently used like `date` or
`ip` to a module.
This commit creates a new module called `mappers-extra` and moves the
`scaled_float` and `token_count` mappers to it. I'd like to eventually move
`range` fields there but it's more complicated due to their intimate
relationship with range queries.
Relates #10368
Today we don't have a pluggable way to validate if the cluster state
is compatible with the node that joins. We already apply some checks for index
compatibility that prevents nodes to join a cluster with indices it doesn't support
but for plugins this isn't possible. This change adds a cluster state validator that
allows plugins to prevent a join if the cluster-state is incompatible.
This test case was leftover from the static bwc tests. There was still
one use for checking we do not load old indices, but this PR moves the
legacy code needed for that directly into the test. I also opened a
follow up issue to completely remove the unsupported test: #26583.
When determining if a build is a snapshot build, we look for a field in
the JAR manifest. However, when running tests, we are not running with a
compiled core Elasticsearch JAR, we are running with the compiled core
classes on the classpath. We have a fallback for this, we always assume
such a situation is a snapshot build. However, when running builds with
-Dbuild.snapshot=false, this is not the case. As such, we need to
fallback to the value of build.snapshot. However, there are cases where
we are not running with a compiled core Elasticsearch JAR (e.g., when
the transport client is embedded in a web container) so we should only
do this fallback if we are in tests. To verify we are in tests, we check
if randomized runner is on the classpath.
Relates #26554
The percolator will add a `_percolator_document_slot` field to all percolator
hits to indicate with what document it has matched. This number matches with
the order in which the documents have been specified in the percolate query.
Also improved the support for multiple percolate queries in a search request.
We currently have a weird relationship between Transport,
TransportService, and TransportServiceAdaptor. At some point I think
that we would like to collapse these all into one concept as we only
support TCP transports.
This commit moves in that direction by eliminating the adaptor and just
passing the transport service to the transport.
The current script service has a script compilation limit for a one
minute window. This is set to a small default value of 15. Instead of
increasing that default value, this commit introduces a new setting
that allows to configure a rate per time unit, so that the script service can deal with bursts better.
The new setting is named `script.max_compilations_rate`,
requires a nonnegative number and a positive time value.
The default is `75/5m`, which is equivalent to the existing 15 per minute.
* Implement adaptive replica selection
This implements the selection algorithm described in the C3 paper for
determining which copy of the data a query should be routed to.
By using the service time EWMA, response time EWMA, and queue size EWMA we
calculate the score of a node by piggybacking these metrics with each search
request.
Since Elasticsearch lacks the "broadcast to every copy" behavior that Cassandra
has (as mentioned in the C3 paper) to update metrics after a node has been
highly weighted, this implementation adjusts a node's response stats using the
average of the its own and the "best" node's metrics. This is so that a long GC
or other activity that may cause a node's rank to increase dramatically does not
permanently keep a node from having requests routed to it, instead it will
eventually lower its score back to the realm where it is a potential candidate
for new queries.
This feature is off by default and can be turned on with the dynamic setting
`cluster.routing.use_adaptive_replica_selection`.
Relates to #24915, however instead of `b=3` I used `b=4` (after benchmarking)
* Randomly use adaptive replica selection for internal test cluster
* Use an action name *prefix* for retrieving pending requests
* Add unit test for replica selection
* don't use adaptive replica selection in SearchPreferenceIT
* Track client connections in a SearchTransportService instead of TransportService
* Bind `entry` pieces in local variables
* Add javadoc link to C3 paper and javadocs for stat adjustments
* Bind entry's key and value to local variables
* Remove unneeded actionNamePrefix parameter
* Use conns.longValue() instead of cached Long
* Add comments about removing entries from the map
* Pull out bindings for `entry` in IndexShardRoutingTable
* Use .compareTo instead of manually comparing
* add assert for connections not being null and gte to 1
* Copy map for pending search connections instead of "live" map
* Increase the number of pending search requests used for calculating rank when chosen
When a node gets chosen, this increases the number of search counts for the
winning node so that it will not be as likely to be chosen again for
non-concurrent search requests.
* Remove unused HashMap import
* Rename rank -> rankShardsAndUpdateStats
* Rename rankedActiveInitializingShardsIt -> activeInitializingShardsRankedIt
* Instead of precalculating winning node, use "winning" shard from ranked list
* Sort null ranked nodes before nodes that have a rank
At current, we do not feel there is enough of a reason to shade the low
level rest client. It caused problems with commons logging and IDE's
during the brief time it was used. We did not know exactly how many
users will need this, and decided that leaving shading out until we
gather more information is best. Users can still shade the jar
themselves. For information and feeback, see issue #26366.
Closes#26328
This reverts commit 3a20922046.
This reverts commit 2c271f0f22.
This reverts commit 9d10dbea39.
This reverts commit e816ef89a2.
This allows plugins to plug rescore implementations into
Elasticsearch. While this is a fairly expert thing to do I've
done my best to point folks to the QueryRescorer as one that at
least documents the tradeoffs that it makes. I've attempted to
limit the API surface area by removing `SearchContext` from the
exposed interface, instead exposing just the IndexSearcher and
`QueryShardContext`. I also tried to make some of the class names
more consistent and do some general cleanup while I was there.
I entertained the notion of moving the `QueryRescorer` to module.
After all, it'd be a wonderful test to prove that you can plug
rescore implementation into Elasticsearch if the only built in
rescore implementation is in the module. But I decided against it
because the new module would require a client jar and it'd require
moving some more things around. I think if we really want to do
it, we should do it as a followup.
I did, on the other hand, create an "example" rescore plugin which
should both be a nice example for anyone wanting to plug in their
own rescore implementation and servers as a good integration test
to make sure that you can indeed plug one in.
Closes#26208
This PR begins the long journey to deprecating Streamable.
The idea here is to add additional method signatures that
support Writeable.Reader, so that the work to migrate objects TransportMessage to
implement Writeable and not Streamable.
One example conversion is done in this PR: SimulatePipelineRequest.
This commit makes the security code aware of the Java 9 FilePermission changes (see #21534) and allows us to remove the `jdk.io.permissionsUseCanonicalPath` system property.
This commit converts script query to use a new FilterScript context. The
new context returns a boolean, so the error that would have previously
happened at runtime if a non boolean was returned would now happen at
script compilation. Also, the leniency of supporting returning a number
and 0 mapping to false, non-zero to true is gone, but it was never
documented. With the new context compilation will now also fail if
special variables are used at compilation time, instead of runtime, eg
ctx.
Right now we use a custom future for the CloseFuture associated with a
channel. This is because we need special unwrapping logic to ensure that
exceptions from a future failure are a certain type (opposed to an
UncategorizedException). However, the current version is limiting
because we can only attach one listener.
This commit changes the CloseFuture to extend the
PlainListenableActionFuture. This change allows us to attach multiple
listeners.
The client sniffer depends on the low-level REST client, while the Java high-level REST client and the transport client depend on Elasticsearch itself. Javadoc are not that useful unless they have links to the Elasticsearch classes in the latter case, and to the low-level REST client in the sniffer javadoc. This commit adds those links.
This chance adds several random test infrastructure improvements that caused
issues in on-going developments but are generally useful. For instance is it impossible
to restart a node with a secure setting source since we close it after the node is started.
This change makes it cloneable such that we can reuse it for a restart.
The following token filters were moved: arabic_stem, brazilian_stem, czech_stem, dutch_stem, french_stem, german_stem and russian_stem.
Relates to #23658
In reindex APIs, when using the `slices` parameter to choose the number of slices, adds the option to specify `slices` as "auto" which will choose a reasonable number of slices. It uses the number of shards in the source index, up to a ceiling. If there is more than one source index, it uses the smallest number of shards among them.
This gives users an easy way to use slicing in these APIs without having to make decisions about how to configure it, as it provides a good-enough configuration for them out of the box. This may become the default behavior for these APIs in the future.
Today we have a `null` invariant on all `ClusterState.Custom`. This makes
several code paths complicated and requires complex state handling in some cases.
This change allows to register a custom supplier that is used to initialize the
initial clusterstate with these transient customs.
The build was ignoring suffixes like "beta1" and "rc1" on the version numbers which was causing the backwards compatibility packaging tests to fail because they expected to be upgrading from 6.0.0 even though they were actually upgrading from 6.0.0-beta1. This adds the suffixes to the information that the build scrapes from Version.java. It then uses those suffixes when it resolves artifacts build from the bwc branch and for testing.
Closes#26017
* Adds ToXContentFragment
This interface is meant for objects that implement `ToXContent` but are not complete objects. It is basically the opposite of `ToXContentObject`. It means that it will be easier to track the migration of classes over to the fragment/not fragment ToXContent model as it will be clear which classes are not migrated. When no classes directly implement `ToXContent` we can make `ToXContent` package private to be sure that all new classes must implement `ToXContentObject` or `ToXContentFragment`.
* review comments
* more review comments
* javadocs
* iter
* Adds tests
* iter
* adds toString test for aggs
* improves tests following review comments
* iter
* iter
We introduced a hack in #25885 to respect the cluster alias if available on the `_index` field. This is important if aggregations or other field data related operations are executed. Yet, we added a small hack that duplicated an implementation detail from the `_index` field data builder to make this work. This change adds a necessary but simple API change that allows us to remove the hack and only have a single implementation.
The goal of this similarity is to help users who would like to keep the
functionality of the `tf-idf` similarity that we want to remove, or to allow
for specific usec-cases (disabling idf, disabling tf, disabling length norm,
etc.) to not have to build a custom plugin and familiarize with the low-level
Lucene API.
Raw requests are supported only by the java yaml test runner and were introduced to test docs snippets. Some yaml tests ended up using them (see #23497) which causes failures for other language clients. This commit migrates those yaml tests to Java tests that send requests through the Java low-level REST client, and also moves the ability to send raw requests to a special client that's only available when testing docs snippets.
Closes#25694
When `refresh=wait_for` is set on an indexing request, we register a listener on the shards that are call during the next refresh. During the recover translog phase, when the engine is open, we have a window of time when indexing operations succeed and they can add their listeners. Those listeners will only be called when the recovery finishes as we do not refresh during recoveries (unless the indexing buffer is full). Next to being a bad user experience, it can also cause deadlocks with an ongoing peer recovery that may wait for those operations to mark the replica in sync (details below).
To fix this, this PR changes refresh listeners to be a noop when the shard is not yet serving reads (implicitly covering the recovery period). It doesn't matter anyway.
Deadlock with recovery:
When finalizing a peer recovery we mark the peer as "in sync". To do so we wait until the peer's local checkpoint is at least as high as the global checkpoint. If an operation with `refresh=wait_for` is added as a listener on that peer during recovery, it is not completed from the perspective of the primary. The primary than may wait for it to complete before advancing the local checkpoint for that peer. Since that peer is not considered in sync, the global checkpoint on the primary can be higher, causing a deadlock. Operation waits for recovery to finish and a refresh to happen. Recovery waits on the operation.
* Adds mutate function to various tests
Relates to #25929
* fix test
* implements mutate function for all single bucket aggs
* review comments
* convert getMutateFunction to mutateIInstance
Currently there is an issue where the send listener is not called in the
nio transport when an exception is throw during channel flush. This
leads to memory leaks. This commit ensures that the listener is called
This commit adds the nio transport as an option in place of the mock tcp
transport for tests. Each test will only use one transport type. The
transport type is decided by a random boolean generated inside of the
`ESTestCase` class.
This commit updates the version for master to 7.0.0-alpha1. It also adds
the 6.1 version constant, and fixes many tests, as well as marking some
as awaits fix.
Closes#25893Closes#25870
ToXContentToBytes is used as a base class that adds toString and buildAsBytes method implementation to classes that implement ToXContent. With the ongoing cleanups, this class is limited and doesn't add a lot of value, given that buildAsBytes can be replaced with XContentHelper.toXContent and toString can be replaced with Strings.toString(this).
The plan would be to remove ToXContentToBytes entirely, and AbstractQueryBuilder is the first place where we can remove its usage.
During peer recoveries, we need to copy over lucene files and replay the operations they miss from the source translog. Guaranteeing that translog files are not cleaned up has seen many iterations overtime. Back in the old 1.0 days, recoveries went through the Engine and actively prevented both translog cleaning and lucene commits. We then moved to a notion called Translog Views, which allowed the recovery code to "acquire" a view into the translog which is then guaranteed to be kept around until the view is closed. The Engine code was free to commit lucene and do what it ever it wanted without coordinating with recoveries. Translog file deletion logic was based on reference counting on the file level. Those counters were incremented when a view was acquired but also when the view was used to create a `Snapshot` that allowed you to read operations from the files. At some point we removed the file based counting complexity in favor of constructs on the Translog level that just keep track of "open" views and the minimum translog generation they refer to. To do so, Views had to be kept around until the last snapshot that was made from them was consumed. This was fine in recovery code but lead to [a subtle bug](https://github.com/elastic/elasticsearch/pull/25862) in the [Primary Replica Resyncer](https://github.com/elastic/elasticsearch/pull/25862).
Concurrently, we have developed the notion of a `TranslogDeletionPolicy` which is responsible for the liveness aspect of translog files. This class makes it very simple to take translog Snapshot into account for keep translog files around, allowing people that just need a snapshot to just take a snapshot and not worry about views and such. Recovery code which actually does need a view can now prevent trimming by acquiring a simple retention lock (a `Closable`). This removes the need for the notion of a View.
* Improves AbstractWireSerializingTestCase equals test
`AbstractWireSerializingTestCase.testEqualsAndHashcode()` now uses `EqualsHashcodeTestUtils` to perform the hashCode and equals checks. To support this `AbstractWireSerializingTestCase` has two new methods: `getCopyFunction()` and `getMutateFunction` which are used when calling `EqualsHashcodeTestUtils`
* Adds TODO
* Makes equivalent changes to AbstractStreamableTestCase
* corrects javadoc error
The following token filters were moved: delimited_payload_filter, keep, keep_types, classic, apostrophe, decimal_digit, fingerprint, min_hash and scandinavian_folding.
Relates to #23658
The Writeble representation is less heavy to parse and that will benefit percolate performance and throughput.
The query builder's binary format has now the same bwc guarentees as the xcontent format.
Added a qa test that verifies that percolator queries written in older versions are still readable by the current version.
This commit fixes tests for environment-aware commands. A previous
change added a check that es.path.conf is not null. The problem is that
this system property is not being set in tests so this check trips every
single time. To fix this, we move the check into a method that can be
overridden, and then override this method in relevant places in tests to
avoid having to set the property in tests. We also add a test that this
check works as expected.
Today we expose `IndexFieldDataService` outside of IndexService to do maintenance
or lookup field data in different ways. Yet, we have a streamlined way to access IndexFieldData
via `QueryShardContext` that should encapsulate all access to it. This also ensures that we control all other functionality like cache clearing etc.
This change also removes the `recycler` option from `ClearIndicesCacheRequest` this option is a no-op and should have been removed long ago.
Currently, NioTransport does start normal socket selectors and the
client when the network server setting is set to false. This commit
makes it so that the client will be started even when the network server
is not enabled.
Additionally, it randomly introduces the NioTransport as an option for
the MockTransportClient throughout tests.
These two methods do do the same thing. The subtle difference between the two is that the former prints out pretty printed content by default while the latter doesn't. There are way more usages of the latter throughout the codebase hence I kept that variant although I do think that it would be much better to print out prettified content by default from a `toString`. That breaks quite some tests so I didn't make that change yet.
Also XContentHelper#toString was outdated as it didn't check the ToXContent#isFragment method to decide whether a new anonymous object has to be created or not. It would simply fail with any ToXContentObject.
Today when we aggregate on the `_index` field the cross cluster search
alias is not taken into account. Neither is it respected when we search
on the field. This change adds support for cluster alias when the cluster
alias is present on the `_index` field.
Closes#25606
Currently we have an option to interrupt the selector thread on close.
This option is not needed as we do not call this method and we should
not be blocking on the network thread. Instead we only need to ever call
wakeup() on the raw selector.
This commit removes all external dependencies from the rest client jar
and shades them in an 'org.elasticsearch.client' package within the jar
using shadowJar gradle plugin. All projects that depended on the
existing jar have been converted to using the 'org.elasticsearch.client'
package prefixes to interact with the rest client.
Closes#25208
Currently we are failing to close socket channels when the initial bind
or connect operation fails. This leaves the file descriptor hanging
around. This closes the channel when an exception occurs during bind or
connect.
Currently an NioChannel is created and it is UNREGISTERED. At some point
it is registered with a selector. From that point on, the channel can
only be closed by the selector. The fact that a channel might not be
associated with a selector has significant implications for concurrency
and the channel shutdown process. The only thing that is simplified by
allowing channels to be in a state independent of a selector is some
testing scenarios.
This PR modifies channels so that they are given a selector at creation
time and are always associated with that selector. Only that selector
can close that channel. This simplifies the channel lifecycle and
closing intricacies.
Removes the primary term from the replication request and pushes it into the transport envelope. This makes it possible to remove the term from the ReplicationOperation universe. The primary term that is to be used for a replication operation is now determined in the reroute phase when the node decides to execute a primary action (and validated once the primary action gets to execute). This makes it possible to validate that the primary action was sent to the correct primary shard instance that it was meant to be sent to (currently we only validate primary actions using the allocation id, which can be reused for failed and reallocated primaries).
When a node tries to join a cluster, it goes through a validation step to make sure the node is compatible with the cluster. Currently we validation that the node can read the cluster state and that it is compatible with the indexes of the cluster. This PR adds validation that the joining node's version is compatible with the versions of existing nodes. Concretely we check that:
1) The node's min compatible version is higher or equal to any node in the cluster (this prevents a too-new node from joining)
2) The node's version is higher or equal to the min compat version of all cluster nodes (this prevents a too old join where, for example, the master is on 5.6, there's another 6.0 node in the cluster and a 5.4 node tries to join).
3) The node's major version is at least as higher as the lowest node in the cluster. This is important as we use the minimum version in the cluster to stop executing bwc code for operations that require multiple nodes. If the nodes are already operating in "new cluster mode", we should prevent nodes from the previous major to join (even if they are wire level compatible). This does mean that if you have a very unlucky partition during the upgrade which partitions all old nodes which are also a minority / data nodes only, the may not be able to re-join the cluster. We feel this edge case risk is well worth the simplification it brings to BWC layers only going one way. This restriction only holds if the cluster state has been recovered (i.e., the cluster has properly formed).
Also, the node join validation can now selectively fail specific nodes (previously the entire batch was failed). This is an important preparation for a follow up PR where we plan to have a rejected joining node die with dignity.
Nhat Nguyen
Tanguy Leroux
Nik Everett
Boaz Leskes
Jason Tedor
Yannick Welsch
Alan Woodward
David Turner
Tim Brooks
Adrien Grand
Daniel Mitterdorfer
Christoph Büscher
Jim Ferenczi
Luca Cavanna
Lee Hinman
Simon Willnauer
Lee Hinman
Martijn van Groningen
Michael Basnight
David Roberts
kel
Colin Goodheart-Smithe
Ryan Ernst
olcbean
Nhat
Md. Abdulla-Al-Sun
Armin Braun
Christoph Büscher
Alexander Reelsen
Tal Levy
Sergey Galkin
Andy Bristol