The ThreadPool#scheduleWithFixedDelay method does not make it clear that all scheduled runnable instances
will be run on the scheduler thread. This becomes problematic if the actions being performed include
blocking operations since there is a single thread and tasks may not get executed due to a blocking task.
This change includes a few different aspects around trying to prevent this situation. The first is that
the scheduleWithFixedDelay method now requires the name of the executor that should be used to execute
the runnable. All existing calls were updated to use Names.SAME to preserve the existing behavior.
The second aspect is the removal of using ScheduledThreadPoolExecutor#scheduleWithFixedDelay in favor of
a custom runnable, ReschedulingRunnable. This runnable encapsulates the logic to deal with rescheduling a
runnable with a fixed delay and mimics the behavior of executing using a ScheduledThreadPoolExecutor and
provides a ScheduledFuture implementation that also mimics that of the typed returned by a
ScheduledThreadPoolExecutor.
Finally, an assertion was added to BaseFuture to detect blocking calls that are being made on the scheduler
thread.
When looking at the logstash template, I noticed that it has definitions for
dynamic temilates with `match_mapping_type` equal to `byte` for instance.
However elasticsearch never tries to find templates that match the byte type
(only long or double as far as numbers are concerned). This commit changes
template parsing in order to ignore bad values of `match_mapping_type` (given
how the logstash template is popular, this would break many upgrades
otherwise). Then I hope to fail the parsing on bad values in 6.0.
We used to mutate it as part of building the aggregation. That
caused assertVersionSerializable to fail because it assumes that
requests aren't mutated after they are sent.
Closes#19481
Primary relocation violates two invariants that ensure proper interaction between document replication and peer recoveries, ultimately leading to documents not being properly replicated.
Invariant 1: Document writes must be replicated based on the routing table of a cluster state that includes all shards which have ongoing or finished recoveries. This is ensured by the fact that do not start a recovery that is not reflected by the cluster state available on the primary node and we always sample a fresh cluster state before starting to replicate write operations.
Invariant 2: Every operation that is not part of the snapshot taken for phase 2, must be succesfully indexed on the target replica (pending shard level errors which will cause the target shard to be failed). To ensure this, we start replicating to the target shard as soon as the recovery start and open it's engine before we take the snapshot. All operations that are indexed after the snapshot was taken are guaranteed to arrive to the shard when it's ready to index them. Note that this also means that the replication doesn't fail a shard if it's not yet ready to recieve operations - it's a normal part of a recovering shard.
With primary relocations, the two invariants can be possibly violated. Let's consider a primary relocating while there is another replica shard recovering from the primary shard.
Invariant 1 can be violated if the target of the primary relocation is so lagging on cluster state processing that it doesn't even know about the new initializing replica. This is very rare in practice as replica recoveries take time to copy all the index files but it is a theoretical gap that surfaces in testing scenarios.
Invariant 2 can be violated even if the target primary knows about the initializing replica. This can happen if the target primary replicates an operation to the intializing shard and that operation arrives to the initializing shard before it opens it's engine but arrives to the primary source after it has taken the snapshot of the translog. Those operations will be currently missed on the new initializing replica.
The fix to reestablish invariant 1 is to ensure that the primary relocation target has a cluster state with all replica recoveries that were successfully started on primary relocation source. The fix to reestablish invariant 2 is to check after opening engine on the replica if the primary has been relocated in the meanwhile and fail the recovery.
Closes#19248
RecoveryTarget increments a reference on the store once it's
created. If we fail to return the instance from the reset method
we leak a reference causing shard locks to not be released. This
change creates the reference in the return statement to ensure no
references are leaked
An initializing replica shard might not have an UnassignedInfo object, for example when it is a relocation target. The method allocatedPostIndexCreate does not account for this situation.
Today when we reset a recovery because of the source not being
ready or the shard is getting removed on the source (for whatever reason)
we wipe all temp files and reset the recovery without respecting any
reference counting or locking etc. all streams are closed and files are
wiped. Yet, this is problematic since we assert that some files are on disk
etc. when we finish writing a file. These assertions don't hold anymore if we
concurrently wipe the tmp files.
This change moves the logic out of RecoveryTarget into RecoveriesCollection which
basically clones the RecoveryTarget on reset instead which allows in-flight operations
to finish gracefully. This means we now have a single path for cleanups in RecoveryTarget
and can safely use assertions in the class since files won't be removed unless the recovery
is either canceled, failed or finished.
Closes #19473
Today they don't because the create index request that is implicitly created
adds an empty mapping for the type of the document. So to Elasticsearch it
looks like this type was explicitly created and `index.mapper.dynamic` is not
checked.
Closes#17592
The test would previously catch Throwable and then decide if it was a critical exception or not. As the catch block was changed from Throwable to Exception this made the test fail for non-critical exceptions. This commit changes the test so that exceptions are only thrown when they're unexpected.
This is the first pipeline aggregation that doesn't have its own
bucket type that needs serializing. It uses InternalHistogram instead.
So that required reworking the new-style `registerAggregation` method
to not require bucket readers. So I built `PipelineAggregationSpec` to
mirror `AggregationSpec`. It allows registering any number of bucket
readers or result readers.
The `client/transport` project adds a new jar build project that
pulls in all dependencies and configures all required modules.
Preinstalled modules are:
* transport-netty
* lang-mustache
* reindex
* percolator
The `TransportClient` classes are still in core
while `TransportClient.Builder` has only a protected construcutor
such that users are redirected to use the new `TransportClientBuilder`
from the new jar.
Closes#19412
Previously if the size of the search request was greater than zero we would not cache the request in the request cache.
This change retains the default behaviour of not caching requests with size > 0 but also allows the `request_cache=true` query parameter
to enable the cache for requests with size > 0
This is a tentative to revive #15939 motivated by elastic/beats#1941.
Half-floats are a pretty bad option for storing percentages. They would likely
require 2 bytes all the time while they don't need more than one byte.
So this PR exposes a new `scaled_float` type that requires a `scaling_factor`
and internally indexes `value*scaling_factor` in a long field. Compared to the
original PR it exposes a lower-level API so that the trade-offs are clearer and
avoids any reference to fixed precision that might imply that this type is more
accurate (actually it is *less* accurate).
In addition to being more space-efficient for some use-cases that beats is
interested in, this is also faster that `half_float` unless we can improve the
efficiency of decoding half-float bits (which is currently done using software)
or until Java gets first-class support for half-floats.
Today the default precision for the cardinality aggregation depends on how many
parent bucket aggregations it had. The reasoning was that the more parent bucket
aggregations, the more buckets the cardinality had to be computed on. And this
number could be huge depending on what the parent aggregations actually are.
However now that we run terms aggregations in breadth-first mode by default when
there are sub aggregations, it is less likely that we have to run the cardinality
aggregation on kagilions of buckets. So we could use a static default, which will
be less confusing to users.
We currently have concurrency issue between the static methods on the Store class and store changes that are done via a valid open store. An example of this is the async shard fetch which can reach out to a node while a local shard copy is shutting down (the fetch does check if we have an open shard and tries to use that first, but if the shard is shutting down, it will not be available from IndexService).
Specifically, async shard fetching tries to read metadata from store, concurrently the shard that shuts down commits to lucene, changing the segments_N file. this causes a file not find exception on the shard fetching side. That one in turns makes the master think the shard is unusable. In tests this can cause the shard assignment to be delayed (up to 1m) which fails tests. See https://elasticsearch-ci.elastic.co/job/elastic+elasticsearch+master+java9-periodic/570 for details.
This is one of the things #18938 caused to bubble up.
* Removed `Template` class and unified script & template parsing logic. Templates are scripts, so they should be defined as a script. Unless there will be separate template infrastructure, templates should share as much code as possible with scripts.
* Removed ScriptParseException in favour for ElasticsearchParseException
* Moved TemplateQueryBuilder to lang-mustache module because this query is hard coded to work with mustache only
integration tests, as they are no longer needed with
index creation now waiting for shards to be started before
returning from the index creation call (by default, it waits
for the primary of each shard to be started before returning,
which is what ensureYellow() was ensuring anyway).
Closes#19452
Relates #19450
This commit adds a log message when bootstrap checks are enforced
informing the user that they are enforced because they are bound to an
external network interface. We also log if bootstrap checks are being
enforced but system checks are being ignored.
Relates #19451
The unit tests for IndicesClusterStateService currently inject random failures upon shard creation/ routing upate / mapping update etc. This commit makes injecting failures optional so that stronger assertions can be made about the local indices / shard state in case of no failures.
Before returning, index creation now waits for the configured number
of shard copies to be started. In the past, a client would create an
index and then potentially have to check the cluster health to wait
to execute write operations. With the cluster health semantics changing
so that index creation does not cause the cluster health to go RED,
this change enables waiting for the desired number of active shards
to be active before returning from index creation.
Relates #9126
For historical reasons, the value associated with Priority.IMMEDIATE is
-1. Yet, with a full-cluster restart required on major version upgrades,
we can reset these values so they are conceptually simpler. This commit
resets the values associated with Priority instances.
Today we have an abstraction Priority for representing
priorities. Ideally, these values are a fixed set of constants with a
well-defined ordering which sounds perfect for an enum. This commit
changes Priority so that it is an enum instead of a class.
In Priority there is a field named values that represents an ordered, by
priority, list of all priorities. Yet, this collection is modifiable and
this collection is exposed via the public API. This means that consumers
can modify this list potentially leading to complete chaos. This commit
modifies this field so that it is unmodifiable, documents that the
returned collection is unmodifiable, and returns total order to the
world. We also punish the bad consumer here by making them make a copy
of the returned collection with which they can do as they please. This
fixes a puzzling test failure which only arises if the two tests
(PrioritizedExecutorsTests#testPriorityQueue and
PriorityTests#testCompareTo run in the same JVM, and run in the right
order).
Relates #19447
Also introduced a `Processor.Parameters` class that is holder for several services processors rely on,
the IngestPlugin#getProcessors(...) method has been changed to accept `Processor.Parameters` instead
of each service seperately.
Today we log all loaded modules and installed plugins in a single
line. The number of modules has grown, and when plugins are installed a
single log line containing the loaded modules and plugins is
lengthy. With this commit, we log a single module or plugin per line,
log these in sorted order, and also log if no modules or no plugins were
loaded.
Relates #19441
This commit renames the Netty 3 transport module from transport-netty to
transport-netty3. This is to make room for a Netty 4 transport module,
transport-netty4.
Relates #19439
Currently custom headers that should be passed through rest requests are
registered by depending on the RestController in guice and calling a
registration method. This change moves that registration to a getter for
plugins, and makes the RestController take the set of headers on
construction.
Invocation counts can be used to help judge the selectivity of individual query components in the context of the entire query. E.g. a query may not look selective when run by itself (matches most of the index), but when run in context of a full search request, is evaluated only rarely due to execution order
Since this is modifying the base timing class, it'll enrich both query and agg profiles (as well as future profile results)
Today `node.mode` and `node.local` serve almost the same purpose, they
are a shortcut for `discovery.type` and `transport.type`. If `node.local: true`
or `node.mode: local` is set elasticsearch will start in _local_ mode which means
only nodes within the same JVM are discovered and a non-network based transport
is used. The _local_ mode it only really used in tests or if nodes are embedded.
For both, embedding and tests explicit configuration via `discovery.type` and `transport.type`
should be preferred.
This change removes all the usage of these settings and by-default doesn't
configure a default transport implemenation since netty is now a module. Yet, to make
the user expericence flawless, plugins or modules can set a `http.type.default` and
`transport.type.default`. Plugins set this via `PluginService#additionalSettings()`
which enforces _set-once_ which prevents node startup if set multiple times. This means
that our distributions will just startup with netty transport since it's packaged as a
module unless `transport.type` or `http.transport.type` is explicitly set.
This change also found a bunch of bugs since several NamedWriteables were not registered if a
transport client is used. Now that we don't rely on the `node.mode` leniency which is inherited
instead of using explicit settings, `TransportClient` uses `AssertingLocalTransport` which detects these problems since it serializes all messages.
Closes#16234
The Java API supports this while mostly used for tests it can also be useful in
production environments. For instance if something is automated like a settings change
and we execute some health right after it the settings update might have some consequences
like a reroute which hasn't been fully applied since the preconditions are not fulfilled yet.
For instance if not all shards started the settings update is applied but the reroute won't move
currently initializing shards like in the shrink API test. Sure this could be done by waiting for
green before but if the cluster moves shards due to some side-effects waiting for all events is
still useful. I also took the chance to add unittests to Priority.java
Closes#19419
This adds an extra method, registerWithDeprecatedHandler, to register both a normal handler and a deprecated handler at the same time. This helps with renaming methods as opposed to _just_ deprecated methods.
Today we assert that the tmp files are present but if the recovery
was canceled this might not be the case while still a valid state.
This chance only throws the AssertionError if the recovery is still active.
This moves all netty related code into modules/transport-netty the module is build as a zip file as well as a JAR to serve as a dependency for transport client. For the time being this is required otherwise we have no network based impl. for transport client users. This might be subject to change given that we move forward http client.
* Clean up the generics around significant terms aggregation results
* Reduce code duplicated between `SignificantLongTerms` and
`SignificantStringTerms` by creating `InternalMappedSignificantTerms`
and moving common things there where possible.
* Migrate to `NamedWriteable`
* Line length fixes while I was there
This commit removes support for properties syntax and config files:
- removed support for elasticsearch.properties
- removed support for logging.properties
- removed support for properties content detection in REST APIs
- removed support for properties content detection in Java API
Relates #19398
Some tests still start http implicitly or miss configuring the transport clients correctly.
This commit fixes all remaining tests and adds a depdenceny to `transport-netty` from
`qa/smoke-test-http` and `modules/reindex` since they need an http server running on the nodes.
This also moves all required permissions for netty into it's module and out of core.
If a nested, has_child or has_parent query's inner query gets rewritten then the InnerHitBuilder should use that rewritten form too, otherwise this can cause exceptions in a later phase.
Also fixes a bug that HasChildQueryBuilder's rewrite method overwrites max_children with min_children value.
Closes#19353
That exception is currently serialized as its current base class IllegalStateException which confuses code supposed to deal with the stepping down of a master. This is an important exception and we should be able to serialize it correctly. This commit fixes it by moving the exception to inherit from ElasticsearchException and properly register it.
As a bonus I adapted CapturingTransport to properly simulate serialized exceptions.
Switches most search behavior extensions from push (`onModule(SearchModule)`)
to pull (`implements SearchPlugin`). This effort in general gives plugin
authors a much cleaner view of how to extend Elasticsearch and starts to
set up portions of Elasticsearch as "the plugin API". This commit in
particular does that for search-time behavior like customized suggesters,
highlighters, score functions, and significance heuristics.
It also switches most such customization to being done at search module
construction time which is much, much easier to reason about from a testing
perspective. It also helps significantly in the process of de-guice-ing
Elasticsearch's startup.
There are at least two major search time extensions that aren't covered in
this commit that will simply have to wait for the next commit on the topic
because this one has already grown large: custom aggregations and custom
queries. These will likely live in the same SearchPlugin interface as well.
This change adds a createComponents() method to Plugin implementations
which they can use to return already constructed componenents/services.
Eventually this should be just services ("components" don't really do
anything), but for now it allows any object so that preconstructed
instances by plugins can still be bound to guice. Over time we should
add basic services as arguments to this method, but for now I have left
it empty so as to not presume what is a necessary service.
If the allocation decision for a primary shard was NO, this should
cause the cluster health for the shard to go RED, even if the shard
belongs to a newly created index or is part of cluster recovery.
Relates #9126
Previously, index creation would momentarily cause the cluster health to
go RED, because the primaries were still being assigned and activated.
This commit ensures that when an index is created or an index is being
recovered during cluster recovery and it does not have any active
allocation ids, then the cluster health status will not go RED, but
instead be YELLOW.
Relates #9126
this commit moves the most of the http related integ tests out into it's own
`qa/smoke-test-http` project where most of the test can run against the external cluster.
Today when a node is removed the cluster (it leaves or it fails), we
submit a cluster state update task. These cluster state update tasks are
processed serially on the master. When nodes are removed en masse (e.g.,
a rack is taken down or otherwise becomes unavailable), the master will
be slow to process these failures because of the resulting reroutes and
publishing of each subsequent cluster state. We improve this in this
commit by processing the node removals using the cluster state update
task batch processing framework.
Relates #19289
Today we have a bunch of tests that use netty transport for several reasons
these tests use it because they need to run some tcp based transport. Yet, this
couples our tests tightly to the netty implementation which should be tested on it's own.
This change adds a plain socket based blocking TcpTransport implementation that is used by
default in tests if local transport is suppressed or if network is selected.
It also adds another tcp network implementation as a showcase how the interface works.
Lucene IndexWriter asserts on files existing on the filesystem but
some tests throw IOException explicitly on those operatiosn such that
some tests trip asserts. We had this before on InternalEngine#ctor
and added some logic there to catch only a specific assertions based
on some excepition stack analysis. This change applies the same logic
to the IndexWriter#commit part of the engine since it can hit the same
issue.
This also fixes a self-suppression issue in Store.java.
Closes#19356
Several tests required http.enabled where it was unnecessary.
We also had RestMainActionIT which tests what two of our REST tests
test already so I removed it.
The explicit use of http.enabled: false is also obsolet since our
test do that by default.
The DiscoveryNodeService exists to register CustomNodeAttributes which
plugins can add. This is not necessary, since plugins can already add
additional attributes, and use the node attributes prefix.
This change removes the DiscoveryNodeService, and converts the only
consumer, the ec2 discovery plugin, to add the ec2 availability zone
in additionalSettings().
This change removes the ability for guice to have child injectors (and
the entire concept of parent injectors) from our fork of guice. The
methodology for removing was simple: I removed createChildInjector, and
continued to remove methods and members that were unused until my head
was spinning. The motivation for this change is to limit what our fork
of guice gives us access to, so we don't regress and start adding back
more complicated uses.
This adds a test that uses transport implementation and sends random requests
to 3 different nodes, the request handlers maybe forwarding the requests to yet another node
etc. until returning the response. This test basically tests that nodes are not deadlocking
in a distributed fashion.
Repository plugins currently use a lot of custom classes like
RepositoryName and RepositorySettings in order to use guice to construct
repository implementations. But repositories now only really need their
settings to be constructed. Anything else they need (eg a cloud client)
can be constructed within the plugin, instead of via guice.
This change makes repository plugins use the new pull model. It removes
guice from the construction of Repository objects (no more child
injectors) and also from all repository plugins.
This exposes a method to start an action and return a task from
`NodeClient`. This allows reindex to use the injected `Client` rather
than require injecting `TransportAction`s
Today when a thread encounters a fatal unrecoverable error that
threatens the stability of the JVM, Elasticsearch marches on. This
includes out of memory errors, stack overflow errors and other errors
that leave the JVM in a questionable state. Instead, the Elasticsearch
JVM should die when these errors are encountered. This commit causes
this to be the case.
Relates #19272
After #13834 many tests that used Groovy scripts (for good or bad reason) in their tests have been moved in the lang-groovy module and the issue #13837 has been created to track these messy tests in order to clean them up.
This commit moves more tests back in core, removes the dependency on Groovy, changes the scripts in order to use the mocked script engine, and change the tests to integration tests.
This commit moves back some messy tests that have been placed in lang-groovy module in https://github.com/elastic/elasticsearch/pull/13834. It removes the dependency on Groovy plugin as well as change back the tests to integration tests (IT suffix).
It also changes the current MockScriptEngine and MockScriptPlugin to make it easier to use.
The api for snapshot/restore was split up between two interfaces,
Repository and IndexShardRepository. There was also complex
initialization and injection between the two. However, there is always a
one to one relationship between the two.
This change moves the IndexShardRepository api into Repository, as well
as updates the API so as not to require any services to be injected for
sublcasses.
This adds a new proxy for RestHandlers and RestControllers so that requests made
to deprecated REST APIs can be automatically logged in the ES logs via the
DeprecationLogger as well as via a "Warning" header (RFC-7234) for all responses.
Calling indicesService.deleteIndex() can trip an assertion if there is an ongoing cluster state applied in
IndicesClusterStateService. This means that the index is possibly deleted after the failMissingShards
check and before we try creating new and updated shards, tripping an assertion that non-existing shards must
have shard state initializing (started in this case).
This change adds the type of the field in the fieldstats response.
It can be one of the following:
* "integer" for byte, short, integer and long
* "float" for float, half-float and double
* "date" for date
* "ip" for ip
* "text" for string, keyword and text.
Closes#17750
This adds a remote option to reindex that looks like
```
curl -POST 'localhost:9200/_reindex?pretty' -d'{
"source": {
"remote": {
"host": "http://otherhost:9200"
},
"index": "target",
"query": {
"match": {
"foo": "bar"
}
}
},
"dest": {
"index": "target"
}
}'
```
This reindex has all of the features of local reindex:
* Using queries to filter what is copied
* Retry on rejection
* Throttle/rethottle
The big advantage of this version is that it goes over the HTTP API
which can be made backwards compatible.
Some things are different:
The query field is sent directly to the other node rather than parsed
on the coordinating node. This should allow it to support constructs
that are invalid on the coordinating node but are valid on the target
node. Mostly, that means old syntax.
This commit renames writeThrowable to writeException. The situation here
stems from the fact that the StreamOutput method for serializing
Exceptions needs to accept Throwables too as Throwables can be the cause
of serialized Exceptions. Yet, we do not serialize Throwables in the
Error sub-hierarchy in a way that they can be deserialized into their
initial type. This leads to an asymmetry in the StreamOutput method for
serializing Exceptions and the StreamInput method for writing
Excpetions. Namely, the former will accept Throwables but the latter
will only return Exceptions. A goal with the stream methods has always
been symmetry in the method names so that serialization/deserialization
routines appear symmetrical in code. It is this asymmetry on the
input/output types for Exceptions on StreamOutput/StreamInput that
clashes with the desired symmetry of naming. Despite this, we should
favor symmetry in the naming of the methods. This commit renames
StreamOutput#writeThrowable to StreamOutput#writeException which leaves
us with Exception StreamInput#readException and void
StreamOutput#writeException(Throwable).
This commit modifies the initial value of the transport client
round-robin index to a random value so that initial requests are more
likely to not all hit the same node.
Relates #14143
Due to some optimization on the netty layer we had quite some code / cruft
added to the TcpTransport to allow for those optimizations. After cleaning
up BytesReference we can now move this optimization into TcpTransport and
have a simple send method on the implementation layer instead. This commit
adds a CompositeBytesReference that also allows message headers to be written
separately which simplify the header code as well since no skips are needed
anymore.
The top-level class Throwable represents all errors and exceptions in
Java. This hierarchy is divided into Error and Exception, the former
being serious problems that applications should not try to catch and the
latter representing exceptional conditions that an application might
want to catch and handle. This commit renames
org.elasticsearch.cli.UserError to org.elasticsearch.UserException to
make its name consistent with where it falls in this hierarchy.
Relates #19254
Today when reading a malformed operation from the translog, we throw an
assertion error that is immediately caught and wrapped into a translog
corrupted exception. This commit replaces this by electing to directly
throw a translog corrupted exception instead.
Additionally, this cleanup also addressed a double-wrapped translog
corrupted exception. Namely, verifying the checksum can throw a translog
corrupted exception which the existing code would catch and wrap again
in a translog corrupted exception.
Relates #19256
We've been slowly improving batch support in `ClusterService` so service won't need to implement this tricky logic themselves. These good changes are blessed but our logging infra didn't catch up and we now log things like:
```
[2016-07-04 21:51:22,318][DEBUG][cluster.service ] [node_sm0] processing [put-mapping [type1],put-mapping [type1]]:
```
Depending on the `source` string this can get quite ugly (mostly in the ZenDiscovery area).
This PR adds some infra to improve logging, keeping the non-batched task the same. As result the above line looks like:
```
[2016-07-04 21:44:45,047][DEBUG][cluster.service ] [node_s0] processing [put-mapping[type0, type0, type0]]: execute
```
ZenDiscovery waiting on join moved from:
```
[2016-07-04 17:09:45,111][DEBUG][cluster.service ] [node_t0] processing [elected_as_master, [1] nodes joined),elected_as_master, [1] nodes joined)]: execute
```
To
```
[2016-07-04 22:03:30,142][DEBUG][cluster.service ] [node_t3] processing [elected_as_master ([3] nodes joined)[{node_t2}{R3hu3uoSQee0B6bkuw8pjw}{p9n28HDJQdiDMdh3tjxA5g}{127.0.0.1}{127.0.0.1:30107}, {node_t1}{ynYQfk7uR8qR5wKIysFlQg}{wa_OKuJHSl-Oyl9Gis-GXg}{127.0.0.1}{127.0.0.1:30106}, {node_t0}{pweq-2T4TlKPrEVAVW6bJw}{NPBSLXSTTguT1So0JsZY8g}{127.0.0.1}{127.0.0.1:30105}]]: execute
```
As a bonus, I removed all `zen-disco` prefixes to sources from that area.
Node IDs are currently randomly generated during node startup. That means they change every time the node is restarted. While this doesn't matter for ES proper, it makes it hard for external services to track nodes. Another, more minor, side effect is that indexing the output of, say, the node stats API results in creating new fields due to node ID being used as keys.
The first approach I considered was to use the node's published address as the base for the id. We already [treat nodes with the same address as the same](https://github.com/elastic/elasticsearch/blob/master/core/src/main/java/org/elasticsearch/discovery/zen/NodeJoinController.java#L387) so this is a simple change (see [here](https://github.com/elastic/elasticsearch/compare/master...bleskes:node_persistent_id_based_on_address)). While this is simple and it works for probably most cases, it is not perfect. For example, if after a node restart, the node is not able to bind to the same port (because it's not yet freed by the OS), it will cause the node to still change identity. Also in environments where the host IP can change due to a host restart, identity will not be the same.
Due to those limitation, I opted to go with a different approach where the node id will be persisted in the node's data folder. This has the upside of connecting the id to the nodes data. It also means that the host can be adapted in any way (replace network cards, attach storage to a new VM). I
It does however also have downsides - we now run the risk of two nodes having the same id, if someone copies clones a data folder from one node to another. To mitigate this I changed the semantics of the protection against multiple nodes with the same address to be stricter - it will now reject the incoming join if a node exists with the same id but a different address. Note that if the existing node doesn't respond to pings (i.e., it's not alive) it will be removed and the new node will be accepted when it tries another join.
Last, and most importantly, this change requires that *all* nodes persist data to disk. This is a change from current behavior where only data & master nodes store local files. This is the main reason for marking this PR as breaking.
Other less important notes:
- DummyTransportAddress is removed as we need a unique network address per node. Use `LocalTransportAddress.buildUnique()` instead.
- I renamed `node.add_lid_to_custom_path` to `node.add_lock_id_to_custom_path` to avoid confusion with the node ID which is now part of the `NodeEnvironment` logic.
- I removed the `version` paramater from `MetaDataStateFormat#write` , it wasn't really used and was just in the way :)
- TribeNodes are special in the sense that they do start multiple sub-nodes (previously known as client nodes). Those sub-nodes do not store local files but derive their ID from the parent node id, so they are generated consistently.
Once all of these are migrated we'll be able to remove aggregation's
custom "streams" which function that same as NamedWriteable. It also
allows us to make most of the fields on aggregations final which is
rather nice.
Also starts to migrate MultiBucketAggregation.Bucket to Writeable,
allowing the buckets to have immutable parts.
Once all of these are migrated we'll be able to remove aggregation's
custom "streams" which function that same as NamedWriteable. It also
allows us to make most of the fields on aggregations final which is
rather nice.
Today throughout the codebase, catch throwable is used with reckless
abandon. This is dangerous because the throwable could be a fatal
virtual machine error resulting from an internal error in the JVM, or an
out of memory error or a stack overflow error that leaves the virtual
machine in an unstable and unpredictable state. This commit removes
catch throwable from the codebase and removes the temptation to use it
by modifying listener APIs to receive instances of Exception instead of
the top-level Throwable.
Relates #19231
Rename `fields` to `stored_fields` and add `docvalue_fields`
`stored_fields` parameter will no longer try to retrieve fields from the _source but will only return stored fields.
`fields` will throw an exception if the user uses it.
Add `docvalue_fields` as an adjunct to `fielddata_fields` which is deprecated. `docvalue_fields` will try to load the value from the docvalue and fallback to fielddata cache if docvalues are not enabled on that field.
Closes#18943
With this commit we also propagate the `canTripCircuitBreaker`
setting for the main action in TransportBroadcastByNodeAction.
Previously, we set it only on the additional action added by
this handler.
This change removes a handful of classes and methods that were simply
unused. Some of the classes were intermediate abstract classes that
added nothing to the base class they extended.
Primary relocation and indexing concurrently can currently lead to a deadlock situation as indexing operations are blocked on a (bounded) thread pool during the hand-off phase between old and new primary. This change replaces blocking of indexing operations by putting operations that cannot be executed during relocation hand-off in a queue to be executed once relocation completes.
Closes#18553.
The only reason for LifecycleComponent taking a generic type was so that
it could return that type on its start and stop methods. However, this
chaining has no practical necessity. Instead, start and stop can be
void, and a whole bunch of confusing generics disappear.
Before, a repository would maintain an index file (named 'index') per
repository, that contained the current snapshots in the repository.
This file was not atomically written, so repositories had to depend on
listing the blobs in the repository to determine what the current
snapshots are, and only rely on the index file if the repository does
not support the listBlobs operation. This could cause an incorrect view
of the current snapshots in the repository if any prior snapshot delete
operations failed to delete snapshot metadata files.
This commit introduces the atomic writing of the index file, and because
atomic writes are not guaranteed if the file already exists, we write to
a generational index file (index-N, where N is the current generation).
We also maintain an index-latest file that contains the current
generation, for those repositories that cannot list blobs.
Closes#19002
Relates #18156
These are the first aggregations with multiple `InternalAggregation`s
backing the same `AggregationBuilder`. This required a change in the
register method's signature.
NodeService has an "service attributes" map, which is only
set by HttpServer on start/stop. But the only thing it puts in this map
is already available as part of the HttpServer info which is added to
node info requests. This change removes the attributes map and removes
the dependency in HttpServer on NodeService.
BytesReference should be a really simple interface, yet it has a gazillion
ways to achieve the same this. Methods like `#hasArray`, `#toBytesArray`, `#copyBytesArray`
`#toBytesRef` `#bytes` are all really duplicates. This change simplifies the interface
dramatically and makes implementations of it much simpler. All array access has been removed
and is streamlined through a single `#toBytesRef` method. Utility methods to materialize a
compact byte array has been added too for convenience.
Migrates the `stats` and `extended_stats` aggregations and pipeline
aggregations from the special purpose aggregations streams to
`NamedWriteable`. These are the first pipeline aggregations so this
adds the infrastructure to support both streams and `NamedWriteable`s
for pipeline aggregations.
Raise IOException on deleteBlob if the blob doesn't exist
This commit raises an IOException on BlobContainer#deleteBlob
if the blob does not exist, in conformance with the BlobContainer
interface contract. Each implementation of BlobContainer now
conforms to this contract (file system, S3, Azure, HDFS). This
commit also contains blob container tests for each of the
repository implementations.
Closes#18530
We'll migrate to NamedWriteable so we can share code with the rest
of the system. So we can work on this in multiple pull requests without
breaking Elasticsearch in between the commits this change supports
*both* old style `InternalAggregations.stream` serialization and
`NamedWriteable` style serialization. As such it creates about a
half dozen `// NORELEASE` comments that will have to be removed
once the migration is complete.
This also introduces a boolean `transportClient` flag to `SearchModule`
which is used to skip inappropriate registrations for for the
transport client while still registering the things it needs. In
this case that means that the `InternalAggregation` subclasses are
registered with the `NamedWriteableRegistry` but the `AggregationBuilder`
subclasses are not.
Finally, this moves aggregation registration from guice configuration
time to `SearchModule` construction time. This will make it simpler to
work with in the future as we further clean up Elasticsearch's
extension points.
We have long worked to capture different partitioning scenarios in our testing infra. This PR adds a new variant, inspired by the Jepsen blogs, which was forgotten far - namely a partition where one node can still see and be seen by all other nodes. It also updates the resiliency page to better reflect all the work that was done in this area.
Currently there are cases when using TimeIntervalRounding#round() and date1 <
date2 that round(date2) < round(date1). These errors can happen when using a
non-fixed time zone and the values to be rounded are slightly after a time zone
offset change (e.g. DST transition).
Here is an example for the "CET" time zone with a 45 minute rounding interval.
The dates to be rounded are on the left (with utc time stamp), the rounded
values on the right. The error case is marked:
2011-10-30T01:40:00.000+02:00 1319931600000 | 2011-10-30T01:30:00.000+02:00 1319931000000
2011-10-30T02:02:30.000+02:00 1319932950000 | 2011-10-30T01:30:00.000+02:00 1319931000000
2011-10-30T02:25:00.000+02:00 1319934300000 | 2011-10-30T02:15:00.000+02:00 1319933700000
2011-10-30T02:47:30.000+02:00 1319935650000 | 2011-10-30T02:15:00.000+02:00 1319933700000
2011-10-30T02:10:00.000+01:00 1319937000000 | 2011-10-30T01:30:00.000+02:00 1319931000000 *
2011-10-30T02:32:30.000+01:00 1319938350000 | 2011-10-30T02:15:00.000+01:00 1319937300000
2011-10-30T02:55:00.000+01:00 1319939700000 | 2011-10-30T02:15:00.000+01:00 1319937300000
2011-10-30T03:17:30.000+01:00 1319941050000 | 2011-10-30T03:00:00.000+01:00 1319940000000
We should correct this by detecting that we are crossing a transition when
rounding, and in that case pick the largest valid rounded value before the
transition.
This change adds this correction logic to the rounding function and adds this
invariant to the randomized TimeIntervalRounding tests. Also adding the example
test case from above (with corrected behaviour) for illustrative purposes.
Today we have a ton of logic inside the NettyTransport* codebase. The footprint
of the code that has a direct netty dependency is large and alternative implementations
are pretty hard today since they need to know all about our proticol etc.
This change moves most of the code into TCPTransport* baseclasses and moves all
the protocol send code together. The base classes now contain the majority of the logic
while NettyTransport* classes remain to implement the glue code, configuration and optimization.
The factory for ingest processor is generic, but that is only for the
return type of the create mehtod. However, the actual consumer of the
factories only cares about Processor, so generics are not needed.
This change removes the generic type from the factory. It also removes
AbstractProcessorFactory which only existed in order pull the optional
tag from config. This functionality is moved to the caller of the
factories in ConfigurationUtil, and the create method now takes the tag.
This allows the covariant return of the implementation to work with
tests not needing casts.
Previously all rest handlers would take Client in their injected ctor.
However, it was only to hold the client around for runtime. Instead,
this can be done just once in the HttpService which handles rest
requests, and passed along through the handleRequest method. It also
should always be a NodeClient, and other types of Clients (eg a
TransportClient) would not work anyways (and some handlers can be
simplified in follow ups like reindex by taking NodeClient).
jaymode
Adrien Grand
Nik Everett
Yannick Welsch
Simon Willnauer
Martijn van Groningen
Simon Willnauer
Colin Goodheart-Smithe
Boaz Leskes
Ali Beyad
Jason Tedor
Ryan Ernst
Zachary Tong
Mathias Fussenegger
Tal Levy
gfyoung
Chris Earle
Lee Hinman
Britta Weber
Christoph Büscher
Nicholas Knize
javanna
Tanguy Leroux
Clinton Gormley
Alexander Reelsen
Jim Ferenczi
Igor Motov
Justin Patrin
LeeDr
Daniel Mitterdorfer