This adds a new boolean (index.merge.scheduler.auto_throttle) dynamic
setting, default true (matching Lucene), to adaptively set the IO rate
limit for merges over time.
This is more flexible than the previous fixed rate throttling because
it responds depending on the incoming merge rate, so search-heavy
applications that are not doing much indexing will see merges heavily
throttled while indexing-heavy cases will lighten the throttle so
merges can keep up within incoming indexing.
The fixed rate throttling is still available as a fallback if things
go horribly wrong.
Closes#9243Closes#9133
previous push was partial by mistake, we still need the wrapped dirs around after being closed for the test infra, for now, explicitly clear it in the leak test (which is still bad apple)
I ran the bad apple test for index memory leaks and still saw leaks, it seems like we don't properly clean the dirs from the static mock test dir wrapper
`cluster.routing.allocation.disk.include_relocations` and
`cluster.routing.allocation.disk.reroute_interval` are both boolean
settings, so they should have the `Validator.BOOLEAN` applied.
Fixes#9309
validation tests for constants
Currently the snapshot flag for Version constants is only set to true
for CURRENT. However, this means that the snapshot state changes from
branch to branch. Instead, snapshot should be "is this version
released?". This change also adds a validation test checking that
ID -> constant and vice versa are correct, and fixes one bug found there
(for an unreleased version).
- don't allow for soft references anymore in the recycler
- remove some abusive thread locals
- don't recycle independently float/double and int/long pages, they are the
same and just interpret bits differently.
Close#9272
BucketAggregationMode used to be part of the framework, now it's only an
implementation detail of the terms, histogram, geohash grid and scripted
aggregators.
Aggregator.estimatedBucketCount() was a complicated way to do the initial sizing
of the data structures, but it did not work very well in practice and was rather
likely to lead to over-sized data-structures causing OOMEs. It's removed now and
all data-structures start with a size of 1 and grow exponentially.
Aggregator.preCollection() is now symetric with postCollection(): it exists on
all aggregation objects where postCollection() also is and recursively calls its
children.
Fixed other minor issues related to generics and exceptions.
Close#9097
The query cache has a mechanism that disables it automatically when
SearchContext.nowInMillis() is used. One issue with that is that the date math
parser always evaluates the current timestamp when parsing a date, even if it
is not needed. As a consequence, whenever you use a date expression in your
queries, the query cache would not be used.
Close#9225
Some of our Java API requests have public setters but their corresponding getters are package private only. This commit makes those getters public as well.
Closes#9273
This change fixes _timestamp's serialization method to write out
`doc_values` and `doc_values_format`, which could already be set,
but would not be written out.
closes#8893closes#8967
Today we give the HTTP status back within the HTTP response itself and within the JSON response as well:
```sh
curl localhost:9200/
```
```js
{
"status" : 200,
"name" : "Red Wolf",
"version" : {
"number" : "2.0.0",
"build_hash" : "6837a61d8a646a2ac7dc8da1ab3c4ab85d60882d",
"build_timestamp" : "2014-08-19T13:55:56Z",
"build_snapshot" : true,
"lucene_version" : "4.9"
},
"tagline" : "You Know, for Search"
}
```
This commits adds a test that simulate disconnecting nodes and dropping requests during the various stages of recovery and solves all the issues that were raised by it. In short:
1) On going recoveries will be scheduled for retry upon network disconnect. The default retry period is 5s (cross node connections are checked every 10s by default).
2) Sometimes the disconnect happens after the target engine has started (but the shard is still in recovery). For simplicity, I opted to restart the recovery from scratch (where little to no files will be copied again, because there were just synced).
3) To protected against dropped requests, a Recovery Monitor was added that fails a recovery if no progress has been made in the last 30m (by default), which is equivalent to the long time outs we use in recovery requests.
4) When a shard fails on a node, we try to assign it to another node. If no such node is available, the shard will remain unassigned, causing the target node to clean any in memory state for it (files on disk remain). At the moment the shard will remain unassigned until another cluster state change happens, which will re-assigned it to the node in question but if no such change happens the shard will remain stuck at unassigned. The commits adds an extra delayed reroute in such cases to make sure the shard will be reassinged
5) Moved all recovery related settings to the RecoverySettings.
Closes#8720
You can now specify `format` in the request definition for most numeric metric aggregations. The exceptions are Percentile_Ranks, Cardinality and Value_Count as the response type of these can be different from the field type so the formatter won't work.
Closes#6812
Today the internal engine closes itself it the engine hits an exception
it can not recover from. This complicates a lot of refcounting issues
if such an exception happens during engine creation. This commit
only markes the engine as failed and let the user close it once the exception
bubbles up. Additionally it rolls back the indexwriter to prevent any changes after
the engine is failed.
I have a field with a `null` [default `_timestamp` value](http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/mapping-timestamp-field.html#mapping-timestamp-field-default) and when I try to update the mapping I get a server error caused by a `NullPointerException`
```
[2015-01-08 17:28:56,040][DEBUG][action.admin.indices.mapping.put] [...] failed to put mappings on indices [[feed_170_v1, feed_204_v1, feed_229_v1, feed_232_v1, feed_239_v1, feed_248_v1, feed_268_v1, feed_256_v1, feed_272_v1, feed_159_v1, feed_255_v1, feed_164_v1, feed_259_v1, feed_266_v1, feed_188_v1, feed_240_v1, feed_233_v1, feed_13_v1, feed_184_v1, feed_261_v1, feed_267_v1, feed_271_v1, feed_257_v1, feed_172_v1, feed_238_v1, feed_254_v1, feed_223_v1, feed_274_v1, feed_203_v1, feed_269_v1, feed_262_v1, feed_205_v1, feed_168_v1, feed_219_v1, feed_253_v1, feed_251_v1, feed_173_v1, feed_252_v1, feed_210_v1, feed_216_v1, feed_218_v1, feed_118_v1, feed_273_v1, feed_227_v1, feed_166_v1, feed_213_v1, feed_226_v1]], type [history]
java.lang.NullPointerException
at org.elasticsearch.index.mapper.internal.TimestampFieldMapper.merge(TimestampFieldMapper.java:287)
at org.elasticsearch.index.mapper.object.ObjectMapper.merge(ObjectMapper.java:936)
at org.elasticsearch.index.mapper.DocumentMapper.merge(DocumentMapper.java:693)
at org.elasticsearch.cluster.metadata.MetaDataMappingService$4.execute(MetaDataMappingService.java:508)
at org.elasticsearch.cluster.service.InternalClusterService$UpdateTask.run(InternalClusterService.java:329)
at org.elasticsearch.common.util.concurrent.PrioritizedEsThreadPoolExecutor$TieBreakingPrioritizedRunnable.run(PrioritizedEsThreadPoolExecutor.java:153)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:745)
```
https://github.com/elasticsearch/elasticsearch/blob/v1.4.2/src/main/java/org/elasticsearch/index/mapper/internal/TimestampFieldMapper.java#L286
Looks like the existence of default timestamp is not checked before use. The next line also has the same issue -- uses of default timestamp without checked to see if it's not null.
To reproduce:
```
$ curl -XPUT localhost:9200/twitter2
$ curl -XPUT localhost:9200/twitter2/tweet/_mapping -d '{
"tweet" : {
"_timestamp" : {
"enabled" : true,
"default" : null
}
}
}'
$ curl -XPUT localhost:9200/twitter2/tweet/_mapping -d '{
"tweet" : {
"_timestamp" : {
"enabled" : true,
"default" : null
},
"properties": {
"user": {"type": "string"}
}
}
}'
```
Closes#9204.
(cherry picked from commit 62c6d63)
Before Elasticsearch 1.0, the type was allowed to be passed as the root
element when uploading a document. However, this was ambiguous if the
mappings also contained a field with the same name as the type. The
behavior was changed in 1.0 to not allow this, but a setting was added
for backwards compatibility. This change removes the setting for 2.0.
The header indicates to how many shard copies (primary and replicas shards) a write was supposed to go to, to how many
shard copies to write succeeded and potentially captures shard failures if writing into a replica shard fails.
For async writes it also includes the number of shards a write is still pending.
Closes#7994
This commit removes most of the Engine abstractions and removes
Engine exposure via dependency injection. It also removes the Holder
abstraction and makes the engine itself start at constrcution time.
It removes the start method from the engine entire which means no engine
instances exists until it's started. There is also no way to stop the
engine to restart, it needs to be an entire new Engine
This fix ensures that calls to the GET alias/mappings/settings/warmers APIs return the aliases/mappings/settings/warmers object even if there is no content within them.. This make them consistent with the GET Index API docs and the breaking changes in 1.4 docs
Closes#9148
This currently only adds checks to BaseFuture, but this should already cover
lots of client code. We could add more in the future, like interactions with
the filesystem and so on.
Close#9164
The multi percolate shard responses are collected in an atomic array which uses the shard id is used as index, but the number of shards the multi percolate request was meant to go to was used as size of this array instead the total number of shards an index has. This caused the exception when routing was used.
Closes#6214
Once we delete the the index on a node we are closing all resources
and subsequently need to delete all shards contents from disk. Yet
this happens today under a lock (the shard lock) that needs to be
acquried in order to execute any operation on the shards data
path. We try to delete all the index meta-data once we acquired
all the shard lock but this operation can run into a timeout which causes
the index to remain on disk. Further, all shard data will be left on
disk if the timeout is reached.
This commit removes all the shards data just before the shard lock
is release as the last operation on a shard that belongs to a deleted
index.
additional element per segment.
This commit adds a verbose flag to the _segments api. Currently the
only additional information returned when set to true is the full
ram tree from lucene for each segment.
This is our only cache which is not 'exact' and might allow for stalled results.
Additionally, a similar cache that we have and needs to perform lookups in other
indices in order to run queries is the script index, and for this index we rely
on the filesystem cache, so we should probably do the same with terms filters
lookups.
Close#9056
The `cluster.routing.allocation.balance.primary` setting has caused
a lot of confusion in the past while it has very little benefit form a
shard allocatioon point of view. Users tend to modify this value to
evently distribute primaries across the nodes which is dangerous since
a prmiary flag on it's own can trigger relocations. The primary flag for a shard
is should not have any impact on cluster performance unless the high level feature
suffereing from primary hotspots is buggy. Yet, this setting was intended to be a
tie-breaker which is not necessary anymore since the algorithm is deterministic.
This commit removes this setting entriely.
Today this not populated (null) in these cases. But it would be useful to have
this available, even just for improved error messages.
The trickiest part today is the handling of 1.2.x files written with
lucene 4.8+ which have both ES checksums and lucene ones. This is now simplified:
when the lucene checksum is there, we always use it.
Closes#9152
In some situations the shard balanceing weight delta becomes negative. Yet,
a negative delta is always treated as `well balanced` which is wrong. I wasn't
able to reproduce the issue in any way other than useing the real world data
from issue #9023. This commit adds a fix for absolute deltas as well as a base
test class that allows to build tests or simulations from the cat API output.
Closes#9023
If someone blocks on it and it is interrupted, we throw an ElasticsearchIllegalStateException. We should not set Thread.currentThread().interrupt(); in this case because we already communicate the interrupt through an exception.
Similar to #9001Closes#9141
A couple of changes that triggerred a refactoring in Elasticsearch:
- LUCENE-6148: Accountable.getChildResources returns a collection instead of
a list.
- LUCENE-6121: CachingTokenFilter now propagates reset(), as a result
SimpleQueryParser.newPossiblyAnalyzedQuery has been fixed to not reset both
the underlying stream and the wrapper (otherwise lucene would barf because of
a doubl reset).
- LUCENE-6119: The auto-throttle issue changed a couple of method
names/parameters. It also made
`UpdateSettingsTests.testUpdateMergeMaxThreadCount` dead slow so I muted this
test until we clea up merge throttling to use LUCENE-6119.
Close#9145
In the case you try to merge two settings, one being an array and one being
a field, together, the settings were merged instead of being overridden.
First config
my.value: 1
Second config
my.value: [ 2, 3 ]
If you execute
settingsBuilder().put(settings1).put(settings2).build()
now only values 2,3 will be in the final settings
Closes#8381
Plugin installation failed when bin/, conf/ and plugins/ directories are on different file systems. The method File.move() can't be used to move a non-empty directory between different filesystems.
I didn't find a simple way to unittest that, even with in-memory filesystems like jimfs or the Lucene test framework.
Closes#8999
If bulk index request fails due to a disconnect, unavailable shard etc, the request is
retried once before actually failing. However, even in case of failure the documents
might already be indexed. For autogenerated ids the request must not add the
documents again and therfore canHaveDuplicates must be set to true.
closes#8788
A recent situation occured where a MultiPolygon coordinate array was accidentally defined as a single polygon with multiple holes. Since the intent was a MultiPolygon, the holes of the unintended Polygon fell outside the outer shell. This exposed a bug in the contains logic inside BasePolygonBuilder. An ArrayIndexOutOfBoundsException was being thrown instead of a more useful ElasticsearchParseException( "hole is not within polygon" ). This pull request fixes the bug and adds additional unit tests for verifying proper MultiPolygon type parsing.
closes#9071
if we reopen an index and the majority of the replicas where
not created the reopen will fail sicne on master this runs with
local gatway all the time.
This can lead to sporadic shard creating timeouts if the same shard is
created, closed and created again on the same node. The reason for this is
that we holding on to the store reference while blocking on the mapping update
that will prevent the shard lock from being released. Holding the lock is unnecessary
in this case and can simply be removed.
ShapeBuilder expected coordinates for Envelope types in strict Top-Left, Bottom-Right order. Given that GeoJSON does not enforce coordinate order (as seen in #8672) clients could specify envelope bounds in any order and be compliant with the GeoJSON spec but not the ES ShapeBuilder logic. This change loosens the ShapeBuilder requirements on envelope coordinate order, reordering where necessary.
closes#2544closes#9067closes#9079closes#9080
Cleans up the testReusePeerRecovery test as well
The actual fix is in TransportNodesListShardStoreMetaData.java, which
needs to use `nodeEnv.shardDataPaths` instead of `nodeEnv.shardPaths`.
Due to the difficulty in tracking this down, I've added a lot of
additional logging. This also fixes a logging issue in GatewayAllocator
This allows specifying the path an index will be at.
`index.data_path` is specified in the settings when creating an index,
and can not be dynamically changed.
An example request would look like:
POST /myindex
{
"settings": {
"number_of_shards": 2,
"data_path": "/tmp/myindex"
}
}
And would put data in /tmp/myindex/0/index/0 and /tmp/myindex/0/index/1
Since this can be used to write data to arbitrary locations on disk, it
requires enabling the `node.enable_custom_paths` setting in
elasticsearch.yml on all nodes.
Relates to #8976
The terms filter lookup mechanism today caches filters. Because of this, the
cache values depend on two things: the values that can be found in the lookup
index AND the mapping of the local index, since changing the mapping can change
the way that the filter is parsed. We should make the cache depend solely on
the content of the lookup index.
For instance the issue I was seeing was due to the following scenario:
- create index1 with _id indexed
- run terms filter with lookup, the parsed filter looks like `_id: 1 OR _id: 2`
- remove index1
- create index1 with _id not indexed
- run terms filter without lookup, the parsed filter is `_uid: type#1 OR _uid: type#2` (the _id field mapper knows how to use the _uid field when _id is not indexed)
- run terms filter with lookup, the filter is fetched from the cache: `_id: 1 OR _id: 2` but does not match anything since `_id` is not indexed.
Close#9027
Running a terms filter on a single term is equivalent to loading a postings
list into a bit set and then returning the bit set instead of reading the
postings list on the fly.
Close#9014
PR #8978 included 4 unnecessary enumeration values ('cw', 'clockwise', 'ccw', 'counterclockwise'). Since the ShapeBuilder.parse method handles these as strings and maps them to LEFT and RIGHT enumerators, respectively, their enumeration counterpart is unnecessary. This minor change adds 4 static convenience variables (COUNTER_CLOCKWISE, CLOCKWISE, CCW, CW) for purposes of the API and removes the unnecessary values from the Orientation Enum.
closes#9035
This feature adds an optional orientation parameter to the GeoJSON document and geo_shape mapping enabling users to explicitly define how they want Elasticsearch to interpret vertex ordering. The default uses the right-hand rule (counterclockwise for outer ring, clockwise for inner ring) complying with OGC Simple Feature Access standards. The parameter can be explicitly specified for an entire index using the geo_shape mapping by adding "orientation":{"left"|"right"|"cw"|"ccw"|"clockwise"|"counterclockwise"} and/or overridden on each insert by adding the same parameter to the GeoJSON document.
closes#8764
When I originally wrote the transform feature I didn't think that the
XContentType of the reencoded source mattered. It actually matters because
payloads for the completion suggester are stored and returned exactly
as encoded by this XContentType.
This revision changes the transform feature from always reencoding with smile
to always reencoding with the provided XContentType to support the completion
suggester.
Closes#8959
This commit adds the support for the Ctrl-Close event on Windows using native system calls. This way, it is possible to catch the Ctrl-Close event sent by a 'taskill /pid' command (or when the user closes the console window where elasticsearch.bat was started) and gracefully close the node. Before this commit, the node was simply killed on taskkill/window closing.
There was a race condition in the test in the case where the nodes fault detection would manage to send and initial ping, followed by 2 attempts before the target service was disconnected.
RecoveryTarget initiates the recovery by sending a start recovery request to the source node and then waits for the recovery to complete. During recovery cancellation, we interrupt the thread so it will wake up and clean the recovery. Depending on timing, this can leave an unneeded interrupted thread status causing future IO commands to fail unneeded.
RecoverySource already had a handy utility called CancellableThreads. This extracts it to a top level class, and uses it in RecoveryTarget as well.
Closes#9000
Up to now, all filters could be cached using the `_cache` flag that could be
set to `true` or `false` and the default was set depending on the type of the
`filter`. For instance, `script` filters are not cached by default while
`terms` are. For some filters, the default is more complicated and eg. date
range filters are cached unless they use `now` in a non-rounded fashion.
This commit adds a 3rd option called `auto`, which becomes the default for
all filters. So for all filters a cache wrapper will be returned, and the
decision will be made at caching time, per-segment. Here is the default logic:
- if there is already a cache entry for this filter in the current segment,
then return the cache entry.
- else if the doc id set cannot iterate (eg. script filter) then do not cache.
- else if the doc id set is already cacheable and it has been used twice or
more in the last 1000 filters then cache it.
- else if the filter is costly (eg. multi-term) and has been used twice or more
in the last 1000 filters then cache it.
- else if the doc id set is not cacheable and it has been used 5 times or more
in the last 1000 filters, then load it into a cacheable set and cache it.
- else return the uncached set.
So for instance geo-distance filters and script filters are going to use this
new default and are not going to be cached because of their iterators.
Similarly, date range filters are going to use this default all the time, but
it is very unlikely that those that use `now` in a not rounded fashion will get
reused so in practice they won't be cached.
`terms`, `range`, ... filters produce cacheable doc id sets with good iterators
so they will be cached as soon as they have been used twice.
Filters that don't produce cacheable doc id sets such as the `term` filter will
need to be used 5 times before being cached. This ensures that we don't spend
CPU iterating over all documents matching such filters unless we have good
evidence of reuse.
One last interesting point about this change is that it also applies to compound
filters. So if you keep on repeating the same `bool` filter with the same
underlying clauses, it will be cached on its own while up to now it used to
never be cached by default.
`_cache: true` has been changed to only cache on large segments, in order to not
pollute the cache since small segments should not be the bottleneck anyway.
However `_cache: false` still has the same semantics.
Close#8449
We use PlainTransportFuture as a future for our transport calls. If someone blocks on it and it is interrupted, we throw an ElasticsearchIllegalStateException. We should not set Thread.currentThread().interrupt(); in this case because we already communicate the interrupt through an exception.
Closes#9001
Add a new ignore_idle_threads boolean option (default true) to
/_nodes/hot_threads, to filter out threads in known idle places like
waiting on a socket select or on pulling the next task from an empty
queue.
Closes#8985Closes#8908
This commit makes histogram reduction a bit cleaner by expecting buckets
returned from shards to be sorted by key and merging them on-the-fly on the
coordinating node using a priority queue.
Close#8797
This commit adds support for version and version_type to the Term Vectors API.
This could be useful in the following case whereby the user gets a document
and later wants to generate its TVs. With version, this would ensure that only
the TVs of that particular document are generated, and error out if the
document has been updated in between.
Closes#7480
This commit makes histogram reduction a bit cleaner by expecting buckets
returned from shards to be sorted by key and merging them on-the-fly on the
coordinating node using a priority queue.
Close#8797
When a node tries to join a master, the master may not yet be ready to accept the join request. In such cases we retry sending the join request up to 3 times before going back to ping. To detect this the current logic uses ExceptionsHelper.unwrapCause(t) to unwrap the incoming RemoteTransportException and inspect it's source, looking for ElasticsearchIllegalStateException. However, local ElasticsearchIllegalStateException can also be thrown when the join process should be cancelled (i.e., node shut down). In this case we shouldn't retry.
This commit adds an explicit NotMasterException to indicate the remote node is not a master. A similarly named exception (but meaning something else) in the master fault detection code was given a better name. Also clean up some other exceptions while at it.
Closes#8972
This allows specifying the path an index will be at.
`index.data_path` is specified in the settings when creating an index,
and can not be dynamically changed.
An example request would look like:
POST /myindex
{
"settings": {
"number_of_shards": 2,
"data_path": "/tmp/myindex"
}
}
And would put data in /tmp/myindex/0/index/0 and /tmp/myindex/0/index/1
Since this can be used to write data to arbitrary locations on disk, it
requires enabling the `node.enable_custom_paths` setting in
elasticsearch.yml on all nodes.
This commit adds the logic necessary for supporting polygon vertex ordering per OGC standards. Exterior rings will be treated in ccw (right-handed rule) and interior rings will be treated in cw (left-handed rule). This feature change supports polygons that cross the dateline, and those that span the globe/map. The unit tests have been updated and corrected to test various situations. Greater test coverage will be provided in future commits.
Addresses #8672
This feature branch implements OGC compliance for Polygon/Multi-polygon. That is, vertex order for the exterior ring follows the right-hand rule (ccw) and all holes follow the left-hand rule (cw). While GeoJSON imposes no restrictions, a user that wants to specify a complex poly across the dateline must do so in compliance with the OGC spec, otherwise a polygon that spans the globe will be assumed.
Reference issue #8672
Fix orientation of outer and inner ring for polygon with holes. Updated unit tests. Bug exists in boundary condition on negative side of dateline.
This provides a fix to issue #7644. A new Stats object must be created, and
not a reference to the retrieved stats, before we can add stats to it.
Otherwise, we would keep on adding to the same object on subsequent calls to
IndicesStatsResponse#getPrimaries() or IndicesStatsResponse#getTotal().
Closes#7644 and #8950
The "compressed" format was removed, so this caused warnings in the log
like:
```
[WARN ][index.fielddata ] [node_0] [test] failed to find format
[compressed] for field [test-num], will use default
```
Now that we do not automatically call .cleanUp() when clearing the field
data cache, we need to call it after the cache clear in
RandomExceptionCircuitBreakerTests
We do wait for shards to be closed in IndicesService for 30 second.
Yet, if somebody holds on to a store reference ie. an open scroll request
the 30 seconds time-out and node shutdown takes very long. We should
release all other resources first before we shutdown IndicesService.
Closes#8940
The setting `mapping.date.round_ceil` (and the undocumented setting
`index.mapping.date.parse_upper_inclusive`) affect how date ranges using
`lte` are parsed. In #8556 the semantics of date rounding were
solidified, eliminating the need to have different parsing functions
whether the date is inclusive or exclusive.
This change removes these legacy settings and improves the tests
for the date math parser (now at 100% coverage!). It also removes the
unnecessary function `DateMathParser.parseTimeZone` for which
the existing `DateTimeZone.forID` handles all use cases.
Any user previously using these settings can refer to the changed
semantics and change their query accordingly. This is a breaking change
because even dates without datemath previously used the different
parsing functions depending on context.
closes#8598closes#8889
In cases of heavy contention, it's possible for more than 2 threads
to race to a circuit breaking exception.
Essentially this means that if we have 3 threads all trying to add 3 and
simultaneously cause a circuit breaking exception (due to retry), when
adjusting after circuit breaking we can "rewind" past what this test
expects the child breaker to be at.
This adds leeway into the check, where it's okay to be within
NUM_THREADS from the parentLimit, because each thread should only add 1
to the breaker at a time.
We only have a single gatweway since es 1.3. There is no need to keep all
these abstractsion and nested packages. We can fold most of it into simpler
structures.
IndexEngine was an abstraction where we had index-level engines (instead
of shard-level) that could store meta information about the index. It
was never actually used by Elasticsearch, and only there for plugins.
This removes it, because it is a confusing abstraction and not needed,
no plugins should be implementing their own IndexEngines.
When a node fails (or closes), the master processes the network disconnect event and removes the node from the cluster state. If multiple nodes fail (or shut down) in rapid succession, we process the events and remove the nodes one by one. During this process, the intermediate cluster states may cause the node fault detection to signal the failure of nodes that are not yet removed from the cluster state. While this is fine, it currently causes unneeded reroutes and cluster state publishing, which can be cumbersome in big clusters.
Closes#8804Closes#8933
When we close a node all pending / active search requests need to be
cleared otherwise a node will wait up to 30 sec for shutdown sicne there
could be open scroll requests. This behavior was introduces in 1.5 such that
versions <= 1.4.x are not affected.
Closes#8940
Occasionally a the join thread successfully connected to a just closed node and which causes the subsequent join request to time out. It's default timeout 60s throws the test off when it waits for a cluster to form.
Calling cache.cleanUp() is kind of like calling System.gc(), meaning
that we should never have (non-test) things that rely on this
functionality.
For the field data and filter cache, we already have a periodic process
that runs this .cleanUp(), so there is no need to block index
closing/clearing on it. Instead, we can clean the field data cache in
InternalTestCluster before we check the circuit breaker.
This can help tests that time out because cleaning the cache is taking
too long
Today, shard stats are blocked while phase 3 of recovery (replay xlog)
is running; this change removes the engine readLock from shard stats
so it's not blocked.
Closes#8910
We have lots of boilerplate code that is unnecessarily abstracting
services ie InternalIndexShard and IndexShard or InternalIndexService and
IndexService. It's enough to have concrete classes for these core classes.
Closes#8904
If you use the java-package tool to create java packages, those
paths also should be added to the debian init script.
Also updated the docs, that it is ok to install java8.
Closes#7383
This change adds a 'http.publish_port' setting to the HTTP module to configure
the port which HTTP clients should use when communicating with the node. This
is useful when running on a bridged network interface or when running behind
a proxy or firewall.
Closes#8807Closes#8137
we currently have operationThreaded set to false when indexing a script. This setting
means that if we are executing the operation locally that we don't spawn a new thread for
it althought incoming thread in this case is the network thread. Now sicne we are indexing here
the engine is currently recovering which sometimes locks the engine for finalization blocks on
a network call waiting for the recovery target to come back the internal lock in engine will never be
released since we are waiting with our network thread for it to be released.
Upgrades lucene to latest, and supports the BEST_COMPRESSION parameter
now supported (with backwards compatibility, etc) in Lucene.
This option uses deflate, tuned for highly compressible data.
index.codec::
The default value compresses stored data with LZ4 compression, but
this can be set to best_compression for a higher compression ratio,
at the expense of slower stored fields performance.
IMO its safest to implement as a named codec here, because ES already
has logic to handle this correctly, and because its unrealistic to have
a plethora of options to Lucene's default codec... we are practically
limited in Lucene to what we can support with back compat, so I don't
think we should overengineer this and add additional unnecessary plumbing.
See also:
https://issues.apache.org/jira/browse/LUCENE-5914https://issues.apache.org/jira/browse/LUCENE-6089https://issues.apache.org/jira/browse/LUCENE-6090https://issues.apache.org/jira/browse/LUCENE-6100Closes#8863
This fix adds better error handling for parsing multipoint, linestring, and polygon GeoJSONs. Current logic throws a NPE when parsing a multipoint, linestring, or polygon that does not comply with the GeoJSON specification. That is, if a user provides a single coordinate instead of an array of coordinates, or array of linestrings, the ShapeParser throws a NPE wrapped in a SearchParseException instead of a more useful error message.
Closes#8432
After the refactoring in #8784 some settings didn't get passed to the
actual engine and there exists a race if the settings are updated while
the engine is started such that the actual starting engine doesn't see
the latest settings. This commit fixes the concurrency issue as well as
adds tests to ensure the settings are reflected.
After upgrading shard might start relocating again. If there are no
replicas the cluster state of a node might not be up to data for
a few miliseconds and direct a search request to a node that does not
have the shard anymore. This result in the following test failures:
1> java.lang.AssertionError: Count is 99 but 101 was expected. Total shards: 13 Successful shards: 12 & 0 shard failures:
1> __randomizedtesting.SeedInfo.seed([1932F73B458703CA:6F4FAD3DAC55591C]:0)
1> [...org.junit.*]
1> org.elasticsearch.test.hamcrest.ElasticsearchAssertions.assertHitCount(ElasticsearchAssertions.java:184)
1> org.elasticsearch.bwcompat.BasicBackwardsCompatibilityTest.testIndexRollingUpgrade(BasicBackwardsCompatibilityTest.java:358)
Waiting for relocation finished should fix this.
Today we have several upgrade methods that can read state written
pre 0.90 or even pre 0.19. Version 2.0 should not support these state
formats. Users of these version should upgrade to a 1.x or 0.90.x version
first.
Closes#8850
This commit removes the deprecated constant for the main
version and uses the real lucene version we are running instead.
Behind the scenes the same value was used and is now obsolet.
Modifications to LoggingListener pushed with #8820 caused the original logger levels not to be reset after modifications, as the new state was saved for restore instead of the previous one.
Added unit tests for LoggingListener as well.
Closes#8845
Restrict use of java.io.File to 5 methods (excluded), but otherwise ban.
This is a prerequisite to do any mocking here.
I don't try to do any heavy cleanup on these tests, I am not familiar with them.
So this is mostly a rote straightforward conversion.
Closes#8836
This is a start to exposing memory stats improvements from Lucene 5.0.
This adds the following categories of Lucene index pieces to index stats:
* Terms
* Stored fields
* Term Vectors
* Norms
* Doc values
This commit add the engines reference to the store out of the actual
implementation into the hodler since the holder manages the actual lifcycle.
Engine internal references like per searcher or per recovery are kept inside
the actual implemenation since the have a different lifecycle.
Once the current engine is started you can only close it once. Once closed the engine cannot be started again. This commit adds a stop method which signals the engine to free it's resources but in a way that allows restarting.
This is done by introducing InternalEngineHolder which is a wrapper around InternalEngine. This allows to add the stop() method without adding complexity the engine implementation. InternalEngineHolder also serves an entry point for listeners (incoming and outgoing) to other ES components, which removes the needs add/remove them if the engine is stopped.
Closes#8784
Today we try to fetch a shard Id for a given IndexReader / LeafReader
by walking it's tree until the lucene internal SegmentReader and then
casting the directory into a StoreDirecotory. This class is fully internal
to Elasticsearch and should not be exposed outside of the Store.
This commit makes StoreDirectory a private inner class and adds dedicated
ElasticsearchDirectoryReader / ElasticserachLeafReader exposing a ShardId
getter to obtain information about the shard the index / segment belogs to.
These classes can be used to expose other segment specific information in
the future more easily.
When a scoring script returns not a number, the current message is confusing (IllegalArgumentException[docID must be >= 0 and < maxDoc=3 (got docID=2147483647)]). This commit adds the error message ScriptException[script score function returns a wrong score: NaN].
Closes#2426
The packaged init scripts could return an error, if the file
/proc/sys/vm/max_map_count was not existing and we still called
sysctl.
This is primarly to prevent confusing error messages when elasticsearch
is started under virtualized environments without a proc file system.
Closes#4978
Instead of iterating all shards of all indices to get all relocating
shards for a given node we can just use the RoutingNode#shardsWithState
method and fetch all INITIALIZING / RELOCATING shards and check if they
are relocating. This operation is much faster and uses pre-build
data-structures.
Relates to #6372
This commit changes internal callback to be clear
about when they are called and also provide the
exception that was potentially thrown as a callback argument.
Closes#5945
Previously it was possible for the field data clearing in this test to
take too long, causing the test to time out.
This also switches to using `scaledRandomIntBetween` for the number of
fields.
This change will chown /usr/share/elasticsearch/plugins to the elasticsearch
user (the directory was formerly owned by root). This enables the ES user to
manage plugins.
Also, /usr/share/elasticsearch/plugins is now removed when the elasticsearch
package is un-installed. Previously it was left lying there.
Closes#8732
Signed-off-by: Thilo Fromm <github@thilo-fromm.de>
This change adds a systemd service configuration file, and adds systemd logic
to installation and de-installation scripts. The upcoming Debian 8 "Jessie"
release will use systemd.
fixes#8943
Signed-off-by: Thilo Fromm <github@thilo-fromm.de>
the recovery diff can return file in the `different` category
since it's conservative if it can't tell if the files are the same.
Yet, the cleanup code only needs to ensure both ends of the recovery
are consistent. If we have a very old segments_N file no checksum is present
but for the delete files they might be such that the segments file passes
the consistency check but the .del file doesn't sicne it's in-fact the same
but this check was missing in the last commit.
Original indices are optional in ShardDeleteByQueryRequest only for backwards compatibility, see #7406. We can remove this in master since 2.0 will require a full cluster restart.
Closes#8777
This commit factors out the PID file creation from bootstrap and adds
tests for error conditions etc. We also can't rely on DELETE_ON_CLOSE
since it might not even write the file depending on the OS and JVM implementation.
This impl uses a shutdown hook to best-effort remove the pid file if it was written.
Closes#8771
This commit adds a very lightweight action to the transport
serivce that allows to fetch clustername and the discovery node
from a node. This is used by transport clients to test liveness of
a node without using the nodesinfo API which can be blocking if management
threadpools are busy.
Closes#8763
The bwc layer added with #7105 is not needed in master as a full cluster restart will be required, thus from 2.0 on the only supported action names are compliant to the defined conventions and don't need to be converted to the old format
Closes#8758
Related to #8667:
Some QueryBuilders have been deprecated in 1.x branches. We removed them in 2.0.
Removed
-------
* `textPhrase(...)`
* `textPhrasePrefix(...)`
* `textPhrasePrefixQuery(...)`
* `filtered(...)`
* `inQuery(...)`
* `commonTerms(...)`
* `queryString(...)`
* `simpleQueryString(...)`
Closes#8721.
the extract location might be on a different filesystem where
atomic move won't work. Yet this operation is not critical in terms
of visibility so there is no need to do this.
REST tests are being shuffled before their execution. To guarantee their repeatability given the seed, their order needs to be always the same before the shuffling happens.
Closes#8745
The conversion to the Path API doesn't work if the path points
to a file inside a JAR like a config. These path must be read
while the ZIP filesystem is opened which can't be guaranteed across
the board. This commit reverts back the relevant changes to java.net.URL
and adds a util method to read UTF-8 Encoded files from URLs correctly.
This commit cuts over all of core (not quite all tests) to java.nio.Path
It also adds the file class to the core forbidden APIs to prevent its usage.
This commit also resolves#8254 since we now consistently useing the NIO Path
API. The Changes in this commit allow for more information if IO operations fail
since the NIO API throws exceptions instead of boolean return values. The build-in
methods used in this commit are also more resillient to encodeing errors like
unmappable characters and throw exceptions if those chars are present in a file.
Closes#8254Closes#8666
Today we don't check if the recovery target has all the
files that we expect there after the recovery. This commit
adds aditional safety to ensure all files are present with the
correct checksums on recovery finalization.
Closes#8723
Today we wait 500ms before we retry a recovery if the target node is not ready.
This happens if the source starts the recovery before the target has
processed the clusterstate moving the target shard into the right state.
This can cause a 500ms delay each time it happens while the shard is ready
way earlier on the target node. This commit makes this delay configurable
to mainly speed up test processing and shard allocation in tests.
Inner hits allows to embed nested inner objects, children documents or the parent document that contributed to the matching of the returned search hit as inner hits, which would otherwise be hidden.
Closes#8153Closes#3022Closes#3152
Some QueryBuilders are missing or have a different naming than the other ones.
This patch is applied to branch 1.x and master (elasticsearch 1.5 and 2.0):
Added
-----
* `templateQuery(...)`
* `commonTermsQuery(...)`
* `queryStringQuery(...)`
* `simpleQueryStringQuery(...)`
Deprecated
----------
* `commonTerms(...)`
* `queryString(...)`
* `simpleQueryString(...)`
Ryan Ernst
Michael McCandless
Adrien Grand
Robert Muir
Shay Banon
Lee Hinman
Luca Cavanna
David Pilato
Sebastian Utz
Martijn van Groningen
Boaz Leskes
Simon Willnauer
Colin Goodheart-Smithe
Alexander Reelsen
tlrx
Clinton Gormley
Britta Weber
Nicholas Knize
Tomas Varaneckas
Nik Everett
Guillaume Hiron
Alex Ksikes
Peter Fabian Mitchell
Jun Ohtani
Patrick Koenig
Igor Motov
Thilo Fromm
Louis Sautier
John Michael Luy