Today, the TransportReplicationAction checks the global level blocks and
the index level blocks before routing the operation to the primary, in the
ReroutePhase, and it happens at the very beginning of the transport
replication action execution. For the upcoming rework of the Close Index
API and in order to deal with primary relocation, we'll need to also check
for blocks before executing the operation on the primary (while holding a
permit) but before routing to the new primary.
This pull request change the AsyncPrimaryAction so that it checks for
replication action's blocks before executing the operation locally or before
routing the primary action to the newly primary shard. The check is done
while holding a PrimaryShardReference.
Related to #33888
The way ScoreAccessor implements `compareTo()` is problematic because it doesn't
completely follow the Comparable contract, specificaly symmetry (if x is a
ScoreAccessor and y any Number then x.comparTo(y) works, but y.compareTo(x)
generally does not even compile). Fortunately we don't seem to use the fact that
ScoreAccessor is a Comparable anywhere, so we can simply remove it.
Today the `PeerFinder` probes each address it obtains, identifies the node to
which it just connected, and then returns all such nodes. However, this can
lead to duplicates if a node manages to connect to another node via two
distinct addresses. This causes bootstrapping to fail since
`BootstrapConfiguration#resolve` forbids duplicates.
This change alters the behaviour of the `PeerFinder` to remove duplicates in
this situation.
If shutting down half or more of the master-eligible nodes, their votes must
first be explicitly withdrawn to ensure that the cluster doesn't lose its
quorum. This works via _voting tombstones_, stored in the cluster state, which
tell the reconfigurator to remove nodes from the voting configuration.
This change introduces voting tombstones to the cluster state, together with
transport APIs for adding and removing them, and makes use of these APIs in
`InternalTestCluster` to support tests which remove at least half of the
master-eligible nodes at once (e.g. shrinking from two master-eligible nodes to
one).
AbstractComponent was deprecated in #35140 and is looking like it will be
removed at some point by #34888. Today all it does is provide a logger. This
change removes the usages of AbstractComponent that live solely in the zen2
feature branch to avoid some future merge pain, and replaces it where necessary
with some directly-created loggers.
This change adds a special caching reader that caches all relevant
values for a range query to rewrite correctly in a can_match phase
without actually opening the underlying directory reader. This
allows frozen indices to be filtered with can_match and in-turn
searched with wildcards in a efficient way since it allows us to
exclude shards that won't match based on their date-ranges without
opening their directory readers.
Relates to #34352
Depends on #34357
The ParsedReverseNested implementation should implement the ReverseNested
interface and not the Nested interface. Although this is an empty marker
interface it is confusing and can lead to casting errors. Also adding a test to
check that both ParsedNested and ParsedReverseNested implement the correct
interface.
Closes#35449
Some very old ancient versions of Linux do not have /etc/os-release. For
example, old Red Hat-like OS. This commit adds a fallback for handling
pretty name for these OS.
This is a follow up to #35357. That commit failed to register the new
cluster.remote.cluster_name.transport.compress setting with
`ClusterSettings`. This commit fixes that.
Some OS (e.g., Oracle Linux Server 6.9) have a trailing space at the end
of the PRETTY_NAME line in /etc/os-release. This commit addresses this
by accounting for this trailing space when extracting the pretty name.
Implements serialization compatibility between Zen1 and Zen2 transport action, allowing a Zen1 node to join a fully formed Zen2 cluster and vice-versa.
The MockTcpTransport is not friendly in regards to memory usage. It must
allocate multiple byte arrays for every message. This improves the
memory situation by failing fast if the message is improperly formatted.
Additionally, it uses reusable big arrays for at least half of the
allocated byte arrays.
This change adds a logger for the query and fetch phases that prints all requests
before their execution at the trace level. This will help debugging cases where an issue
occurs during the execution since only completed queries are logged by the slow logs.
This is related to #34483. It introduces a namespaced setting for
compression that allows users to configure compression on a per remote
cluster basis. The transport.tcp.compress remains as a fallback
setting. If transport.tcp.compress is set to true, then all requests
and responses are compressed. If it is set to false, only requests to
clusters based on the cluster.remote.cluster_name.transport.compress
setting are compressed. However, after this change regardless of any
local settings, responses will be compressed if the request that is
received was compressed.
Today our OS information returned in node stats only returns a
high-level name of the OS (e.g., "Linux"). Yet, for some uses this is
too high-level and knowing at a finer level of granularity the
underlying OS can be useful. This commit extracts the pretty name on
Linux from /etc/os-release. This pretty name usually includes the Linux
vendor and the Linux vendor version number (e.g., Fedora 28).
Enables diff-based publishing, which is an optimization where only the changing parts of the cluster
state are published to the nodes in the cluster, falling back to full cluster state publishing if the
receiver does not have the previous cluster state.
- Introduces a transport API for bootstrapping a Zen2 cluster
- Introduces a transport API for requesting the set of nodes that a
master-eligible node has discovered and for waiting until this comprises the
expected number of nodes.
- Alters ESIntegTestCase to use these APIs when forming a cluster, rather than
injecting the initial configuration directly.
Currently we introduced a hard limit of 1024 to the number of fields a query can
be expanded to in #26541. Instead of using a hard limit, we should make this
configurable. This change removes the hard limit check and uses the existing
`max_clause_count` setting instead.
Closes#34778
Currently when aggregating on an unmapped date field (e.g. using a
date_histogram) we don't preserve the aggregations `format` setting but instead
use the default format. This can lead to loosing the aggregations `format` when
aggregating over several indices where some of them contain unmapped date fields
and are encountered first in the reduce phase.
Related to #31760
Today the `composite` aggregation throws an error if a source targets an
unmapped field and `missing_bucket` is set to false. Documents without a
value for a source cannot produce any bucket if `missing_bucket` is not
activated so the error is a shortcut to say that the response will be empty.
However this is not consistent with the `terms` aggregation which accepts
unmapped field by default even if the response is also guaranteed to be empty.
This commit removes this restriction, if a source contains an unmapped field
we now return an empty response (no buckets).
Closes#35317
The current implementation of asyncBlockOperations() can be used to
execute some code once all indexing operations permits have been acquired,
then releases all permits immediately after the code execution. This
immediate release is not suitable for treatments that need to keep all
permits over multiple execution steps.
This commit adds a new asyncBlockOperations() that exposes a Releasable,
making it possible to acquire all permits and only release them all
when needed by closing the Releasable. The existing blockOperations()
method has been modified to delegate permit acquisition/releasing to this new
method.
Relates to #33888
This commit uses the index settings version so that a follower can
replicate index settings changes as needed from the leader.
Co-authored-by: Martijn van Groningen <martijn.v.groningen@gmail.com>
The lookup vars under params (namely _fields and _source) were
inadvertently removed when scoring scripts were converted to using
script contexts. This commit adds them back, along with deprecation
warnings for those that should not be used.
A CCR test failure shows that the approach in #34474 is flawed.
Restoring the LocalCheckpointTracker from an index commit can cause both
FollowingEngine and InternalEngine to incorrectly ignore some deletes.
Here is a small scenario illustrating the problem:
1. Delete doc with seq=1 => engine will add a delete tombstone to Lucene
2. Flush a commit consisting of only the delete tombstone
3. Index doc with seq=0 => engine will add that doc to Lucene but soft-deleted
4. Restart an engine with the commit (step 2); the engine will fill its
LocalCheckpointTracker with the delete tombstone in the commit
5. Replay the local translog in reverse order: index#0 then delete#1
6. When process index#0, an engine will add it into Lucene as a live doc
and advance the local checkpoint to 1 (seq#1 was restored from the
commit - step 4).
7. When process delete#1, an engine will skip it because seq_no=1 is
less than or equal to the local checkpoint.
We should have zero document after recovering from translog, but here we
have one.
Since all operations after the local checkpoint of the safe commit are
retained, we should find them if the look-up considers also soft-deleted
documents. This PR fills the disparity between the version map and the
local checkpoint tracker by taking soft-deleted documents into account
while resolving strategy for engine operations.
Relates #34474
Relates #33656
Today when a percolator query contains a date range then the query
analyzer extracts that range, so that at search time the `percolate` query
can exclude percolator queries efficiently that are never going to match.
The problem is that if 'now' is used it is evaluated at index time.
So the idea is to rewrite date ranges with 'now' to a match all query,
so that the query analyzer can't extract it and the `percolate` query
is then able to evaluate 'now' at query time.
This change adds a `frozen` engine that allows lazily open a directory reader
on a read-only shard. The engine wraps general purpose searchers in a LazyDirectoryReader
that also allows to release and reset the underlying index readers after any and before
secondary search phases.
Relates to #34352
Today we only apply `ingore_throttled` to expansions from wildcards,
date math expressions and aliases. Yet, this is tricky since we might
have resolved certain expressions in pre-filter steps like security.
It's more consistent to apply this logic to all expressions including
concrete indices.
Relates to #34354
With this change, `Version` no longer carries information about the qualifier,
we still need a way to show the "display version" that does have both
qualifier and snapshot. This is now stored by the build and red from `META-INF`.
This is related to #29023. Additionally at other points we have
discussed a preference for removing the need to unnecessarily block
threads for opening new node connections. This commit lays the groudwork
for this by opening connections asynchronously at the transport level.
We still block, however, this work will make it possible to eventually
remove all blocking on new connections out of the TransportService
and Transport.
We've decided that the bulk, delete, get, index, update, and search APIs should not
contain this request parameter, and we will instead accept both typed and typeless calls.
Today we allow the user to set the minimum size of a voting configuration. On
reflection we would rather this was simply '3' where possible, and we can use
the retirement API to control the removal of nodes more explicitly.
This change replaces the old reconfigurator setting with a new one,
`cluster.auto_shrink_voting_configuration`, which determines whether
Elasticsearch should automatically remove nodes from the voting configuration
or not.
We have seen an improvement when we bumped the timeout from 1s to 5s, but there are still a few failures for this tests. With this commit we bump the timeout to 10 seconds hoping it will stop all the failures.
Today if a wildcard, date-math expression or alias expands/resolves
to an index that is search-throttled we still search it. This is likely
not the desired behavior since it can unexpectedly slow down searches
significantly.
This change adds a new indices option that allows `search`, `count`
and `msearch` to ignore throttled indices by default. Users can
force expansion to throttled indices by using `ignore_throttled=true`
on the rest request to expand also to throttled indices.
Relates to #34352
This test has a bug that got introduced during the refactoring of #32442. With 2 concurrent term increments,
we can only assert under the operation permit that we are in the correct operation term, not that there is
not already another term bump pending.
Closes#34862
If the random query string is "now" by accident _and_ we are also not setting
some field names to use explicitely, then we can hit the "mapped_date" field
from default test setup. This correctly leads to the query being was marked as
not cacheable, but we assume and check so later. This change fixes this rare
edge case by making sure we don't hit the "date" field in this rare cases.
Closes#35183
When the engine is asked for historical operations, we check if some of the requested operations
are not yet refreshed and if so we refresh before returning the operations. The refresh check is
based on capturing the local checkpoint before each refresh and comparing that value to the one
requested when `newChangesSnapshot` was called. If the requested range is above the captured
local checkpoint we issue a refresh.
This can currently cause unneeded extra refreshes if the method is called concurrently which may cause unwanted degradation in indexing performance. This is especially relevant for CCR where we always ask for a range below the global checkpoint. That range is guaranteed to be below the local
checkpoint of the shard and one refresh is enough to serve multiple changes requests.
This commit fixes this by introducing a dedicated mutex to make sure the test for whether a refresh
is needed actually wait for concurrents for concurrent refreshes that were caused by another
change refresh.
Note that this is not a big change in semantics as refreshes are serialized by lucene anyway. I also
opted not to keep the synchronization to the changes snapshot request only even if in theory we
can apply it to all refreshes, not matter where they come from.
This changes the current script.max_size_in_bytes to be dynamic so it can be
set through the cluster settings API. This setting is also applied to inline scripts
in the compile method of ScriptService to prevent excessively long inline
scripts from being compiled. The script length limit is removed from Painless as
this is no longer necessary with the protection in compile.
Currently we create a new netty event loop group for client connections
and all server profiles. Each new group creates new threads for io
processing. This means 2 * num of processors new threads for each group.
A single group should be able to handle all io processing (for the
transports). This also brings the netty module inline with what we do
for nio.
Additionally, this PR renames the worker threads to be the same for
netty and nio.
Currently, we assume that rollback always happens in the test
testRestoreLocalHistoryFromTranslogOnPromotion. However, if the global
checkpoint equals max_seq_no, we won't rollback. This causes the
max_seq_no_of_updates assertion failed because max_seq_no_of_updates
won't be advanced to the global checkpoint. With this commit, we assert
max_seq_no_of_updates in two different paths.
Today we always allocate a full buffer (1024 elements) in a
LuceneChangesSnapshot even though the requesting size is smaller.
With this change, we will use the requesting size as the buffer size if
it's smaller than the default batch size; otherwise uses the default
batch size.
With this commit we differentiate between permanent circuit breaking
exceptions (which require intervention from an operator and should not
be automatically retried) and transient ones (which may heal themselves
eventually and should be retried). Furthermore, the parent circuit
breaker will categorize a circuit breaking exception as either transient
or permanent based on the categorization of memory usage of its child
circuit breakers.
Closes#31986
Relates #34460
Stop passing `Settings` to `AbstractComponent`'s ctor. This allows us to
stop passing around `Settings` in a *ton* of places. While this change
touches many files, it touches them all in fairly small, mechanical
ways, doing a few things per file:
1. Drop the `super(settings);` line on everything that extends
`AbstractComponent`.
2. Drop the `settings` argument to the ctor if it is no longer used.
3. If the file doesn't use `logger` then drop `extends
AbstractComponent` from it.
4. Clean up all compilation failure caused by the `settings` removal
and drop any now unused `settings` isntances and method arguments.
I've intentionally *not* removed the `settings` argument from a few
files:
1. TransportAction
2. AbstractLifecycleComponent
3. BaseRestHandler
These files don't *need* `settings` either, but this change is large
enough as is.
Relates to #34488
* NETWORKING: MockTransportService Wait for Close
* Make `MockTransportService` wait `30s` for close listeners to run before failing the assertion
* Closes#34990
When we connect to remote clusters, there may be a few more routers/firewalls in-between compared to when we connect to nodes in the same cluster. We've experienced cases where firewalls drop connections completely and keep-alives seem not to be enough, or they are not properly configured. With this commit we allow to enable application-level pings specifically from CCS nodes to the selected remote nodes through the new setting `cluster.remote.${clusterAlias}.transport.ping_schedule`. The new setting is similar `transport.ping_schedule` but it does not affect intra-cluster communication, pings are only sent to specific remote cluster when specifically enabled, as they are disabled by default.
Relates to #34405
* DISCOVERY: Cleanup AbstractDisruptionTestCase
* Make the internal test cluster manage minimum master nodes where we used the default of (nodes / 2 + 1) before
* Remove use of the `NodeConfigurationSource` indirection
* Relates #33675
Drops the `Settings` member from `AbstractComponent`, moving it from the
base class on to the classes that use it. For the most part this is a
mechanical change that doesn't drop `Settings` accesses. The one
exception to this is naming threads where it switches from an invocation
that passes `Settings` and extracts the node name to one that explicitly
passes the node name.
This change doesn't drop the `Settings` argument from
`AbstractComponent`'s ctor because this change is big enough as is.
We'll do that in a follow up change.
This commit filters out usage of deprecated tzs by tests. These are
tested separately and should not require checking for warnings on any
test using random timezones.
closes#34188
This commit adds a new single value metric aggregation that calculates
the statistic called median absolute deviation, which is a measure of
variability that works on more types of data than standard deviation
Our calculation of MAD is approximated using t-digests. In the collect
phase, we collect each value visited into a t-digest. In the reduce
phase, we merge all value t-digests, then create a t-digest of
deviations using the first t-digest's median and centroids
Conflicts during the merge:
1. >=140 chars line length fixed for a lot of project files and warnings
for those files are no longer suppressed
2. Node name is removed from AbstractComponent, it’s no longer taken
from settings, but is explicitly passed as constructor argument and
there were quite a few new classes on zen2 branch that require this
change
3. TransportResponseHandler interface changed (new method added) and
Zen2 makes a lot of subclasses in tests
4. Deprecated way of obtaining logger was changed
Bulk Request in High level rest client should be consistent with what is
possible in Rest API, therefore should support global parameters. Global
parameters are passed in URL in Rest API.
Some parameters are mandatory - index, type - and would fail validation
if not provided before before the bulk is executed.
Optional parameters - routing, pipeline.
The usage of these should be consistent across sync/async execution,
bulk processor and BulkRequestBuilder
closes#26026
Deprecates `_source_include` and `_source_exclude` url parameters
in favor of `_source_inclues` and `_source_excludes` because those
are consistent with the rest of Elasticsearch's APIs.
Relates to #22792
Drops the `deprecationLogger` from `AbstractComponent`, moving it to
places where we need it. This saves us from building a bunch of
`DeprecationLogger`s that we don't need.
Relates to #34488
`AbstractComponent` is trouble because its name implies that
*everything* should extend from it. It *is* useful, but maybe too
broadly useful. The things it offers access too, the `Settings` instance
for the entire server and a logger are nice to have around, but not
really needed *everywhere*. The `Settings` instance especially adds a
fair bit of ceremony to testing without any value.
This removes the `nodeName` method from `AbstractComponent` so it is
more clear where we actually need the node name.
* Removed methods are just unused (the exceptions being isGeoPoint() and is
isFloatingPoint() but those could more efficiently be replaced by enum comparisons to simplify the code)
* Remove exceptions aren't thrown
In order to remove Streamable from the codebase, Response objects need
to be read using the Writeable.Reader interface which this change
enables. This change enables the use of Writeable.Reader by adding the
`Action#getResponseReader` method. The default implementation simply
uses the existing `newResponse` method and the readFrom method. As
responses are migrated to the Writeable.Reader interface, Action
classes can be updated to throw an UnsupportedOperationException when
`newResponse` is called and override the `getResponseReader` method.
Relates #34389
This commit adds a new ParentJoinAggregator that implements a join using global ordinals
in a way that can be reused by the `children` and the upcoming `parent` aggregation.
This new aggregator is a refactor of the existing ParentToChildrenAggregator with two main changes:
* It uses a dense bit array instead of a long array when the aggregation does not have any parent.
* It uses a single aggregator per bucket if it is nested under another aggregation.
For the latter case we use a `MultiBucketAggregatorWrapper` in the factory in order to ensure that each
instance of the aggregator handles a single bucket. This is more inlined with the strategy we use for other
aggregations like `terms` aggregation for instance since the number of buckets to handle should be low (thanks to the breadth_first strategy).
This change is also required for #34210 which adds the `parent` aggregation in the parent-join module.
Relates #34508
* Fix linelength suppressions in index.fielddata
* Some lines that were too long were dead code => Removed them and all code that became dead because of it
* Relates #34884
testClusterFormsByScanningPorts is flaky: sometimes in CI it's not possible to
bind to any of the ports we need to in order for the port scanning to work.
This change removes this test, and #34809 describes a better way to test this
behaviour.
After discussing on the team's FixItFriday, we concluded that
static final instance variables that are mutable should be lowercased.
Historically, DeprecationLogger was uppercased more frequently than lowercased.
This is related to #30876. The AbstractSimpleTransportTestCase initiates
many tcp connections. There are normally over 1,000 connections in
TIME_WAIT at the end of the test. This is because every test opens at
least two different transports that connect to each other with 13
channel connection profiles. This commit modifies the default
connection profile used by this test to 6. One connection for each
type, except for REG which gets 2 connections.
* Check self references in metric agg after last doc collection (#33593)
* Revert 0aff5a30c5dbad9f476be14f34b81e2d1991bb0f (#33593)
* Check self refs in metric agg only once in post collection hook (#33593)
* Remove unnecessary mocking (#33593)
The check for null argument is already done in `splitStringByCommaToArray`, hence it can be removed, which allows us to remove the whole splitScrollIds private method.
With this change, we apply the common test config automatically to all
newly created tasks instead of opting in specifically.
For plugin authors using the plugin externally this means that the
configuration will be applied to their RandomizedTestingTasks as well.
The purpose of the task is to simplify setup and make it easier to
change projects that use the `test` task but actually run integration
tests to use a task called `integTest` for clarity, but also because
we may want to configure and run them differently.
E.x. using different levels of concurrency.
Currently, if MetaDataStateFormat.write throws an IOExceptions if there was some problem with persisting state to disk. If an exception is thrown, loadLatestState may read either old state or new state. This is not enough for the Zen2 algorithm. In case of failure, we need to distinguish between 2 cases: storage is left in clean state or storage is left in a dirty state.
If storage is left in the clean state, loadLatestState may read only old state. If storage is left in a dirty state, loadLatestState may read either old or new state.
If an exception occurs when writing the manifest file to disk this distinction is important for Zen2. If storage is clean, the node can continue to be a part of the cluster and may try to accept further cluster state updates (if it fails to accept cluster state updates it will be kicked off from the cluster using different mechanism). But if storage is dirty, the node should be restarted and it will be able to start up successfully only once it successfully re-writes manifest file to disk.
This commit changes MetaDataStateFormat.write signature, replacing IOException with WriteStateException, which “isDirty” method could be used to distinguish between 2 failure cases.
We need to minimise the number of failures, that leave storage in a dirty state. That’s why this PR changes the algorithm that is used to store state to disk. It has the following layout:
1. For the first state location, create and fsync tmp file with state content.
2. For each extra location, copy and fsync tmp file with state content.
2. Atomically rename tmp file in the first location.
3. For each extra location, atomically rename tmp file.
4. For each location, fsync state directory.
5. Perform cleanup of old files, ignoring exceptions.
If an exception occurs in steps 1-3, storage is clearly in the clean state. If an exception occurs in step 5, storage is clearly in dirty state. Exception in step 4 is questionable, there are 2 options:
1. Consider it as a failure. If the first disk fails, state disappears. So this is a failure and storage is in a dirty state.
2. Do not consider it as failure at all, ignore disk failures.
This commit prefers 1st approach and MetaDataTestFormatTests.testFailRandomlyAndReadAnyState tests for disk failures.
Access to special variables _source and _fields were accidentally
removed in recent refactorings. This commit adds them back, along with a
test.
closes#33884
In a future major version, we will be introducing a soft limit on the
number of shards in a cluster based on the number of nodes in the
cluster. This limit will be configurable, and checked on operations
which create or open shards and issue a warning if the operation would
take the cluster over the limit.
There is an option to enable strict enforcement of the limit, which
turns the warnings into errors. In a future release, the option will be
removed and strict enforcement will be the default (and only) behavior.
- Restrict visibility of Aggregators and Factories
- Move PipelineAggregatorBuilders up a level so it is consistent with
AggregatorBuilders
- Checkstyle line length fixes for a few classes
- Minor odds/ends (swapping to method references, formatting, etc)
This commit introduces two corrections to the way simulate?verbose
handles conditionals on processors.
1) Prior to this change when executing simulate?verbose for
processors with conditionals that evaluate to false, that processor
would still be displayed in the result set. What was displayed was
correct, such that no changes to the document occurred. However, if the
conditional evaluates to false, the processor should not even be
displayed.
2) Prior to this change when executing simulate?verbose for
pipeline processors with conditionals, the individual steps would no
longer be displayed. Commit e37e5df addressed the issue, but
failed account for a conditional on the pipeline processor. Since
a pipeline processor can introduce cycles and is effectively a
single processor that encapsulates multiple other processors that
are potentially guarded by a single conditional, special handling is
needed to for pipeline and conditional pipeline processors.
We should delete a job by directly talking to the allocated
task and telling it to shutdown. Today we shut down a job
via the persistent task framework. This is not ideal because,
while the job has been removed from the persistent task
CS, the allocated task continues to live until it gets the
shutdown message.
This means a user can delete a job, immediately delete
the rollup index, and then see new documents appear in
the just-deleted index. This happens because the indexer
in the allocated task is still running and indexes a few
more documents before getting the shutdown command.
In this PR, the transport action is changed to a TransportTasksAction,
and we invoke onCancelled() directly on the matching job.
The race condition still exists after this PR (albeit less likely),
but this was a precursor to fixing the issue and a self-contained
chunk of code. A second PR will followup to fix the race itself.
This fixes a bug about aliases authorization.
That is, a user might see aliases which he is not authorized to see.
This manifests when the user is not authorized to see any aliases
and the `GetAlias` request is empty which normally is a marking
that all aliases are requested. In this case, no aliases should be
returned, but due to this bug, all aliases will have been returned.
The `ignore` set contains entries of type Class<?>, but the check is performed
on Path objects. This always returns false so is useless currently. Looking at
the first commit of this test that already shows this behaviour this never
excluded anything, so it can be removed.
This change introduces stats per processors. Total, time, failed,
current are currently supported. All pipelines will now show all
top level processors that belong to it. Failure processors are not
displayed, however, the time taken to execute the failure chain is part
of the stats for the top level processor.
The processor name is the type of the processor, ordered as defined in
the pipeline. If a tag for the processor is found, then the tag is
appended to the type.
Pipeline processors will have the pipeline name appended to the name of
the name of the processors (before the tag if one exists). If more
then one pipeline is used to process the document, then each pipeline
will carry its own stats. The outer most pipeline will also include the
inner most pipeline stats.
Conditional processors will only included in the stats if the condition evaluates
to true.
Adds checks for parsed geo distance query. It is a bit hack-ish since it
compares with query's toString() output, but it is better than no
checks. The parsed query itself has default visibility, so we cannot
access it here unless we move the test to org.apache.lucene.document
package.
Fixes#34043
This switches from joda time to java time when resolving index names
using date math. This commit also removes two non registered settings
from the code, which could not be used anyway. An unused method was
removed as well.
Relates #27330
* Replace custom type names with _doc in REST examples.
* Avoid using two mapping types in the percolator docs.
* Rename doc -> _doc in the main repository README.
* Also replace some custom type names in the HLRC docs.
* Make accounting circuit breaker settings dynamic
These missed the original property making them dynamic. This fixes the issue so
these can now be set at any time.
Resolves#34368
Integrates the failure detectors with the Connection lifecycle, to fail nodes as soon as:
- a leader detects one of his followers disconnecting.
- a follower detects its leader disconnecting.
This change introduces stats per processors. Total, time, failed,
current are currently supported. All pipelines will now show all
top level processors that belong to it. Failure processors are not
displayed, however, the time taken to execute the failure chain is part
of the stats for the top level processor.
The processor name is the type of the processor, ordered as defined in
the pipeline. If a tag for the processor is found, then the tag is
appended to the type.
Pipeline processors will have the pipeline name appended to the name of
the name of the processors (before the tag if one exists). If more
then one pipeline is used to process the document, then each pipeline
will carry its own stats. The outer most pipeline will also include the
inner most pipeline stats.
Conditional processors will only included in the stats if the condition evaluates
to true.
As master-eligible nodes join or leave the cluster we should give them votes or
take them away, in order to maintain the optimal level of fault-tolerance in
the system. #33924 introduced the `Reconfigurator` to calculate the optimal
configuration of the cluster, and in this change we add the plumbing needed to
actually perform the reconfigurations needed as the cluster grows or shrinks.
Since #34288, we might hit deadlock if the FollowTask has more fetchers
than writers. This can happen in the following scenario:
Suppose the leader has two operations [seq#0, seq#1]; the FollowTask has
two fetchers and one writer.
1. The FollowTask issues two concurrent fetch requests: {from_seq_no: 0,
num_ops:1} and {from_seq_no: 1, num_ops:1} to read seq#0 and seq#1
respectively.
2. The second request which fetches seq#1 completes before, and then it
triggers a write request containing only seq#1.
3. The primary of a follower fails after it has replicated seq#1 to
replicas.
4. Since the old primary did not respond, the FollowTask issues another
write request containing seq#1 (resend the previous write request).
5. The new primary has seq#1 already; thus it won't replicate seq#1 to
replicas but will wait for the global checkpoint to advance at least
seq#1.
The problem is that the FollowTask has only one writer and that writer
is waiting for seq#0 which won't be delivered until the writer completed.
This PR proposes to replicate existing operations with the old primary
term (instead of the current term) on the follower. In particular, when
the following primary detects that it has processed an process already,
it will look up the term of an existing operation with the same seq_no
in the Lucene index, then rewrite that operation with the old term
before replicating it to the following replicas. This approach is
wait-free but requires soft-deletes on the follower.
Relates #34288
When a envelope that crosses the dateline is specified as a part of
geo_shape query is parsed it shouldn't have its left and right points
flipped.
Fixes#34418
The shard suggestion sort uses a different tie-break than the one that is used
to merge different shards responses. The former uses the internal document identifier
when scores are the same whereas the latter compares the surface form first.
Because of this discrepancy some suggestion outputs are linked to the wrong documents
because the merge sort reorders the shard suggestions differently. This change
fixes this bug by duplicating the Lucene collector in order to be able to apply the
same tiebreak strategy than the merge sort. This logic will be removed when
https://issues.apache.org/jira/browse/LUCENE-8529 is fixed.
Closes#34378
Today we rely on the LocalCheckpointTracker to ensure no duplicate when
enabling optimization using max_seq_no_of_updates. The problem is that
the LocalCheckpointTracker is not fully reloaded when opening an engine
with an out-of-order index commit. Suppose the starting commit has seq#0
and seq#2, then the current LocalCheckpointTracker would return "false"
when asking if seq#2 was processed before although seq#2 in the commit.
This change scans the existing sequence numbers in the starting commit,
then marks these as completed in the LocalCheckpointTracker to ensure
the consistent state between LocalCheckpointTracker and Lucene commit.
Exclusion setting `cluster.routing.allocation.exclude._host` default value
is an empty string.
When an exclusion setting is sent with a null value the
o.e.c.s.Setting#innerGetRaw API return an empty string (probably to
avoid a NullPointerException to be raised).
The o.e.c.r.a.d.FilterAllocationDecider class is developed to omit
updates of default values for exclusion setting.
That's why a null exclusion setting value is translated to an empty
string which is equals to the exclusion default value which is
configured to be ignored.
A simple fix would be to not omit default values for exclusion setting
and keep the NullPointerException guard. This is the purpose of this
commit.
Closes#32721
The `term` and `phrase` suggesters have different options to filter candidates
based on their frequencies. The `popular` mode for instance filters candidate
terms that occur in less docs than the original term. However when we compute this threshold
we use the total term frequency of a term instead of the document frequency. This is not inline
with the actual filtering which is always based on the document frequency. This change fixes
this discrepancy and clarifies the meaning of the different frequencies in use in the suggesters.
It also ensures that the threshold doesn't overflow the maximum allowed value (Integer.MAX_VALUE).
Closes#34282
With this commit we cleanup hand-coded duplicate checks in XContent
parsing. They were necessary previously but since we reconfigured the
underlying parser in #22073 and #22225, these checks are obsolete and
were also ineffective unless an undocumented system property has been
set. As we also remove this escape hatch, we can remove the additional
checks as well.
Closes#22253
Relates #34588
In order to stay BWC compatible with joda time, the epoch millis date
formatter needs to parse dates with a dot like `123.45`. This
adds this functionality for the epoch millis parser in the same way as
for the epoch seconds parser. It also adds support for scientific
notations like `1.0e3` and fixes parsing of negative values for epoch
seconds and epoch millis.
We wish to commit a cluster state update after having received a response from
more than half of the master-eligible nodes in the cluster. This is optimal:
requiring either more or fewer votes than half harms resilience. For instance
if we have three master nodes then, we want to be able to commit a cluster
state after receiving responses from any two nodes; requiring responses from
all three is clearly not resilient to the failure of any node, and if we could
commit an update after a response from just one node then that node would be
required for every commit, which is also not resilient.
However, this means we must adjust the configuration (the set of voting nodes
in the cluster) whenever a master-eligible node joins or leaves. The
calculation of the best configuration for the cluster is the job of the
Reconfigurator, introduced here.
When we upgrade an index, we set the settings version upgraded
setting. This should be considered a settings change, and therefore we
need to increment the settings version. This commit addresses that.
Applies our line length guidance for all classes in the server in `lucene`
directories *except* `XMoreLikeThis`. The only long line in
`XMoreLikeThis` says "remove this when we upgrade to Lucene 5. Given
that we're on Lucene 8, this is a little terrifying and deserves another
look.
In remote cluster setup if we see a configured proxy we should set
the seed nodes host name as the `server_name` to trigger SNI based
routing even for seed nodes. Since remote cluster connections are
plain TCP connections we have to set the host manually since the other
side can't take it from the request URL like in the HTTP case.
This also adds some more informative logging to remote cluster connection.
Switch MetaDataStateFormat to Lucene directory abstraction
This commit switches MetaDataStateFormat class to Lucene directory abstraction to make it easier to test MetaDataStateFormat for different IO failures.
This commits also adds different IO failures tests to MetaDataStateFormatTests.
in #28741 RolloverIT fails because we are cutting over to the
next day while the test executes. We assume that this doesn't happen
based on the assertions in the test. This adds a assumeTrue to ensure
we are at least 5 min away form a date-flip.
Closes#28741
`Engine.Searcher` is non-final today which makes it error prone
in the case of wrapping the underlying reader or lucene `IndexSearcher`
like we do in `IndexSearcherWrapper`. Yet, there is no subclass of it yet
that would be dramatic to just drop on the floor. With the start of development
of frozen indices this changed since in #34357 functionality was added to
a subclass which would be dropped if a `IndexSearcherWrapper` is installed on an index.
This change locks down the `Engine.Searcher` to prevent such a functionality trap.
The `AutoFollowTests` needs to restart the clusters between each tests, because
it is using auto follow stats in assertions. Auto follow stats are only reset
by stopping the elected master node.
Extracted the `testGetOperationsBasedOnGlobalSequenceId()` test to its own test, because it just tests the shard changes api.
* Renamed AutoFollowTests to AutoFollowIT, because it is an integration test.
Renamed ShardChangesIT to IndexFollowingIT, because shard changes it the name
of an internal api and isn't a good name for an integration test.
* move creation of NodeConfigurationSource to a seperate method
* Fixes issues after merge, moved assertSeqNos() and assertSameDocIdsOnShards() methods from ESIntegTestCase to InternalTestCluster, so that ccr tests can use these methods too.
This change disallows negative query boosts. Negative scores are not allowed in Lucene 8 so
it is easier to just disallow negative boosts entirely. We should also deprecate negative boosts
in 6x in order to ensure that users are aware when they'll upgrade to ES 7.
Relates #33309
This commit introduces settings version to index metadata. This value is
monotonically increasing and is updated on settings updates. This will
be useful in cross-cluster replication so that we can request settings
updates from the leader only when there is a settings update.
Today when submitting an update settings request to update the number of
replicas with a wildcard that does not match any indices and allow no
indices is set to true, the request ends up being interpreted as
updating the number of replicas for all indices. That is, consider the
following sequence:
PUT /test-index
{
"settings": {
"index.number_of_replicas": 0
}
}
PUT /non-existent-*/_settings?expand_wildcards=open&allow_no_indices=true
{
"settings": {
"index.number_of_replicas": 1
}
}
GET /test-index/_settings
The latter will show that the number of replicas on test-index is now
one. This is surprising, and should be considered a bug.
The underlying problem here is treating no indices in the underlying
methods used to update the routing table and the metadata as meaning all
indices. This commit takes away this assumption. Tests that relied on
this behavior have been changed to no longer rely on this.
A test for this situation is added in UpdateNumberOfReplicasIT.
This removes another two methods from `AbstractComponent`. One isn't
used at all and another is only used in a single class in watcher. I've
moved the method that watcher uses into the single class that uses it.
This commit handles cases testing withLocale and withZone when the zone
and locale in question is the same as the special base case. This can
happen sometimes since the locale and zoneids are randomized.
Empty values on keyword fields are filtered by the `map` execution mode
of the `terms` aggregation. This commit restores them as valid buckets.
Closes#34434
This commit removes randomization of locale for DateFormatter equals
tests, instead using explicit locales. The test framework already
randomizes locales, so the random choice of the second locale can
sometimes be equal to the already chosen locale. Randomization also does
not provide any extra protection, as the equality of DateFormatter does
not implement equality of the locales itself.
closes#34337
Today we assume the storage layer operates perfectly in CoordinatorTests, which
means we are not testing that the system's invariants are preserved if the
storage layer fails for some reason. This change injects (rare) storage-layer
failures during the safety phase to cover these cases.
Today, fixLag() waits for a new cluster state to be committed. However, it does
not account for the fact that a term bump may occur, requiring a new election
to take place after the cluster state is committed. This change fixes this.
Today we may schedule two elections very close together, which can cause the
first election to fail even if there are no other nodes. This change adds a
delay in between subsequent elections on the same node, effectively allowing
time for each election to complete before scheduling the next one.
ListenableFuture may run a listener on the same thread that called the
addListener method or it may execute on another thread after the future
has completed. Whenever the ListenableFuture stores the listener for
execution later, it should preserve the thread context which is what
this change does.
Today we rewrite the operations from the leader with the term of the
following primary because the follower should own its history. The
problem is that a newly promoted primary may re-assign its term to
operations which were replicated to replicas before by the previous
primary. If this happens, some operations with the same seq_no may be
assigned different terms. This is not good for the future optimistic
locking using a combination of seqno and term.
This change ensures that the primary of a follower only processes an
operation if that operation was not processed before. The skipped
operations are guaranteed to be delivered to replicas via either
primary-replica resync or peer-recovery. However, the primary must not
acknowledge until the global checkpoint is at least the highest seqno of
all skipped ops (i.e., they all have been processed on every replica).
Relates #31751
Relates #31113
Questions on how to work with `ActionPlugin#getRestHandlerWrapper()`
come up in discuss forums all the time. This change adds an example
to the javadoc how this method should/could be used.
Today we inject the initial configuration of the cluster (i.e. the set of
voting nodes) at startup. In reality we must support injecting the initial
configuration after startup too. This commit adds low-level support for doing
so as safely as possible.
ES is scanning for dangling indices on every cluster state update. For this, it lists the subfolders of
the indices directory to determine which extra index directories exist on the node where there's no
corresponding index in the cluster state. These are potential targets for dangling index import. On
certain machine types, and with large number of indices, this subfolder listing can be horribly slow.
This means that every cluster state update will be slowed down by potentially hundreds of
milliseconds. One of the reasons for this poor performance is that Files.isDirectory() is a relatively
expensive call on some OS and JDK versions. There is no need though to do all these isDirectory
calls for folders which we know we are going to discard anyhow in the next step of the dangling
indices logic. This commit allows adding an exclusion predicate to the availableIndexFolders
methods which can dramatically speed up this method when scanning for dangling indices.
The hack to work around lag detection had some issues:
- it always called runFor(), even if no lag was detected
- it looked at the last-accepted state not the last-applied state, so missed
some lag situations.
This fixes these issues.
Since all calls to `ESLoggerFactory` outside of the logging package were
deprecated, it seemed like it'd simplify things to migrate all of the
deprecated calls and declare `ESLoggerFactory` to be package private.
This does that.
Today we accept that some nodes may vote for the wrong master in an election.
This is mostly fine because they do end up joining the correct master in the
end, but the lack of a vote from every follower may prevent a future desirable
reconfiguration from taking place.
The solution is to hold another election in a yet-higher term in order to
collect a complete set of votes. Elections are somewhat disruptive so we should
think carefully about when this election should take place. One option is to
wait as late as possible (on the grounds that it might not ever be necessary).
This unfortunately makes it harder to predict how an
apparently-smoothly-running cluster will react to nodes leaving and joining.
Instead we prefer to perform the election as soon as possible in the leader's
term, adding "votes from all followers" to the invariants that we expect to
hold in a stable cluster. The start of a leader's term is already a somewhat
disrupted time for the cluster, so performing another election at this point
does not materially change the cluster's behaviour.
This change implements the logic needed to trigger a new election in order to
satisfy this extra stabilisation condition.
With this commit we restore the previous behavior in
`BigArraysTests#testMaxSizeExceededOnResize` but lower the sizes that
are tested to the range between 256 bytes to 16 kB so the test does not
produce a whole lot of garbage.
The previous attempt to reduce the amount of garbage produced by that
test was to properly size the array initially but it failed to account
for object alignment which lead to test failures in some cases. While it
would be possible to account for object alignment, we would need to open
up BigArrays or directly use the underlying Lucene API which would
require us to allocate an array upfront only to find its size (incl.
object alignment).
Instead we have fixed this issue by conservatively sizing the array
initially (so the initial allocation will never trip the circuit
breaker) and reduce garbage by reducing the circuit breaker's upper
bound as described previously.
Closes#33750
Relates #34325
From experience with #34257, here are a few things that help with analysing
logs from test runs. Also we prevent trying to stabilise a cluster with raised
delay variability, because lowering the delay variability requires time to
allow all the extra-varied-scheduled tasks to work their way out of the system.
This changes the delete job API by adding
the choice to delete a job asynchronously.
The commit adds a `wait_for_completion` parameter
to the delete job request. When set to `false`,
the action returns immediately and the response
contains the task id.
This also changes the handling of subsequent
delete requests for a job that is already being
deleted. It now uses the task framework to check
if the job is being deleted instead of the cluster
state. This is a beneficial for it is going to also
be working once the job configs are moved out of the
cluster state and into an index. Also, force delete
requests that are waiting for the job to be deleted
will not proceed with the deletion if the first task
fails. This will prevent overloading the cluster. Instead,
the failure is communicated better via notifications
so that the user may retry.
Finally, this makes the `deleting` property of the job
visible (also it was renamed from `deleted`). This allows
a client to render a deleting job differently.
Closes#32836
The `status` part of the tasks API reflects the internal status of a
running task. In general, we do not make backwards breaking changes to
the `status` but because it is internal we reserve the right to do so. I
suspect we will very rarely excercise that right but it is important
that we have it so we're not boxed into any particular implementation
for a request.
In some sense this is policy making by documentation change. In another
it is clarification of the way we've always thought of this field.
I also reflect the documentation change into the Javadoc in a few
places. There I acknowledge Kibana's "special relationship" with
Elasticsearch. Kibana parses `_reindex`'s `status` field and, because
we're friends with those folks, we should talk to them before we make
backwards breaking changes to it. We *want* to be friends with everyone
but there is only so much time in the day and we don't *want* to make
backwards breaking fields to `status` at all anyway. So we hope that
breaking changes documentation should be enough for other folks.
Relates to #34245.
* SCRIPTING: Add Expr. Compile for TermSetQuery Ctx.
* Follow up to #33602 adding the ability to compile TermsSetQuery
scripts with the expressions engine in the same way we support
SearchScript in Expressions
* Duplicated the code here for now to make the change less complex,
the only difference to SearchScript is that `_score` and `_value` are not handled for TermsSetQuery
* remove redundant check
Drops the last logging constructor that takes `Settings` because it is
no longer needed.
Watcher goes through a lot of effort to pass `Settings` to `Logger`
constructors and dropping `Settings` from all of those calls allowed us
to remove quite a bit of log-based ceremony from watcher.
Today's CoordinatorTests have a limited amount of randomisation in how things
are scheduled. However, to be fully confident in Zen2's liveness we require the
system to stabilise after any permitted sequence of events. We can achieve
this by running the system in a much more random fashion for a while, with much
larger variation in when things are scheduled (simulating GC pressure and
network disruption) and then continuing to assert that the system stabilises as
we expect. When running randomly, we do not expect to make significant progress
and merely verify that no safety property is violated.
This change introduces the runRandomly() test method which implements this
idea. It also fixes a handful of liveness bugs that this first version of
runRandomly() exposed.
Today we use the version of a DirectoryReader as a component of the key
of IndicesRequestCache. This usage is perfectly fine since the version
is advanced every time a new change is made into IndexWriter. In other
words, two DirectoryReaders with the same version should have the same
content. However, this invariant is only guaranteed in the context of a
single IndexWriter because the version is reset to the committed version
value when IndexWriter is re-opened.
Since #33473, each IndexShard may have more than one IndexWriter, and
using the version of a DirectoryReader as a part of the cache key can
cause IndicesRequestCache to return stale cached values. For example, in
#27650, we rollback the engine (i.e., re-open IndexWriter), index new
documents, refresh, then make a count request, but the search layer
mistakenly returns the count of the DirectoryReader of the previous
IndexWriter because the current DirectoryReader has the same version of
the old DirectoryReader even their documents are different. This is
possible because these two readers come from different IndexWriters.
This commit replaces the the version with the reader cache key of
IndexReader as a component of the cache key of IndicesRequestCache.
Closes#27650
Relates #33473
This commit adds the support to early terminate the collection of a leaf
in the min/max aggregator. If the query matches all documents the min and max value
for a numeric field can be retrieved efficiently in the points reader.
This change applies this optimization when possible.
* Make text message not required in constructor for slack
* Remove unnecessary comments in test file
* Throw exception when reduce or combine is not provided; update tests
* Update integration tests for scripted metrics to always include reduce and combine
* Remove some old changes from previous branches
* Rearrange script presence checks to be earlier in build
* Change null check order in script builder for aggregated metrics; correct test scripts in IT
* Add breaking change details to PR
Today we reverse the initial order of the nested documents when we
index them in order to ensure that parents documents appear after
their children. This means that a query will always match nested documents
in the reverse order of their offsets in the source document.
Reversing all documents is not needed so this change ensures that parents
documents appear after their children without modifying the initial order
in each nested level. This allows to match children in the order of their
appearance in the source document which is a requirement to efficiently
implement #33587. Old indices created before this change will continue
to reverse the order of nested documents to ensure backwark compatibility.
* Adds trace logging to IndicesRequestCache
This change adds trace level logging to `IndicesrrequestCache` witht eh
primary aim of helping to identify the cause of teh failures in
https://github.com/elastic/elasticsearch/issues/32827. The cache will
log at trace level when a cache hit or miss occurs including the reader
version and the cache key. Note that this change adds a
`cacheKeyRenderer` whcih supplies a human readable String of the cache
key since the actual cache key itself is a `BytesReference` containing
the wire protocol serialised form of the request.
Logging is also added for the case where a search timeout occurs and fr
that reason the cache entry is invalidated.
* Adds comment to remaind us to remove cacheKeyRenderer
In #28941 we changed the computation of cluster state task descriptions but
this introduced a bug in which we only log the empty descriptions (rather than
the non-empty ones). This change fixes that.
Optionals containing boxed primitive types are prohibitively costly because they
have two level of boxing. For Optional<Integer> the analogous OptionalInt can be
used to avoid the boxing of the contained int value.
This change fixes a bug in the cross fields mode of the `query_string`
query. The multi fields query builder must be reseted before parsing
in order to clear the list of expanded fields coming from the previous text block.
Closes#34215
Mappings with completion type and multi-fields, were not able to index array or
object format on completion fields. Only string format was supported.
This is fixed by providing multiField parser with externalValueContext with already parsed object
closes#15115
This adds some method into the `DateFormatter` interface, namely
* `withLocale()` to change the locale of a date formatter
* `getLocale()`
* `getZone()`
* `hashCode()`
* `equals()`
These methods will be needed for aggregations and mapping changes, where
zones and locales can be specified in the mapping or in search/aggs
parts of a search request.
Today we schedule tasks (both immediate and future ones) exactly when
requested. In fact it is more realistic to allow for a small amount of delay in
the scheduling of tasks, and this helps to exercise more interleavings of
actions and therefore to improve test coverage.
This change adds to the DeterministicTaskQueue the ability to add a random
delay to the scheduling of tasks.
This change also provides more explicit timeouts for stabilisation in the
CoordinatorTests.
Using the randomised scheduling feature in the CoordinatorTests also found a
situation in which we could become a leader, then a candidate, and then a
leader again very quickly, causing a clash of the _BECOME_MASTER_ and
_FINISH_ELECTION_ tasks. We change their behaviour to not consider these
duplicates to be problematic.
When nested objects are present in the mappings, we add a filter in
queries to exclude them if there is no evidence that the query cannot
match in this space. In 6x we visit the query in order to find a mandatory
clause that can match root documents only. If we find one we can omit the
nested documents filter. Currently only `term` and `range` queries are checked,
this change adds the support for `terms` query to effectively remove the nested filter
if a mandatory `terms` clause targets a non-nested field.
Closes#34067
Mainly this fixes a warning by replacing the unchecked `new ActionListener`
with the checked `new ActionListener<Response>`, and it also fixes the line
length violations in this class.
This commit adds a check for "enabled" attribute change for types when
a RestPutMappingAction is received. A MappingException is thrown when
such a change is detected. Change are prevented in both ways: "false -> true"
and "true -> false".
Closes#33566
#32281 adds elasticsearch-shard to provide bwc version of elasticsearch-translog for 6.x; have to remove elasticsearch-translog for 7.0
Relates to #31389
The unfollow API changes a follower index into a regular index, so that it will accept write requests from clients.
For the unfollow api to work the index follow needs to be stopped and the index needs to be closed.
Closes#33931
This change introduces the indexing optimization using sequence numbers
in the FollowingEngine. This optimization uses the max_seq_no_updates
which is tracked on the primary of the leader and replicated to replicas
and followers.
Relates #33656
Tanguy Leroux
Andrey Ershov
Hendrik Muhs
Christoph Büscher
David Turner
Vladimir Dolzhenko
Simon Willnauer
Jason Tedor
Tim Brooks
Igor Motov
Yannick Welsch
Nick Knize
Jim Ferenczi
Daniel Mitterdorfer
Alpar Torok
Ryan Ernst
Nhat Nguyen
Martijn van Groningen
Vladimir Dolzhenko
Julie Tibshirani
javanna
Nik Everett
Jim Ferenczi
Armin Braun
Alexander Reelsen
Boaz Leskes
Jack Conradson
Colin Goodheart-Smithe
Andy Bristol
Tal Levy
Seong-hyun, Oh
lipsill
Luca Cavanna
Alan Woodward
Przemyslaw Gomulka
Mark Tozzi
Jake Landis
Jay Modi
Lee Hinman
Gordon Brown
Lee Hinman
Christophe Bismuth
Andrey Atapin
Stéphane Campinas
Mayya Sharipova
Andrey Ershov
Zachary Tong
Albert Zaharovits
markharwood
fonxian
eray
Dimitris Athanasiou
Kazuhiro Sera
albendz
Dimitrios Liappis