We call `ensureConnections()` to undo the effects of a disruption. However, it
is possible that one or more targets are currently CONNECTING and have been
since the disruption was active, and that the connection attempt was thwarted
by a concurrent disruption to the connection. If so, we cannot simply add our
listener to the queue because it will be notified when this CONNECTING activity
completes even though it was disrupted. We must therefore wait for all the
current activity to finish and then go through and reconnect to any missing
nodes.
Closes#40030.
Today we load the shard history retention leases from disk whenever opening the
engine, and treat a missing file as an empty set of leases. However in some
cases this is inappropriate: we might be restoring from a snapshot (if the
target index already exists then there may be leases on disk) or
force-allocating a stale primary, and in neither case does it make sense to
restore the retention leases from disk.
With this change we write an empty retention leases file during recovery,
except for the following cases:
- During peer recovery the on-disk leases may be accurate and could be needed
if the recovery target is made into a primary.
- During recovery from an existing store, as long as we are not
force-allocating a stale primary.
Relates #37165
Today we test Zen1/Zen2 compatibility by running 7.x nodes with a "fake" Zen1
implementation. However this is not a truly faithful test because these nodes
do known how to properly deserialize a 7.x cluster state, voting configurations
and all, whereas a real Zen1 node is in 6.7 and ignores the coordination
metadata.
We only ever apply a cluster state that's been committed, which in Zen2
involves setting the last-committed configuration to equal the last-accepted
configuration. Zen1 knows nothing about this adjustment, so it is possible for
these to differ. This breaks the assertion that the cluster states are equal on
all nodes after integration tests.
This commit fixes this by implementing this adjustment in Zen1 before applying
a cluster state.
Fixes#40055.
This change ensures that we do not make assumptions about the length
of the input that we can read from the stdin. It still consumes only
one line, as the previous implementation
* Handle UTF8 values in the keystore
Our current implementation uses CharBuffer#array to get the chars
that were decoded from the UTF-8 bytes. The backing array of
CharBuffer is created in CharsetDecoder#decode and gets an initial
length that is the same as the length of the ByteBuffer it decodes,
hence the number of UTF-8 bytes.
This works fine for the first 128 characters where each one needs
one bytes, but for the next UTF-8 characters (other latin alphabets
Greek, Cyrillic etc.) where we need 2 to 4 bytes per character, this
backing char array has a larger size than the number of the actual
chars this CharBuffer contains. Calling `array()` on it will return
a char array that can potentially have extra null chars so the
SecureString we get from the KeystoreWrapper, is not the same as the
one we entered.
This commit changes the behavior to use Arrays#copyOfRange to get
the necessary chars from the CharBuffer and adds a test with
random ( maybe not printable ) UTF-8 strings
Computing the compressed size of the cluster state on every invocation
of cluster:monitor/state action is expensive, and the value of this
field is dubious anyway. Therefore we want to remove computing this
field. As a first step, we stop computing and return this field by
default. To avoid breaking users, we will give them a system property to
use to tide them over until the next major release when we will actually
remove this field. This comes with a deprecation warning too, and the
backport to the appropriate minor will also include a note in the
migration guide. There will be a follow-up to remove this field in the
next major version.
After the joda-java time migration we were formatting zone ids with zoneOrOffsetId method. This when a date was provided with a ZoneRegion for instance America/Edmonton it was appending this zone identifier instead of zone formatted as +HH:MM.
This fix is changing the format of zone suffix for all printers and also always wrapping a Temporal into a ZonedDateTime when formatting.
closes#38471
backport #39568
Currently there is a method `Recycler#obtain(size)` that allows a size
parameter to be passed. However all implementations ignore this
parameter and just allocate a page size based on other settings. This
commit removes this method.
When performing the test with 57 master-eligible nodes and one node
crash, we saw messy elections, when multiple nodes were attempting to
become master.
JoinHelper has logged 105 long log messages with lengthy stack
traces during one such election.
To address this, we decided to log these messages every time only on
debug level.
We will log last unsuccessful join attempt (along with a timestamp)
if any with WARN level if the cluster is failing to form.
(cherry picked from commit 17a148cc27b5ac6c2e04ef5ae344da05a8a90902)
Currently token filter settings are treated as fixed once they are declared and
used in an analyzer. This is done to prevent changes in analyzers that are already
used actively to index documents, since changes to the analysis chain could
corrupt the index. However, it would be safe to allow updates to token
filters at search time ("search_analyzer"). This change introduces a new
property of token filters that allows to mark them as only being usable at search
or at index time. Any analyzer that uses these tokenfilters inherits that property
and can be rejected if they are used in other contexts. This is a first step towards
making specific token filters (e.g. synonym filter) updateable.
Relates to #29051
Today, when applying new cluster state we attempt to connect to all of its
nodes as a blocking part of the application process. This is the right thing to
do with new nodes, and is a no-op on any already-connected nodes, but is
questionable on known nodes from which we are currently disconnected: there is
a risk that we are partitioned from these nodes so that any attempt to connect
to them will hang until it times out. This can dramatically slow down the
application of new cluster states which hinders the recovery of the cluster
during certain kinds of partition.
If nodes are disconnected from the master then it is likely that they are to be
removed as part of a subsequent cluster state update, so there's no need to try
and reconnect to them like this. Moreover there is no need to attempt to
reconnect to disconnected nodes as part of the cluster state application
process, because we periodically try and reconnect to any disconnected nodes,
and handle their disconnectedness reasonably gracefully in the meantime.
This commit alters this behaviour to avoid reconnecting to known nodes during
cluster state application.
Resolves#29025.
If a primary on 6.7 and a replica on 5.6 are running more than 5 minutes
(retention leases background sync interval), the retention leases
background sync will be triggered, and it will trip 6.7 node due to the
illegal checkpoint value. We can fix the problem by making the returned
checkpoint depends on the node version. This PR, however, chooses to
enforce retention leases require soft deletes, and make retention leases
sync noop if soft deletes is disabled instead.
Closes#39914
* The test failure in #39852 is caused by a file in the initial repository when there should not be any
* It seems that on a normal consistent file system no left-over file should exist ever here after the validation finishes and I can't reproduce or see any other path to a dangling file in the fresh respository
=> added a more verbose and strict assertion that will log what file is left over next time
* Relates #39852
This commit removes the "doc" type from monitoring internal indexes.
The template still carries the "_doc" type since that is needed for
the internal representation.
This change impacts the following templates:
monitoring-alerts.json
monitoring-beats.json
monitoring-es.json
monitoring-kibana.json
monitoring-logstash.json
As part of the required changes, the system_api_version has been
bumped from "6" to "7" and support for version "2" has been dropped.
A new empty pipeline is now introduced for the version "7", and
the formerly empty "6" pipeline will now remove the type and re-direct
the request to the "7" index.
Additionally, to due to a difference in the internal representation
(which requires the inclusion of "_doc" type) and external representation
(which requires the exclusion of any type) a helper method is introduced
to help convert internal to external representation, and used by the
monitoring HTTP template exporter.
Relates #38637
Executors of type fixed_auto_queue_size (i.e. search / search_throttled) wrap runnables into
TimedRunnable, which is an AbstractRunnable. This is dangerous as it might silently swallow
exceptions, and possibly miss calling a response listener. While this has not triggered any failures in
the tests I have run so far, it might help uncover future problems.
Follow-up to #36137
This test started failing since decreasing the leader and follower check timeouts (#38298). The
reason is that the test was relying on the default publication timeout to come into effect before
leader / follower check timeouts, which is now not always true anymore.
Closes#38867
Currently we log exceptions due to channel close at the debug level in
the normal exception handler. Currently we log all send failures due to
channel close at the warn level. This commit changes that to only log at
warn if the send failure is not due to channel closed. Additionally, it
adds the ssl engine closed as a channel close exception.
Today we use a ConcurrentHashSet to track the in-flight outgoing joins in the
`JoinHelper`. This is fine for adding and removing elements but not for the
emptiness test in `isJoinPending()` which might return false if one join
finishes just after another one starts, even though joins were pending
throughout.
As used today this is ok: it means the node was trying to join a master but
this join attempt just finished unsuccessfully, and causes it to (rightfully)
reject a `FollowerCheck` from the failed master. However this kind of API
inconsistency is trappy and there is no need to be clever here, so this change
replaces the set with a `synchronizedSet()`.
When ESRejectedExecutionException gets thrown on the coordinating node while trying to fetch hits, the resulting exception will hold no shard failures, hence `503` is used as the response status code. In that case, `429` should be returned instead. Also, the status code should be taken from the cause if available whenever there are no shard failures instead of blindly returning `503` like we currently do.
Closes#38586
The monitoring bulk API accepts the same format as the bulk API, yet its concept
of types is different from "mapping types" and the deprecation warning is only
emitted as a side-effect of this API reusing the parsing logic of bulk requests.
This commit extracts the parsing logic from `_bulk` into its own class with a
new flag that allows to configure whether usage of `_type` should emit a warning
or not. Support for payloads has been removed for simplicity since they were
unused.
@jakelandis has a separate change that removes this notion of type from the
monitoring bulk API that we are considering bringing to 8.0.
This commit propagates some exceptions that were previously swallowed and also
makes sure that exceptions closing streams are either propagated if the try
block succeeded or added as suppressed exceptions otherwise.
* [ML] refactoring lazy query and agg parsing
* Clean up and addressing PR comments
* removing unnecessary try/catch block
* removing bad call to logger
* removing unused import
* fixing bwc test failure due to serialization and config migrator test
* fixing style issues
* Adjusting DafafeedUpdate class serialization
* Adding todo for refactor in v8
* Making query non-optional so it does not write a boolean byte
This commit consolidates more mapping validation logic into the same class.
`FieldTypeLookup` is now a bit simpler, and has the sole responsibility of quickly
resolving field names to their types.
I have a broader refactor planned around mapping merge validation, but this
change should at least be a step in the right direction.
These simplifications to `MapperMergeValidator` are possible now that there is
always a single mapping definition.
* Remove the type argument in `validateMapperStructure`.
* Remove unnecessary checks against existing mappers.
If TransportService is stopped before a shard-failure request is sent
but after the request is registered, TransportService will notify
ReplicationOperation a TransportException with an error message:
"transport stop, action: internal:cluster/shard/failure".
Relates #39584
* Bundle java in distributions
Setting up a jdk is currently a required external step when installing
elasticsearch. This is particularly problematic for the rpm/deb packages
as installing a jdk in the same package installation command does not
guarantee any order, so must be done in separate steps. Additionally,
JAVA_HOME must be set and often causes problems in selecting a correct
jdk when, for example, the system java is an older unsupported version.
This commit bundles platform specific openjdks into each distribution.
In addition to eliminating the issues above, it also presents future
possible improvements like using jlink to build jdk images only
containing modules that elasticsearch uses.
closes#31845
Prior to this commit (and after 6.5.0), if an ingest node changes
the _index in a pipeline, the original target index would be created.
For daily indexes this could create an extra, empty index per day.
This commit changes the TransportBulkAction to execute the ingest node
pipeline before attempting to create the index. This ensures that the
only index created is the original or one set by the ingest node pipeline.
This was the execution order prior to 6.5.0 (#32786).
The execution order was changed in 6.5 to better support default pipelines.
Specifically the execution order was changed to be able to read the settings
from the index meta data. This commit also includes a change in logic such
that if the target index does not exist when ingest node pipeline runs, it
will now pull the default pipeline (if one exists) from the settings of the
best matched of the index template.
Relates #32786
Relates #32758Closes#36545
This commit introduces the forget follower API. This API is needed in cases that
unfollowing a following index fails to remove the shard history retention leases
on the leader index. This can happen explicitly through user action, or
implicitly through an index managed by ILM. When this occurs, history will be
retained longer than necessary. While the retention lease will eventually
expire, it can be expensive to allow history to persist for that long, and also
prevent ILM from performing actions like shrink on the leader index. As such, we
introduce an API to allow for manual removal of the shard history retention
leases in this case.
We need to unwrap and use the actual cause when determining if the node
with primary shard is shutting down because TransportService will throw
a TransportException wrapped in a SendRequestTransportException.
Relates #39584
Today when a replicated write operation fails to execute on a replica,
the primary will reach out to the master to fail that replica (and mark
it stale). We then won't ack that request until the master removes the
failing replica; otherwise, we will lose the acked operation if the
failed replica is still in the in-sync set. However, if a node with the
primary is shutting down, we might ack such request even though we are
unable to send a shard-failure request to the master. This happens
because we ignore NodeClosedException which is triggered when the
ClusterService is being closed.
Closes#39467
This adds the capability to snapshot replicated closed indices.
It also changes snapshot requests in v8.0.0 to automatically expand wildcards to closed indices and hence start snapshotting closed indices by default. For v7.1.0 and above, wildcards are by default only expanded to open indices, which can be changed by explicitly setting the expand_wildcards option either to all or closed.
Note that indices are always restored as open indices, even if they have been snapshotted as closed replicated indices.
Relates to #33888
Lucene added an optimization to leave the term dictionary on disk
for non-id like fields. This change happened very late in the release
processes such that it's better to have an escape hatch if certain
use-cases are hurt by this optimization. This setting might be
removed in the future if it turns out to be unnecessary.
Currently SearchServiceTests.testCloseSearchContextOnRewriteException can fail
if a refresh happens while we test for the SearchPhaseExecutionException that is
thrown later in the test. The test takes the current Store#refCount and expects
it to be the same after the exception is thrown. If a refresh happens in that
interval however, the refCound will be different, causing the test to fail. This
can be provoked e.g. by running this section in a tight loop.
Switching of refresh for this tests solves the issue.
When preparing the state to send to other nodes, we're serializing it
for each node, despite using putIfAbsent.
This commit checks if the state was already serialized for this node
version before performing the potentially expensive computation.
The map is not used by multiple threads, so computeIfAbsent is not
needed (and could not be used here easily, because IOException could
be thrown).
(cherry picked from commit c99be63b43f5250f3cd220130df73c5e9e097459)
David Turner
Ioannis Kakavas
Tamara Braun
Jason Tedor
Yogesh Gaikwad
Przemyslaw Gomulka
Jim Ferenczi
Tim Brooks
Andrey Ershov
Christoph Büscher
Andy Bristol
lzh3636
Nhat Nguyen
Armin Braun
Jake Landis
Yannick Welsch
Ankit Jain
Adrien Grand
Benjamin Trent
Julie Tibshirani
Ryan Ernst
Gordon Brown
David Roberts
markharwood
Simon Willnauer