Pruning tombstones is best effort and should not block if a key is currently
locked. This can cause a deadlock in rare situations if we switch of append
only optimization while heavily updating the same key in the engine
while the LiveVersionMap is locked. This is very rare since this code
patch only executed every 15 seconds by default since that is the interval
we try to prune the deletes in the version map.
Closes#28714
This commit fixes an issue with setting plugin.mandatory to include a
meta-plugin. The issue here is that the names that we collect are the
underlying plugins, not the meta-plugin. We should not use the
underlying plugins instead using the names of non-meta plugins and the
names of meta-plugins. This commit addresses this. The strategy here is
that when we look at the installed plugins on the filesystem, we keep
track of which ones are meta-plugins and carry this information up to
where check which plugins are installed against the mandatory plugins.
Relates #28710
We previously specified the -server flag to force the JVM to use the
server JVM. This is the default on all the systems that we support when
using a 64-bit JVM (and we no longer support 32-bit JVMs). There was
some trouble with this flag for the Windows service since procrun did
not understand what to do with it; as such, we had to filter this flag
out in the service. When we migrated to parsing JVM options in Java (via
the JVM options parser) we simplified this situation and removed
specifying the -server flag. This commit removes a leftover statement
that we are forcing the server JVM.
Relates #28738
GceDiscoverTests can be simplified in a similar manner than #27945. It
now uses a mocked GceInstancesService that exposes internal test cluster
nodes as if they were real GCE nodes. It should also make the test more
robust by not using a HTTP server anymore.
closes#24313
Today we have several levels of indirection to acquire an Engine.Searcher.
We first acquire a the reference manager for the scope then acquire an
IndexSearcher and then create a searcher for the engine based on that.
This change simplifies the creation into a single method call instead of
3 different ones.
The node stats API enables filtlering the top-level stats for only
desired top-level stats. Yet, this was never enabled for adaptive
replica selection stats. This commit enables this. We also add setting
these stats on the request builder, and fix an inconsistent name in a
setter.
Relates #28721
This is a fix for #28729. Currently if a write operation is not properly
queued with a selector we notify the listener. However, we are doing
this by calling a method that is only meant for the selector thread to
call. This commit fixes that issue.
We currently revisit the index deletion policy whenever the global
checkpoint has advanced enough. We should also revisit the deletion
policy after releasing the last snapshot of a snapshotting commit. With
this change, the old index commits will be cleaned up as soon as
possible.
Follow-up of #28140https://github.com/elastic/elasticsearch/pull/28140#discussion_r162458207
Today we maintain a copy of every delete in the live version maps. This is unnecessary
and might add quite some overhead if maps grow large. This change moves out the deletes
tracking into the tombstone map only and relies on the cleaning of tombstones when deletes
are collected.
The AdaptiveSelectionStats object serializes the clientOutgoingConnections map that's concurrently updated in SearchTransportService. Serializing the map consists of first writing the size of the map and then serializing the entries. If the number of entries changes while the map is being serialized, the size and number of entries go out of sync. The deserialization routine expects those to be in sync though.
Closes#28713
The acquireIndexCommit was separated into acquireSafeIndexCommit and
acquireLastIndexCommit, however the test was not updated accordingly.
Relates #28271
The Windows service will use a private temporary directory under the
user that is performing the installation. In cases when the service will
run as a different user, operators need a method to set this temporary
directory elsewhere. We have such a mechanism, so this commit merely
adds a note to the documentation on how to utilize it.
Relates #28712
Previously we introduced a new parameter to `acquireIndexCommit` to
allow acquire either a safe commit or a last commit. However with the
new parameters, callers can provide a nonsense combination - flush first
but acquire the safe commit. This commit separates acquireIndexCommit
method into two different methods to avoid that problem. Moreover, this
change should also improve the readability.
Relates #28038
We added a rest test for the translog last modified age without a
version check. This causes BWC failed because the stats are not
available in the old versions.
Relates #28613
* Remove deprecated createParser methods
This removes the final instances of the callers of `XContent.createParser` and
`XContentHelper.createParser` that did not pass in the `DeprecationHandler`. It
also removes the now-unused deprecated methods and fully removes any mention of
Log4j or LoggingDeprecationHandler from the XContent code.
Relates to #28504
* Add comments in JsonXContentGenerator
This test has a race condition. The action listener used to listen for
connections has a guard against being executed twice. However, this
listener can be executed twice. After on success is invoked the test
starts to tear down. At this point, the threads the test forked will
terminate and the remote cluster connection will be closed. However, a
thread forked to the management thread pool by the remote cluster
connection can still be executing and try to continue connecting. This
thread will be cancelled when the remote cluster connection is closed
and this leads to the action listener being invoked again. To address
this, we explicitly check that the reason that on failure was invoked
was cancellation, and we assert that the listener was already previously
invoked. Interestingly, this issue has always been present yet a recent
change (#28667) exposed errors that occur on tasks submitted to the
thread pool and were silently being lost.
Relates #28695
Previously a user could set a custom config path to a relative directory
using ES_PATH_CONF. In a previous change related to enabling GC logging
by default, we forced the working directory for Elasticsearch to be
ES_HOME. This had the impact of causing all relative paths to be
relative to ES_HOME, against the intent of the user. This commit
addresses this by making ES_PATH_CONF absolute before we switch the
working directory to ES_HOME.
Relates #28700
In order to allow us to gradually move to passing the deprecation handler is, we
need a shim that contains both the non-passed and passed version.
Relates to #28504
When we submit a task to a thread pool for asynchronous execution, we
are returned a future. Since we submitted to go asynchronous, these
futures are not inspected for failure (we would have to block a thread
to do that). While we have on failure handlers for exceptions that are
thrown during execution, we do not handle throwables that are not
exceptions and these end up silently lost. This commit adds a check
after the runnable returns that inspects the status of the future. If an
unhandled throwable occurred during execution, this throwable is
propogated out where it will land in the uncaught exception handler.
Relates #28667
The recent addition of the _cluster/settings API was merged together with another change that added encoding of the different URL parts. That broke the cluster PUT settings API straight-away. This commit fixes this problem. The '/' that's part of the /_cluster/settings endpoint should not be encoded.
The REST high-level client supports now encoding of path parts, so that for instance documents with valid ids, but containing characters that need to be encoded as part of urls (`#` etc.), are properly supported. We also make sure that each path part can contain `/` by encoding them properly too.
Closes#28625
We have code used in the networking layer to search for errors buried in
other exceptions. This code will be useful in other locations so with
this commit we move it to our exceptions helpers.
Relates #28691
Elasticsearch 6.x indices do not allow multiple types index. Instead, they use "_doc" as default if created internally (Elasticsearch), or "doc" default if sent by Logstash.
Currently the Translog constructor is capable both of opening an existing translog and creating a
new one (deleting existing files). This PR separates these two into separate code paths. The
constructors opens files and a dedicated static methods creates an empty translog.
Simon Willnauer
Luca Cavanna
Jason Tedor
Tanguy Leroux
Nhat Nguyen
Martijn van Groningen
Tim Brooks
Alejandro Carstens Cattori
Yannick Welsch
Clinton Gormley
Ryan Ernst
Lee Hinman
Justin Wyer
Colin Goodheart-Smithe
Andy Bristol
Cladis
javanna
olcbean
Paul Schwarz
Kuaaaly
Alex Moros Marco
Boaz Leskes
Ke Li