The NodeConnectionsService currently determines which nodes to connect to / disconnect from by inspecting cluster state changes and connecting to added nodes / disconnecting from removed nodes. When a master steps down (for example due to another master-eligible node shutting down which brings the number of master-eligible nodes below minimum_master_master), and the connection to other existing nodes was dropped while pinging, however, the connection to these nodes is not re-established while publishing the first cluster state that establishes the node as master.
This commit changes the NodeConnectionsService connect / disconnect logic to always rely on the state that is to be / was published, looking not only at the added / removed nodes, but validating that exactly all nodes that are currently registered in NodeConnectionsService are connected (corresponds to a NOOP if the node is already connected).
Right now closing a shard looks like it strands refresh listeners,
causing tests like
`delete/50_refresh/refresh=wait_for waits until changes are visible in search`
to fail. Here is a build that fails:
https://elasticsearch-ci.elastic.co/job/elastic+elasticsearch+multi_cluster_search+multijob-darwin-compatibility/4/console
This attempts to fix the problem by implements `Closeable` on
`RefreshListeners` and rejecting listeners when closed. More importantly
the act of closing the instance flushes all pending listeners
so we shouldn't have any stranded listeners on close.
Because it was needed for testing, this also adds the number of
pending listeners to the `CommonStats` object and all API to which
that flows: `_cat/nodes`, `_cat/indices`, `_cat/shards`, and
`_nodes/stats`.
This is related to #22116. A logIfNecessary() call makes a call to
NetworkInterface.getInterfaceAddresses() requiring SocketPermission
connect privileges. By moving this to bootstrap the logging call can be
made before installing the SecurityManager.
Today we execute the low level handshake on the TCP layer in #connectToNode.
If #openConnection is used directly, which is truly expert, no handshake is executed
which allows connecting to nodes that are not necessarily compatible. This change
moves the handshake to #openConnection to prevent bypassing this logic.
This integrates the mocksocket jar with elasticsearch tests. Mocksocket wraps actions requiring SocketPermissions in doPrivilege blocks. This will eventually allow SocketPermissions to be assigned to the mocksocket jar opposed to the entire elasticsearch codebase.
Notifications for request tracing are invoked concurrently and can still
be in flight once a tracer is installed in the test. This can lead to side-effects
since the test relied on exact invocations. This commit adds action filtering
to the test tracer to only count invocations for the relevant actions.
Closes#22418
There could be an issue creating the REST clients and/or making the first request to the external cluster. If that happens, the blacklist has already been assigned and the following tests will fail because of an assertion that checks that the blacklist is not already assigned when the contexts are not.
Today we silently ignore invalid test logging annotations. This commit
rejects these annotations, failing the processing of the annotation and
aborting the test.
This commit adds a test for applying logging levels in hierarchical
order, and addresses an issue with restoring the logging levels at the
end of a test or suite.
This commit factors out the cluster state update tasks that are published (ClusterStateUpdateTask) from those that are not (LocalClusterUpdateTask), serving as a basis for future refactorings to separate the publishing mechanism out of ClusterService.
We have to sort the logger names so they wouldn't override each other. Processing org.elasticsearch:DEBUG after org.elasticsearch.transport:TRACE resets the setting of the later
This change is the first towards providing the ability to store
sensitive settings in elasticsearch. It adds the
`elasticsearch-keystore` tool, which allows managing a java keystore.
The keystore is loaded upon node startup in Elasticsearch, and used by
the Setting infrastructure when a setting is configured as secure.
There are a lot of caveats to this PR. The most important is it only
provides the tool and setting infrastructure for secure strings. It does
not yet provide for keystore passwords, keypairs, certificates, or even
convert any existing string settings to secure string settings. Those
will all come in follow up PRs. But this PR was already too big, so this
at least gets a basic version of the infrastructure in.
The two main things to look at. The first is the `SecureSetting` class,
which extends `Setting`, but removes the assumption for the raw value of the
setting to be a string. SecureSetting provides, for now, a single
helper, `stringSetting()` to create a SecureSetting which will return a
SecureString (which is like String, but is closeable, so that the
underlying character array can be cleared). The second is the
`KeyStoreWrapper` class, which wraps the java `KeyStore` to provide a
simpler api (we do not need the entire keystore api) and also extend
the serialized format to add metadata needed for loading the keystore
with no assumptions about keystore type (so that we can change this in
the future) as well as whether the keystore has a password (so that we
can know whether prompting is necessary when we add support for keystore
passwords).
* Remove a checked exception, replacing it with `ParsingException`.
* Remove all Parser classes for the yaml sections, replacing them with static methods.
* Remove `ClientYamlTestFragmentParser`. Isn't used any more.
* Remove `ClientYamlTestSuiteParseContext`, replacing it with some static utility methods.
I did not rewrite the parsers using `ObjectParser` because I don't think it is worth it right now.
This adds test classes that can be used to test the wire serialisation and (optionally) the XContent serialisation of objects that implement Streamable/Writeable and ToXContent.
These test classes will enable classes sich as InternalAggregation (or at least its implementations) to be tested in a consistent way when is comes to testing serialisation.
As the translog evolves towards a full operations log as part of the
sequence numbers push, there is a need for the translog to be able to
represent operations for which a sequence number was assigned, but the
operation did not mutate the index. Examples of how this can arise are
operations that fail after the sequence number is assigned, and gaps in
this history that arise when an operation is assigned a sequence number
but the operation never completed (e.g., a node crash). It is important
that these operations appear in the history so that they can be
replicated and replayed during recovery as otherwise the history will be
incomplete and local checkpoints will not be able to advance. This
commit introduces a no-op to the translog to set the stage for these
efforts.
Relates #22291
The `UnicastZenPing` shows it's age and is the result of many small changes. The current state of affairs is confusing and is hard to reason about. This PR cleans it up (while following the same original intentions). Highlights of the changes are:
1) Clear 3 round flow - no interleaving of scheduling.
2) The previous implementation did a best effort attempt to wait for ongoing pings to be sent and completed. The pings were guaranteed to complete because each used the total ping duration as a timeout. This did make it hard to reason about the total ping duration and the flow of the code. All of this is removed now and ping should just complete within the given duration or not be counted (note that it was very handy for testing, but I move the needed sync logic to the test).
3) Because of (2) the pinging scheduling changed a bit, to give a chance for the last round to complete. We now ping at the beginning, 1/3 and 2/3 of the duration.
4) To offset for (3) a bit, incoming ping requests are now added to on going ping collections.
5) UnicastZenPing never establishes full blown connections (but does reuse them if there). Relates to #22120
6) Discovery host providers are only used once per pinging round. Closes#21739
7) Usage of the ability to open a connection without connecting to a node ( #22194 ) and shorter connection timeouts helps with connections piling up. Closes#19370
8) Beefed up testing and sped them up.
9) removed light profile from production code
If we conditionally do random things, e.g. initialize a node only after the first test, we have to make sure that we unconditionally create a new seed calling random.nextLong(), then initialize the node under a private randomness context. This makes sure that any random usage through Randomness.get() will retrieve the proper random instance through RandomizedContext.current().getRandom(). When running under private randomness, the context will return the Random instance that was created with the provided seed (forked from the main random instance) rather than the main Random that's exposed to tests as well. Otherwise tests become non repeatable because that initialization part happens only before the first executed test.
Introduces `XContentParser#namedObject which works a little like
`StreamInput#readNamedWriteable`: on startup components register
parsers under names and a superclass. At runtime we look up the
parser and call it to parse the object.
Right now the parsers take a context object they use to help with
the parsing but I hope to be able to eliminate the need for this
context as most what it is used for at this point is to move
around parser registries which should be replaced by this method
eventually. I make no effort to do so in this PR because it is
big enough already. This is meant to the a start down a road that
allows us to remove classes like `QueryParseContext`,
`AggregatorParsers`, `IndicesQueriesRegistry`, and
`ParseFieldRegistry`.
The goal here is to reduce the amount of plumbing required to
allow parsing pluggable things. With this you don't have to pass
registries all over the place. Instead you must pass a super
registry to fewer places and use it to wrap the reader. This is
the same tradeoff that we use for NamedWriteable and it allows
much, much simpler binary serialization. We think we want that
same thing for xcontent serialization.
The only parsing actually converted to this method is parsing
`ScoreFunctions` inside of `FunctionScoreQuery`. I chose this
because it is relatively self contained.
* Internal: Refactor SettingCommand into EnvironmentAwareCommand
This change renames and changes the behavior of SettingCommand to have
its primary method take in a fully initialized Environment for
elasticsearch instead of just a map of settings. All of the subclasses
of SettingCommand already did this at some point, so this just removes
duplication.
We are currenlty checking that no deprecation warnings are emitted in our query tests. That can be moved to ESTestCase (disabled in ESIntegTestCase) as it allows us to easily catch where our tests use deprecated features and assert on the expected warnings.
We return deprecation warnings as response headers, besides logging them. Strict parsing mode stayed around, but was only used in query tests, though we also introduced checks for deprecation warnings there that don't need strict parsing anymore (see #20993).
We can then safely remove support for strict parsing mode. The final goal is to remove the ParseFieldMatcher class, but there are many many users of it. This commit prepares the field for the removal, by deprecating ParseFieldMatcher and making it effectively not needed. Strict parsing is removed from ParseFieldMatcher, and strict parsing is replaced in tests where needed with deprecation warnings checks.
Note that the setting to enable strict parsing was never ported to the new settings infra hance it cannot be set in production. It is really only used in our own tests.
Relates to #19552
Rename the method to assertToXContentEquivalent to highlight that it's tailored to ToXContent comparisons.
Rather than parsing into a map and replacing byte[] in both those maps, add custom equality assertions that recursively walk maps and lists and call Arrays.equals whenever a byte[] is encountered.
Moved field values `toXContent` logic to `GetField` (from `GetResult`), which outputs its own fields, and can also parse them now. Also added `fromXContent` to `GetResult` and `GetResponse`.
The start object and end object for `GetResponse` output have been moved to `GetResult#toXContent`, from the corresponding rest action. This makes it possible to have `toXContent` and `fromXContent` completely symmetric, as parsing requires looping till an end object is found which is weird when the corresponding `toXContent` doesn't print that out.
This also introduces the foundation for testing retrieval of _source and stored field values.
This commit makes mapping updates atomic when multiple types in an index are updated. Mappings for an index are now applied in a single atomic operation, which also allows to optimize some of the cross-type updates and checks.
Sequence BWC logic consists of two elements:
1) Wire level BWC using stream versions.
2) A changed to the global checkpoint maintenance semantics.
For the sequence number infra to work with a mixed version clusters, we have to consider situation where the primary is on an old node and replicas are on new ones (i.e., the replicas will receive operations without seq#) and also the reverse (i.e., the primary sends operations to a replica but the replica can't process the seq# and respond with local checkpoint). An new primary with an old replica is a rare because we do not allow a replica to recover from a new primary. However, it can occur if the old primary failed and a new replica was promoted or during primary relocation where the source primary is treated as a replica until the master starts the target.
1) Old Primary & New Replica - this case is easy as is taken care of by the wire level BWC. All incoming requests will have their seq# set to `UNASSIGNED_SEQ_NO`, which doesn't confuse the local checkpoint logic (keeping it at `NO_OPS_PERFORMED`)
2) New Primary & Old replica - this one is trickier as the global checkpoint service currently takes all in sync replicas into consideration for the global checkpoint calculation. In order to deal with old replicas, we change the semantics to say all *new node* in sync replicas. That means the replicas on old nodes don't count for the global checkpointing. In this state the seq# infra is not fully operational (you can't search on it, because copies may miss it) but it is maintained on shards that can support it. The old replicas will have to go through a file based recovery at some point and will get the seq# information at that point. There is still an edge case where a new primary fails and an old replica takes over. I'lll discuss this one with @ywelsch as I prefer to avoid it completely.
This PR also re-enables the BWC tests which were disabled. As such it had to fix any BWC issue that had crept in. Most notably an issue with the removal of the `timestamp` field in #21670.
The commit also includes a fix for the default value of the seq number field in replicated write requests (it was 0 but should be -2), that surface some other minor bugs which are fixed as well.
Last - I added some debugging tools like more sane node names and forcing replication request to implement a `toString`
With this commit we enable the Jackson feature 'STRICT_DUPLICATE_DETECTION'
by default for all XContent types (not only JSON).
We have also changed the name of the system property to disable this feature
from `es.json.strict_duplicate_detection` to the now more appropriate name
`es.xcontent.strict_duplicate_detection`.
Relates elastic/elasticsearch#19614
Relates elastic/elasticsearch#22073
In #22094 we introduce a test-only setting to simulate transport
impls that don't support handshakes. This commit implements the same logic
without a setting.
This commit touches addresses issues related to recovery and sequence numbers:
- A sequence number can be assigned and a Lucene commit created with a
maximum sequence number at least as large as that sequence number,
yet the operation corresponding to that sequence number can be
missing from both the Lucene commit and the translog. This means that
upon recovery the local checkpoint will be stuck at or below this
missing sequence number. To address this, we force the local
checkpoint to the maximum sequence number in the Lucene commit when
opening the engine. Note that there can still be gaps in the history
in the translog but we do not address those here.
- The global checkpoint is transferred to the target shard at the end
of peer recovery.
- Additionally, we reenable the relocation integration tests.
Lastly, this work uncovered some bugs in the assignment of sequence
numbers on replica operations:
- setting the sequence number on replica write requests was missing,
very likely introduced as a result of resolving merge conflicts
- handling operations that arrive out of order on a replica and have a
version conflict with a previous operation were never marked as
processed
Relates #22212
Some expert users like UnicastZenPing today establishes real connections to nodes during it's ping
phase that can be used by other parts of the system. Yet, this is potentially dangerous
and undesirable unless the nodes have been fully verified and should be connected to in the
case of a cluster state update or if we join a newly elected master. For use-cases like this, this change adds the infrastructure to manually handle connections that are not publicly available on the node ie. should not be managed by `Transport`/`TransportSerivce`
Sends the `error_trace` parameter with all requests sent by the
yaml test framework, including the doc snippet tests. This can be
overridden by settings `error_trace: false`. While this drift's
core's handling of the yaml tests from the client's slightly this
should only be a problem for tests that rely on the default value,
both of which I've fixed by setting the value explicitly.
This also escapes `\n` and `\t` in the `Stash dump on failure` so
the `stack_trace` is more readable.
Also fixes `RestUpdateSettingsAction` to not think of the `error_trace`
parameter as a setting.
`ClusterService` is responsible of updating the cluster state on every node (as a response to an API call on the master and when non-masters receive a new state from the master). When a new cluster state is processed, it is made visible via the `ClusterService#state` method and is sent to series of listeners. Those listeners come in two flavours - one is to change the state of the node in response to the new cluster state (call these cluster state appliers), the other is to start a secondary process. Examples for the later include an indexing operation waiting for a shard to be started or a master node action waiting for a master to be elected.
The fact that we expose the state before applying it means that samplers of the cluster state had to worry about two things - working based on a stale CS and working based on a future, i.e., "being applied" CS. The `ClusterStateStatus` was used to allow distinguishing between the two. Working with a stale cluster state is not avoidable. How this PR changes things to make sure consumers don't need to worry about future CS, removing the need for the status and simplifying the waiting logic.
This change does come with a price as "cluster state appliers" can't sample the cluster state from `ClusterService` whenever they want as the cluster state isn't exposed yet. However, recent clean ups made this is situation easier and this PR takes the last steps to remove such sampling. This also helps clarify the "information flow" and helps component separation (and thus potential unit testing). It also adds an assertion that will trigger if the cluster state is sampled by such listeners.
Note that there are still many "appliers" that could be made a simpler, unrestricted "listener" but this can be done in smaller bits in the future. The commit also makes it clear what the `appliers` and what the `listeners` are by using dedicated interfaces.
Also, since I had to change the listener types I went ahead and changed the data structure for temporary/timeout listeners (used for the observer) so addition and removal won't be an O(n) operation.
Since #22094 has been back-ported to 5.2 we can remove all BWC layers from master since all supported version will handle handshake requests.
Relates to #22094
Today sending a message on a closed channel doesn't throw an exception. The channel
might just swallow the exception and informs the internal async exception handler
that a channel got disconnected. This change adds a safety check that we fail
the handshake if we registered a handler but the channel has been closed already
for instance due to a reset by peer.
Low level handshake code doesn't handle situations gracefully if the connection
is concurrently closed or reset by peer. This commit adds the relevant code to
fail the handshake if the connection is closed.
In order to start clusters with min master nodes set without setting `discovery.initial_state_timeout`, #21846 has changed the way we start nodes. Instead to the previous serial start up, we now always start the nodes in an async fashion (internally). This means that starting a cluster is unsafe without `min_master_nodes` being set. We should therefore make it mandatory.
Boaz Leskes
Yannick Welsch
Nik Everett
Ryan Ernst
Tim B
javanna
Simon Willnauer
Adrien Grand
Simon Willnauer
Daniel Mitterdorfer
Igor Motov
Jason Tedor
Colin Goodheart-Smithe
Luca Cavanna
Daniel Mitterdorfer
Masaru Hasegawa