Today we rely on background syncs to relay the global checkpoint under
the mandate of the primary to its replicas. This means that the global
checkpoint on a replica can lag far behind the primary. The commit moves
to inlining global checkpoints with replication requests. When a
replication operation is performed, the primary will send the latest
global checkpoint inline with the replica requests. This keeps the
replicas closer in-sync with the primary.
However, consider a replication request that is not followed by another
replication request for an indefinite period of time. When the replicas
respond to the primary with their local checkpoint, the primary will
advance its global checkpoint. During this indefinite period of time,
the replicas will not be notified of the advanced global
checkpoint. This necessitates a need for another sync. To achieve this,
we perform a global checkpoint sync when a shard falls idle.
Relates #24513
* Replicate write failures
Currently, when a primary write operation fails after generating
a sequence number, the failure is not communicated to the replicas.
Ideally, every operation which generates a sequence number on primary
should be recorded in all replicas.
In this change, a sequence number is associated with write operation
failure. When a failure with an assinged seqence number arrives at a
replica, the failure cause and sequence number is recorded in the translog
and the sequence number is marked as completed via executing `Engine.noOp`
on the replica engine.
* use zlong to serialize seq_no
* Incorporate feedback
* track write failures in translog as a noop in primary
* Add tests for replicating write failures.
Test that document failure (w/ seq no generated) are recorded
as no-op in the translog for primary and replica shards
* Update to master
* update shouldExecuteOnReplica comment
* rename indexshard noop to markSeqNoAsNoOp
* remove redundant conditional
* Consolidate possible replica action for bulk item request
depanding on it's primary execution
* remove bulk shard result abstraction
* fix failure handling logic for bwc
* add more tests
* minor fix
* cleanup
* incorporate feedback
* incorporate feedback
* add assert to remove handling noop primary response when 5.0 nodes are not supported
This change simplifies how the rest test runner finds test files and
removes all leniency. Previously multiple prefixes and suffixes would
be tried, and tests could exist inside or outside of the classpath,
although outside of the classpath never quite worked. Now only classpath
tests are supported, and only one resource prefix is supported,
`/rest-api-spec/tests`.
closes#20240
We currently have the last minor version of the previous major hardcoded
in tests like rolling upgrade. This change programatically finds this
during gradle initialization by parsing versions from Version.java.
The current rest backcompat tests, which run against a mixed cluster of
5.x and 6.0 nodes, depend on snapshot builds of 5.x. However, this has
the potential for inconsistency that results in CI failures, and happens
quite often, whenever some backcompat logic is added to 5.x, but the bwc
test on master fails because the 5.x code has not yet been published as
a snapshot.
This change creates a git clone of the 5.x branch,
builds the zip distribution, and ties that into gradle substitutions for
the 5.x version.
Gradle's finalizedBy on tasks only ensures one task runs after another,
but not immediately after. This is problematic for our integration tests
since it allows multiple project's integ test clusters to be
simultaneously. While this has not been a problem thus far (gradle 2.13
happened to keep the finalizedBy tasks close enough that no clusters
were running in parallel), with gradle 3.3 the task graph generation has
changed, and numerous clusters may be running simultaneously, causing
memory pressure, and thus generally slower tests, or even failure if the
system has a limited amount of memory (eg in a vagrant host).
This commit reworks how integ tests are configured. It adds an
`integTestCluster` extension to gradle which is equivalent to the current
`integTest.cluster` and moves the rest test runner task to
`integTestRunner`. The `integTest` task is then just a dummy task,
which depends on the cluster runner task, as well as the cluster stop
task. This means running `integTest` in one project will both run the
rest tests, and shut down the cluster, before running `integTest` in
another project.
* Make document write requests immutable
Previously, write requests were mutated at the
transport level to update request version, version type
and sequence no before replication.
Now that all write requests go through the shard bulk
transport action, we can use the primary response stored
in item level bulk requests to pass the updated version,
seqence no. to replicas.
* incorporate feedback
* minor cleanup
* Add bwc test to ensure correct index version propagates to replica
* Fix bwc for propagating write operation versions
* Add assertion on replica request version type
* fix tests using internal version type for replica op
* Fix assertions to assert version type in replica and recovery
* add bwc tests for version checks in concurrent indexing
* incorporate feedback
This commit enforces the requirement of Content-Type for the REST layer and removes the deprecated methods in transport
requests and their usages.
While doing this, it turns out that there are many places where *Entity classes are used from the apache http client
libraries and many of these usages did not specify the content type. The methods that do not specify a content type
explicitly have been added to forbidden apis to prevent more of these from entering our code base.
Relates #19388
When a primary is relocated from an old node to a new node, it can have
ops in its translog that do not have a sequence number assigned. When a
file-based recovery is started, this can lead to skipping these ops when
replaying the translog due to a bug in the recovery logic. This commit
addresses this bug and adds a test in the BWC tests.
Relates #22945
This change adds a strict mode for xcontent parsing on the rest layer. The strict mode will be off by default for 5.x and in a separate commit will be enabled by default for 6.0. The strict mode, which can be enabled by setting `http.content_type.required: true` in 5.x, will require that all incoming rest requests have a valid and supported content type header before the request is dispatched. In the non-strict mode, the Content-Type header will be inspected and if it is not present or not valid, we will continue with auto detection of content like we have done previously.
The content type header is parsed to the matching XContentType value with the only exception being for plain text requests. This value is then passed on with the content bytes so that we can reduce the number of places where we need to auto-detect the content type.
As part of this, many transport requests and builders were updated to provide methods that
accepted the XContentType along with the bytes and the methods that would rely on auto-detection have been deprecated.
In the non-strict mode, deprecation warnings are issued whenever a request with body doesn't provide the Content-Type header.
See #19388
This commit introduces sequence-number-based recovery. When a replica
has fallen out of sync, rather than performing a file-based recovery we
first attempt to replay operations since the last local checkpoint on
the replica. To do this, at the start of recovery the replica tells the
primary what its local checkpoint is. The primary will then wait for all
operations between that local checkpoint and the current maximum
sequence number to complete; this is to ensure that there are no gaps in
the operations that will be replayed from the primary to the
replica. This is a best-effort attempt as we currently have no
guarantees on the primary that these operations will be available; if we
are not able to replay all operations in the desired range, we just
fallback to file-based recovery. Later work will strengthen the
guarantees.
Relates #22484
`ToXContentObject` extends `ToXContent` without adding new methods to it, while allowing to mark classes that output complete xcontent objects to distinguish them from classes that require starting and ending an anonymous object externally.
Ideally ToXContent would be renamed to ToXContentFragment, but that would be a huge change in our codebase, hence we simply document the fact that toXContent outputs fragments with no guarantees that the output is valid per se without an external ancestor.
Relates to #16347
* 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.
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`
There is not yet a BWC layer in sequence numbers. This commit sets the
BWC version to 6.0.0 for the BWC and rolling upgrade tests until this
BWC layer is built.
This commit enables real BWC testing against a 5.1 snapshot. All
REST tests plus rolling upgrade test now run against a mixed version
cross major version cluster.
At one point in the past when moving out the rest tests from core to
their own subproject, we had multiple test classes which evenly split up
the tests to run. However, we simplified this and went back to a single
test runner to have better reproduceability in tests. This change
removes the remnants of that multiplexing support.
This makes it obvious that these tests are for running the client yaml
suites. Now that there are other ways of running tests using the REST
client against a running cluster we can't go on calling the shared
client yaml tests "REST tests". They are rest tests, but they aren't
**the** rest tests.
This adds a header that looks like `Location: /test/test/1` to the
response for the index/create/update API. The requirement for the header
comes from https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.htmlhttps://tools.ietf.org/html/rfc7231#section-7.1.2 claims that relative
URIs are OK. So we use an absolute path which should resolve to the
appropriate location.
Closes#19079
This makes large changes to our rest test infrastructure, allowing us
to write junit tests that test a running cluster via the rest client.
It does this by splitting ESRestTestCase into two classes:
* ESRestTestCase is the superclass of all tests that use the rest client
to interact with a running cluster.
* ESClientYamlSuiteTestCase is the superclass of all tests that use the
rest client to run the yaml tests. These tests are shared across all
official clients, thus the `ClientYamlSuite` part of the name.
Change version, required a minor fix in the RPM building.
In case of a alpha/beta version, the release will contain alpha/beta
as the RPM version cannot contains dashes/tildes.
This change adds the infrastructure to run the rest tests on a multi-node
cluster that users 2 different minor versions of elasticsearch. It doesn't implement
any dedicated BWC tests but rather leverages the existing REST tests.
Since we don't have a real version to test against, the tests uses the current version
until the first minor / RC is released to ensure the infrastructure works.
Relates to #14406Closes#17072
Jason Tedor
Areek Zillur
Ryan Ernst
javanna
Jay Modi
jaymode
Boaz Leskes
Ali Beyad
Nik Everett
Simon Willnauer
Clinton Gormley
Adrien Grand
Alexander Reelsen