Restoring a shard from snapshot throws the primary back in time violating assumptions and bringing the validity of global checkpoints in question. To avoid problems, we should make sure that a shard that was restored will never be the source of an ops based recovery to a shard that existed before the restore. To this end we have introduced the notion of `histroy_uuid` in #26577 and required that both source and target will have the same history to allow ops based recoveries. This PR make sure that a shard gets a new uuid after restore.
As suggested by @ywelsch , I derived the creation of a `history_uuid` from the `RecoverySource` of the shard. Store recovery will only generate a uuid if it doesn't already exist (we can make this stricter when we don't need to deal with 5.x indices). Peer recovery follows the same logic (note that this is different than the approach in #26557, I went this way as it means that shards always have a history uuid after being recovered on a 6.x node and will also mean that a rolling restart is enough for old indices to step over to the new seq no model). Local shards and snapshot force the generation of a new translog uuid.
Relates #10708Closes#26544
This commit updates the version for master to 7.0.0-alpha1. It also adds
the 6.1 version constant, and fixes many tests, as well as marking some
as awaits fix.
Closes#25893Closes#25870
This commit removes a rolling upgrade test for scripting that is totally
busted yet is preventing builds from succeeding. We elect to remove this
test as opposed to skipping the test as:
- it has beeen being skipped for months with no apparent loss
- it appears to need significant work to get to an unbusted state
In the rolling upgrade tests, there is a test to create an index with
replica shards and ensure that in the mixed cluster environment, the
cluster health is green before any other tests are executed. However,
there were two problems with this. First, if the replica shard was
residing on the restarted node, then delayed allocation will kick in and
cause the cluster health request to timeout after 1m. The fix to this
was to drastically lower the delayed allocation setting. Second, if the
primary exists on the higher version node, then the replica cannot be
assigned to the lower version node because recovery cannot happen from
lower lucene versions. The fix here was to wait for the cluster health
to be yellow instead of green in the mixed cluster environment. In the
fully upgraded cluster, the cluster health check waits for a green
cluster as before.
Closes#25185
In 6.x we prevent multiple types and default to `index.mapping.single_type: false`
This change removes the registered setting and ensures that it's preserved for
5.x indices.
Relates to #24961
This commit adds a gradle project, set inside the root build.gradle,
which controls all our bwc tests. This allows for seamless (ie no errant
CI failures) backporting of behavior.
In #25201, a setting was added to allow setting the retry timeout for the rest client under the
impression that this would allow requests to go longer than 30s. However, there is also a socket
timeout that needs to be set to greater than 30s, which this change adds a setting for.
This commit adds a setting to change the request timeout for the rest client. This is useful as the
default timeout is 30s, which is also the same default for calls like cluster health. If both are
the same then the response from the cluster health api will not be received as the client usually
times out first making test failures harder to debug.
Relates #25185
This commit adds back "id" as the key within a script to specify a
stored script (which with file scripts now gone is no longer ambiguous).
It also adds "source" as a replacement for "code". This is in an attempt
to normalize how scripts are specified across both put stored scripts and script usages, including search template requests. This also deprecates the old inline/stored keys.
We default to 0 replicas in the rolling restart scenario already to ensure
we test against worst case. Yet, this adds a dummy index to ensure we also
recover and index with replicas just fine.
These tests spin up two nodes of an older version of Elasticsearch,
create some stuff, shut down the nodes, start the current version,
and verify that the created stuff works.
You can run `gradle qa:full-cluster-restart:check` to run these
tests against the head of the previous branch of Elasticsearch
(5.x for master, 5.4 for 5.x, etc) or you can run
`gradle qa:full-cluster-restart:bwcTest` to run this test against
all "index compatible" versions, one after the other. For master
this is every released version in the 5.x.y version *and* the tip
of the 5.x branch.
I'd love to add more to these tests in the future but these
currently just cover the functionality of the `create_bwc_index.py`
script and start to cover the assertions in the
`OldIndexBackwardsCompatibilityIT` test.
This commit changes the rolling upgrade test to create a set of rest
test tasks per wire compat version. The most recent wire compat version
is always tested with the `integTest` task, and all versions can be
tested with `bwcTest`.
This commit expands the logic for version extraction from Version.java
to include a list of all versions for backcompat purposes. The tests
using bwcVersion are converted to use this list, but those tests
(rolling upgrade and backwards-5.0) are still not randomized; that will
happen in another followup.
This commit renames all rest test files to use the .yml extension
instead of .yaml. This way the extension used within all of
elasticsearch for yaml is consistent.
In #24251 we fix an issue with stored search templates that
this test would have discovered: stored search templates cause
the node to refuse to start. Technically a "restart" test would
have caught this as well and would have caught it more quickly.
But we already *have* an upgrade test and we don't have restart tests.
And testing this on upgrade is a good thing too.
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
After splitting integ tests into cluster configuration and the test
runner task, we still have dependencies of the test runner added as deps
of the cluster. This commit adds dependencies directly to the cluster,
so that the runner can have other dependencies independent of what is
needed for the cluster.
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.
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.
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
Currently, stored scripts use a namespace of (lang, id) to be put, get, deleted, and executed. This is not necessary since the lang is stored with the stored script. A user should only have to specify an id to use a stored script. This change makes that possible while keeping backwards compatibility with the previous namespace of (lang, id). Anywhere the previous namespace is used will log deprecation warnings.
The new behavior is the following:
When a user specifies a stored script, that script will be stored under both the new namespace and old namespace.
Take for example script 'A' with lang 'L0' and data 'D0'. If we add script 'A' to the empty set, the scripts map will be ["A" -- D0, "A#L0" -- D0]. If a script 'A' with lang 'L1' and data 'D1' is then added, the scripts map will be ["A" -- D1, "A#L1" -- D1, "A#L0" -- D0].
When a user deletes a stored script, that script will be deleted from both the new namespace (if it exists) and the old namespace.
Take for example a scripts map with {"A" -- D1, "A#L1" -- D1, "A#L0" -- D0}. If a script is removed specified by an id 'A' and lang null then the scripts map will be {"A#L0" -- D0}. To remove the final script, the deprecated namespace must be used, so an id 'A' and lang 'L0' would need to be specified.
When a user gets/executes a stored script, if the new namespace is used then the script will be retrieved/executed using only 'id', and if the old namespace is used then the script will be retrieved/executed using 'id' and 'lang'
This is related to #22116. URLRepository requires SocketPermission
connect. This commit introduces a new module called "repository-url"
where URLRepository will reside. With the new module, permissions can
be removed from core.
* Fix Translog.Delete serialization for sequence numbers
Translog.Delete used `.writeVLong` instead of `.writeLong` for the sequence
number and primary term (and their respective "read" variants). This could lead
to issues where a 5.x node sent a translog operation with a negative sequence
number (-2 for unassigned seq no) that tripped an assertion serializing a
negative number and causing ES to exit.
Adds a unit test for serialization and a mixed-cluster REST test, since that was
how this was originally caught.
* Use more realistic values for random seqNum and primary term
* Add comment with TODO for removal in 7.0
* Change comment into an assert
* 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.
automatically between tasks, as we want some of the nodes from
the previous task to continue running in the next task. This
commit enables a cluster configuration setting to not stop
nodes automatically after a task runs, but instead the creator
of the test task must stop the running nodes explicitly in a
cleanup phase.
cluster, we wait for the cluster health to indicate the
necessary nodes have formed a cluster. This check was an
exact value (equality) check. However, if we are trying to
connect the nodes in the cluster to nodes from a previously
formed cluster (of the same name), then we will have more
nodes returned by the cluster health check than the current
task's configured number of nodes. Hence, this check needs
to be a >= check. This commit fixes it.