This commit avoids a situation where we might stack overflow in the
auto-follower coordinator. In the face of repeated failures to get the
remote cluster state, we would previously be called back on the same
thread and then recurse to try again. If this failure persists, the
repeated callbacks on the same thread would lead to a stack
overflow. The repeated failures can occur, for example, if the connect
queue is full when we attempt to make a connection to the remote
cluster. This commit avoids this by truncating the call stack if we are
called back on the same thread as the initial request was made on.
This commit avoids an NPE when checking for privileges to follow
indices. The problem here is that in some cases we might not be able to
read the authentication info from the thread context. In that case, a
null user would be returned and we were not guarding against this.
Test clusters currently has its own set of logic for dealing with
finding different versions of Elasticsearch, downloading them, and
extracting them. This commit converts testclusters to use the
DistributionDownloadPlugin.
This commit creates new base classes for master node actions whose
response types still implement Streamable. This simplifies both finding
remaining classes to convert, as well as creating new master node
actions that use Writeable for their responses.
relates #34389
* The incompatible snapshots logic was created to track 1.x snapshots that
became incompatible with 2.x
* It serves no purpose at this point
* It adds an additional GET request to every loading of
RepositoryData (from loading the incompatible snapshots blob)
The base classes for transport requests and responses currently
implement Streamable and Writeable. The writeTo method on these base
classes is implemented with an empty implementation. Not only does this
complicate subclasses to think they need to call super.writeTo, but it
also can lead to not implementing writeTo when it should have been
implemented, or extendiong one of these classes when not necessary,
since there is nothing to actually implement.
This commit removes the empty writeTo from these base classes, and fixes
subclasses to not call super and in some cases implement an empty
writeTo themselves.
relates #34389
This PR enables the indexing optimization using sequence numbers on
replicas. With this optimization, indexing on replicas should be faster
and use less memory as it can forgo the version lookup when possible.
This change also deactivates the append-only optimization on replicas.
Relates #34099
Action is a class that encapsulates meta information about an action
that allows it to be called remotely, specifically the action name and
response type. With recent refactoring, the action class can now be
constructed as a static constant, instead of needing to create a
subclass. This makes the old pattern of creating a singleton INSTANCE
both misnamed and lacking a common placement.
This commit renames Action to ActionType, thus allowing the old INSTANCE
naming pattern to be TYPE on the transport action itself. ActionType
also conveys that this class is also not the action itself, although
this change does not rename any concrete classes as those will be
removed organically as they are converted to TYPE constants.
relates #34389
The Action base class currently works for both Streamable and Writeable
response types. This commit intorduces StreamableResponseAction, for
which only the legacy Action implementions which provide newResponse()
will extend. This eliminates the need for overriding newResponse() with
an UnsupportedOperationException.
relates #34389
This commit replaces usages of Streamable with Writeable for the
AcknowledgedResponse and its subclasses, plus associated actions.
Note that where possible response fields were made final and default
constructors were removed.
This is a large PR, but the change is mostly mechanical.
Relates to #34389
Backport of #43414
Local and global checkpoints currently do not correctly reflect what's persisted to disk. The issue is
that the local checkpoint is adapted as soon as an operation is processed (but not fsynced yet). This
leaves room for the history below the global checkpoint to still change in case of a crash. As we rely
on global checkpoints for CCR as well as operation-based recoveries, this has the risk of shard
copies / follower clusters going out of sync.
This commit required changing some core classes in the system:
- The LocalCheckpointTracker keeps track now not only of the information whether an operation has
been processed, but also whether that operation has been persisted to disk.
- TranslogWriter now keeps track of the sequence numbers that have not been fsynced yet. Once
they are fsynced, TranslogWriter notifies LocalCheckpointTracker of this.
- ReplicationTracker now keeps track of the persisted local and persisted global checkpoints of all
shard copies when in primary mode. The computed global checkpoint (which represents the
minimum of all persisted local checkpoints of all in-sync shard copies), which was previously stored
in the checkpoint entry for the local shard copy, has been moved to an extra field.
- The periodic global checkpoint sync now also takes async durability into account, where the local
checkpoints on shards only advance when the translog is asynchronously fsynced. This means that
the previous condition to detect inactivity (max sequence number is equal to global checkpoint) is
not sufficient anymore.
- The new index closing API does not work when combined with async durability. The shard
verification step is now requires an additional pre-flight step to fsync the translog, so that the main
verify shard step has the most up-to-date global checkpoint at disposition.
This commit removes some very old test logging annotations that appeared
to be added to investigate test failures that are long since closed. If
these are needed, they can be added back on a case-by-case basis with a
comment associating them to a test failure.
Backport of: https://github.com/elastic/elasticsearch/pull/43222
This commit replaces usages of Streamable with Writeable for the
SingleShardRequest / TransportSingleShardAction classes and subclasses of
these classes.
Note that where possible response fields were made final and default
constructors were removed.
Relates to #34389
With this change, we will rebuild the live version map and local
checkpoint using documents (including soft-deleted) from the safe commit
when opening an internal engine. This allows us to safely prune away _id
of all soft-deleted documents as the version map is always in-sync with
the Lucene index.
Relates #40741
Supersedes #42979
In these tests, we sleep for a small multiple of the renew interval,
then check that the retention leases are not changed. If a renewal
request takes longer than that interval because of GC or slow CI, then
the retention leases are not the same as before sleep. With this change,
we relax to assert that we eventually stop the renewable process.
Closes#39509
The retention leases stats is null if the processing shard copy is being
closed. In this the case, we should check against null then retry to
avoid failing a test.
Closes#41237
This commit refactors put mapping request validation for reuse. The
concrete case that we are after here is the ability to apply effectively
the same framework to indices aliases requests. This commit refactors
the put mapping request validation framework to allow for that.
Adds a metadata field to snapshots which can be used to store arbitrary
key-value information. This may be useful for attaching a description of
why a snapshot was taken, tagging snapshots to make categorization
easier, or identifying the source of automatically-created snapshots.
This commit adds functionality so that aliases that are manipulated on
leader indices are replicated by the shard follow tasks to the follower
indices. Note that we ignore write indices. This is due to the fact that
follower indices do not receive direct writes so the concept is not
useful.
Relates #41815
testRetentionLeaseIsAddedIfItDisappearsWhileFollowing does not reset the
mock transport service after test. Surviving transport interceptors from
that test can sneaky remove retention leases and make other tests fail.
Closes#39331Closes#39509Closes#41428Closes#41679Closes#41737Closes#41756
We need more information to understand why CcrRetentionLeaseIT is
failing. This commit adds some debug log to retention leases and enables
them in CcrRetentionLeaseIT.
ShardId already implements Writeable so there is no need for it to implement Streamable too. Also the readShardId static method can be
easily replaced with direct usages of the constructor that takes a
StreamInput as argument.
* Remove IndexShard dependency from Repository
In order to simplify repository testing especially for BlobStoreRepository
it's important to remove the dependency on IndexShard and reduce it to
Store and MapperService (in the snapshot case). This significantly reduces
the dependcy footprint for Repository and allows unittesting without starting
nodes or instantiate entire shard instances. This change deprecates the old
method signatures and adds a unittest for FileRepository to show the advantage
of this change.
In addition, the unittesting surfaced a bug where the internal file names that
are private to the repository were used in the recovery stats instead of the
target file names which makes it impossible to relate to the actual lucene files
in the recovery stats.
* don't delegate deprecated methods
* apply comments
* test
Moves the test infrastructure away from using node.max_local_storage_nodes, allowing us in a
follow-up PR to deprecate this setting in 7.x and to remove it in 8.0.
This also changes the behavior of InternalTestCluster so that starting up nodes will not automatically
reuse data folders of previously stopped nodes. If this behavior is desired, it needs to be explicitly
done by passing the data path from the stopped node to the new node that is started.
Previously, TransportSingleShardAction required constructing a new
empty response object. This response object's Streamable readFrom
was used. As part of the migration to Writeable, the interface here
was updated to leverage Writeable.Reader.
relates to #34389.
These tests are failing somewhat mysteriously, indicating that when we
renew retention leaess during a restore that our retention leases that
we added before starting the restore suddenly do not exist. To make
sense of this, this commit enables trace logging.
The run task is supposed to run elasticsearch with the given plugin or
module. However, for modules, this is most realistic if using the full
distribution. This commit changes the run setup to use the default or
oss as appropriate.
With this change, we will verify the consistency of version and source
(besides id, seq_no, and term) of live documents between shard copies
at the end of disruption tests.
Today we choose to initialize max_seq_no_of_updates on primaries only so
we can deal with a situation where a primary is on an old node (before
6.5) which does not have MUS while replicas on new nodes (6.5+).
However, this strategy is quite complex and can lead to bugs (for
example #40249) since we have to assign a correct value (not too low) to
MSU in all possible situations (before recovering from translog,
restoring history on promotion, and handing off relocation).
Fortunately, we don't have to deal with this BWC in 7.0+ since all nodes
in the cluster should have MSU. This change simplifies the
initialization of MSU by always assigning it a correct value in the
constructor of Engine regardless of whether it's a replica or primary.
Relates #33842
Motivated by slow snapshot deletes reported in e.g. #39656 and the fact that these likely are a contributing factor to repositories accumulating stale files over time when deletes fail to finish in time and are interrupted before they can complete.
* Makes snapshot deletion async and parallelizes some steps of the delete process that can be safely run concurrently via the snapshot thread poll
* I did not take the biggest potential speedup step here and parallelize the shard file deletion because that's probably better handled by moving to bulk deletes where possible (and can still be parallelized via the snapshot pool where it isn't). Also, I wanted to keep the size of the PR manageable.
* See https://github.com/elastic/elasticsearch/pull/39656#issuecomment-470492106
* Also, as a side effect this gives the `SnapshotResiliencyTests` a little more coverage for master failover scenarios (since parallel access to a blob store repository during deletes is now possible since a delete isn't a single task anymore).
* By adding a `ThreadPool` reference to the repository this also lays the groundwork to parallelizing shard snapshot uploads to improve the situation reported in #39657
* Introduce Delegating ActionListener Wrappers
* Dry up use cases of ActionListener that simply pass through the response or exception to another listener
It looks like these permissions were copy/pasted from another plugin yet
almost none of these permissions are needed for the CCR plugin. This
commit removes all these unneeded permissions from the CCR plugin.
This is a dependency of #39504
Motivation:
By refactoring `TransportShardBulkAction#shardOperationOnPrimary` to async, we enable using `DeterministicTaskQueue` based tests to run indexing operations. This was previously impossible since we were blocking on the `write` thread until the `update` thread finished the mapping update.
With this change, the mapping update will trigger a new task in the `write` queue instead.
This change significantly enhances the amount of coverage we get from `SnapshotResiliencyTests` (and other potential future tests) when it comes to tracking down concurrency issues with distributed state machines.
The logical change is effectively all in `TransportShardBulkAction`, the rest of the changes is then simply mechanically moving the caller code and tests to being async and passing the `ActionListener` down.
Since the move to async would've added more parameters to the `private static` steps in this logic, I decided to inline and dry up (between delete and update) the logic as much as I could instead of passing the listener + wait-consumer down through all of them.
* Replace usages RandomizedTestingTask with built-in Gradle Test (#40978)
This commit replaces the existing RandomizedTestingTask and supporting code with Gradle's built-in JUnit support via the Test task type. Additionally, the previous workaround to disable all tasks named "test" and create new unit testing tasks named "unitTest" has been removed such that the "test" task now runs unit tests as per the normal Gradle Java plugin conventions.
(cherry picked from commit 323f312bbc829a63056a79ebe45adced5099f6e6)
* Fix forking JVM runner
* Don't bump shadow plugin version
Many gradle projects specifically use the -try exclude flag, because
there are many cases where auto-closeable resource ignore is never
referenced in body of corresponding try statement. Suppressing this
warning specifically in each case that it happens using
`@SuppressWarnings("try")` would be very verbose.
This change removes `-try` from any gradle project and adds it to the
build plugin. Also this change removes exclude flags from gradle projects
that is already specified in build plugin (for example -deprecation).
Relates to #40366
It is important that resync actions are not rejected on the primary even if its
`write` threadpool is overloaded. Today we do this by exposing
`registerRequestHandlers` to subclasses and overriding it in
`TransportResyncReplicationAction`. This isn't ideal because it obscures the
difference between this action and other replication actions, and also might
allow subclasses to try and use some state before they are properly
initialised. This change replaces this override with a constructor parameter to
solve these issues.
Relates #40706
Currently the TransportMessageListener is applied and used in the
Transport class. However, local requests and responses never make it to
this class. This PR moves the listener add/remove methods to the
TransportService. After this change the Transport can only have one
listener set with it. This one listener is the TransportService, which
will then propogate the events to the external listeners.
Additionally this commit back ports #40237
Remove Tracer from MockTransportService
Currently the TransportMessageListener is applied and used in the
Transport class. However, local requests and responses never make it to
this class. This PR moves the listener add/remove methods to the
TransportService. After this change the Transport can only have one
listener set with it. This one listener is the TransportService, which
will then propogate the events to the external listeners.
In some cases, a request to perform a retention lease action can arrive
on a primary shard before it is active. In this case, the primary shard
would not yet be in primary mode, tripping an assertion in the
replication tracker. Instead, we should not attempt to perform such
actions on an initializing shard. This commit addresses this by not
returning the primary shard in the single shard iterator if the primary
shard is not yet active.
If there's a failover on the follower, then its max_seq_no_of_updates is
bootstrapped from its max_seq_no which might be higher than the
max_seq_no_of_updates of the leader. We need to relax this check.
Relates #40249
This commit adjusts the frequency with which CCR renews retention leases
and with which primaries sync retention leases to replicas. This helps
Lucene reclaim soft-deleted documents more aggressively, which we have
found in some use-cases can help improve performance, and either way
will help keep disk space under more control.
Currently, we cannot update index setting index.translog.sync_interval if index is open, because it's
not dynamic which can be updated for closed index only.
Closes#32763
If a replica were first reset due to one primary failover and then
promoted (before resync completes), its MSU would not include changes
since global checkpoint, leading to errors during translog replay.
Fixed by re-initializing MSU before restoring local history.
This named writable was never registered, so it means that we could not
read auto-follow patterns that were registered in the cluster
state. This causes them to be lost on restarts, a bad bug. This commit
addresses this by registering this named writable, and we add a basic
CCR restart test to ensure that CCR keeps functioning properly when the
follower is restarted.
We introduced WAIT_CLUSTERSTATE action in #19287 (5.0), but then stopped
using it since #25692 (6.0). This change removes that action and related
code in 7.x and 8.0.
Relates #19287
Relates #25692
We were leaking a reference to an AutoFollowCoordinator during
construction, violating safe publication according to the JLS
specification. This commit addresses this by waiting to register
AutoFollowCoordinator with the ClusterApplierService after the
AutoFollowCoordinator is fully constructed. We also remove ourselves as
a listener when stopping.
When shutting down a node, auto-followers will keep trying to run. This
is happening even as transport services and other components are being
closed. In some cases, this can lead to a stack overflow as we rapidly
try to check the license state of the remote cluster, can not because
the transport service is shutdown, and then immeidately retry
again. This can happen faster than the shutdown, and we die with stack
overflow. This commit adds a stop command to auto-followers so that this
retry loop occurs at most once on shutdown.
When an auto-follower coordinator times out waiting for the remote
cluster state, we do not log any indication of this. While this is
expected behavior in quiet deployments, it is still useful to see this
information for tracing the behavior of the auto-follow
coordinator. This commit adds a trace log message indicating that the
timeout.
This commit removes the cluster state size field from the cluster state
response, and drops the backwards compatibility layer added in 6.7.0 to
continue to support this field. As calculation of this field was
expensive and had dubious value, we have elected to remove this field.
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
* reduce the number of leader indices to be auto followed
* also check the number of follower indices being created
* also check the whether leader indices are marked as auto followed
Relates to #36761
This test was more complicated than necessary, where we were capturing
requests to prevent removal of retention leases, so that our forget
follower request could remove the retention leases instead. Instead, a
pause is enough to ensure that the retention leases are not re-added
after we remove them by the forget follower request. This commit
simplifies this test, and should remove some spurious failures.
Relates #39850
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.
Today the `GroupedActionListener` accepts a `defaults` parameter but all
callers pass an empty list. Also it is permitted to pass an empty group but
this is trappy because the delegated listener is never be called in that case.
This commit removes the `defaults` parameter and forbids an empty group.
This commit renames the retention lease setting
index.soft_deletes.retention.lease so that it is under the namespace
index.soft_deletes.retention_lease. As such, we rename the setting to
index.soft_deletes.retention_lease.period.
This cleans up the Engine implementation by separating the sequence number generation from the
planning step in the engine, to avoid for the planning step to have any side effects. This makes it
easier to see that every sequence number is properly accounted for.
Backport support for replicating closed indices (#39499)
Before this change, closed indexes were simply not replicated. It was therefore
possible to close an index and then decommission a data node without knowing
that this data node contained shards of the closed index, potentially leading to
data loss. Shards of closed indices were not completely taken into account when
balancing the shards within the cluster, or automatically replicated through shard
copies, and they were not easily movable from node A to node B using APIs like
Cluster Reroute without being fully reopened and closed again.
This commit changes the logic executed when closing an index, so that its shards
are not just removed and forgotten but are instead reinitialized and reallocated on
data nodes using an engine implementation which does not allow searching or
indexing, which has a low memory overhead (compared with searchable/indexable
opened shards) and which allows shards to be recovered from peer or promoted
as primaries when needed.
This new closing logic is built on top of the new Close Index API introduced in
6.7.0 (#37359). Some pre-closing sanity checks are executed on the shards before
closing them, and closing an index on a 8.0 cluster will reinitialize the index shards
and therefore impact the cluster health.
Some APIs have been adapted to make them work with closed indices:
- Cluster Health API
- Cluster Reroute API
- Cluster Allocation Explain API
- Recovery API
- Cat Indices
- Cat Shards
- Cat Health
- Cat Recovery
This commit contains all the following changes (most recent first):
* c6c42a1 Adapt NoOpEngineTests after #39006
* 3f9993d Wait for shards to be active after closing indices (#38854)
* 5e7a428 Adapt the Cluster Health API to closed indices (#39364)
* 3e61939 Adapt CloseFollowerIndexIT for replicated closed indices (#38767)
* 71f5c34 Recover closed indices after a full cluster restart (#39249)
* 4db7fd9 Adapt the Recovery API for closed indices (#38421)
* 4fd1bb2 Adapt more tests suites to closed indices (#39186)
* 0519016 Add replica to primary promotion test for closed indices (#39110)
* b756f6c Test the Cluster Shard Allocation Explain API with closed indices (#38631)
* c484c66 Remove index routing table of closed indices in mixed versions clusters (#38955)
* 00f1828 Mute CloseFollowerIndexIT.testCloseAndReopenFollowerIndex()
* e845b0a Do not schedule Refresh/Translog/GlobalCheckpoint tasks for closed indices (#38329)
* cf9a015 Adapt testIndexCanChangeCustomDataPath for replicated closed indices (#38327)
* b9becdd Adapt testPendingTasks() for replicated closed indices (#38326)
* 02cc730 Allow shards of closed indices to be replicated as regular shards (#38024)
* e53a9be Fix compilation error in IndexShardIT after merge with master
* cae4155 Relax NoOpEngine constraints (#37413)
* 54d110b [RCI] Adapt NoOpEngine to latest FrozenEngine changes
* c63fd69 [RCI] Add NoOpEngine for closed indices (#33903)
Relates to #33888
Today when users upgrade to 7.0, existing indices will automatically
switch to soft-deletes without an opt-out option. With this change,
we only enable soft-deletes by default for new indices.
Relates #36141
This commit is the final piece of the integration of CCR with retention
leases. Namely, we periodically renew retention leases and advance the
retaining sequence number while following.
With this change, we won't wait for the local checkpoint to advance to
the max_seq_no before starting phase2 of peer-recovery. We also remove
the sequence number range check in peer-recovery. We can safely do these
thanks to Yannick's finding.
The replication group to be used is currently sampled after indexing
into the primary (see `ReplicationOperation` class). This means that
when initiating tracking of a new replica, we have to consider the
following two cases:
- There are operations for which the replication group has not been
sampled yet. As we initiated the new replica as tracking, we know that
those operations will be replicated to the new replica and follow the
typical replication group semantics (e.g. marked as stale when
unavailable).
- There are operations for which the replication group has already been
sampled. These operations will not be sent to the new replica. However,
we know that those operations are already indexed into Lucene and the
translog on the primary, as the sampling is happening after that. This
means that by taking a snapshot of Lucene or the translog, we will be
getting those ops as well. What we cannot guarantee anymore is that all
ops up to `endingSeqNo` are available in the snapshot (i.e. also see
comment in `RecoverySourceHandler` saying `We need to wait for all
operations up to the current max to complete, otherwise we can not
guarantee that all operations in the required range will be available
for replaying from the translog of the source.`). This is not needed,
though, as we can no longer guarantee that max seq no == local
checkpoint.
Relates #39000Closes#38949
Co-authored-by: Yannick Welsch <yannick@welsch.lu>
when dealing with TimeoutException
The `IndexFollowingIT#testDeleteLeaderIndex()`` test failed,
because a NPE was captured as fatal error instead of an IndexNotFoundException.
Closes#39308
Finally! This commit should fix the issues with the CCR retention lease
that has been plaguing build failures. The issue here is that we are
trying to prevent the clear session requests from being executed until
after we have been able to validate that retention leases are being
renewed. However, we were only blocking the clear session requests but
not blocking them when they are proxied through another node. This
commit addresses that.
Relates #39268
This commit changes the sort order of shard stats that are collected in
CCR retention lease integration tests. This change is done so that
primaries appear first in sort order.
This test fails rarely but it is flaky in its current form. The problem
here is that we lack a guarantee on the retention leases having been
synced to all shard copies. We need to sleep long enough to ensure that
that occurs, and then we can sample the retention leases, possibly sleep
again (we usually will not have too since the first sleep will have been
long enough to allow a sync and a renewal to happen, if one was going to
happen), and the sample the retention leases for comparison.
Closes#39331
The shard logged here is the leader shard but it should be the follower
shard since this background retention lease renewal is happening on the
follower side. This commit fixes that.
This commit simplifies the use of transport mocking in the CCR retention
lease integration tests. Instead of adding a send rule between nodes, we
add a default send rule. This greatly simplifies the code here, and
speeds the test up a little bit too.
This fixes#39245. Currently it is possible in this test that the clear
session call times-out. This means that the index commit will not be
released and there will be an assertion triggered in the test teardown.
This commit ensures that we wipe the leader index in the test to avoid
this assertion.
It is okay if the clear session call times-out in normal usage. This
scenario is unavoidable due to potential network issues. We have a local
timeout on the leader to clean it up when this scenario happens.
Currently remote compression and ping schedule settings are dynamic.
However, we do not listen for changes. This commit adds listeners for
changes to those two settings. Additionally, when those settings change
we now close existing connections and open new ones with the settings
applied.
Fixes#37201.
Sleeps in tests smell funny, and we try to avoid them to the extent
possible. We are using a small one in a CCR test. This commit clarifies
the purpose of that sleep by adding a comment explaining it. We also
removed a hard-coded value from the test, that if we ever modified the
value higher up where it was set, we could end up forgetting to change
the value here. Now we ensure that these would move in lock step if we
ever maintain them later.
We have some CCR tests where we use mock transport send rules to control
the behavior that we desire in these tests. Namely, we want to simulate
an exception being thrown on the leader side, or a variety of other
situations. These send rules were put in place between the data nodes on
each side. However, it might not be the case that these requests are
being sent between data nodes. For example, a request that is handled on
a non-data master node would not be sent from a data node. And it might
not be the case that the request is sent to a data node, as it could be
proxied through a non-data coordinating node. This commit addresses this
by putting these send rules in places between all nodes on each side.
Closes#39011Closes#39201
Initially in #38910, ShardFollowTask was reusing ImmutableFollowParameters'
serialization logic. After merging, bwc tests failed sometimes and
the binary serialization that ShardFollowTask was originally was using
was added back. ImmutableFollowParameters is using optional fields (optional vint)
while ShardFollowTask was not (vint).
Today we always refresh when looking up the primary term in
FollowingEngine. This is not necessary for we can simply
return none for operations before the global checkpoint.
The follower won't always have the same history as the leader for its
soft-deletes retention can be different. However, if some operation
exists on the history of the follower, then the same operation must
exist on the leader. This change relaxes the history check in
ShardFollowTaskReplicationTests.
Closes#39093
This commit fixes a broken CCR retention lease unfollow test. The
problem with the test is that the random subset of shards that we picked
to disrupt would not necessarily overlap with the actual shards in
use. We could take a non-empty subset of [0, 3] (e.g., { 2 }) when the
only shard IDs in use were [0, 1]. This commit fixes this by taking into
account the number of shards in use in the test.
With this change, we also take measure to ensure that a successful
branch is tested more frequently than would otherwise be the case. On
that branch, we want to sometimes pretend that the retention lease is
already removed. The randomness here was also sometimes selecting a
subset of shards that did not overlap with the shards actually in use
during the test. While this does not break the test, it is confusing and
reduces the amount of coverage of that branch.
Relates #39185
This commit attempts to remove the retention leases on the leader shards
when unfollowing an index. This is best effort, since the leader might
not be available.
Prior to this commit, if during fetch leader / follower GCP
a fatal error occurred, then the shard follow task was removed.
This is unexpected, because if such an error occurs during the lifetime of shard follow task then replication is stopped and the fatal error flag is set. This allows the ccr stats api to report the fatal exception that has occurred (instead of the user grepping through the elasticsearch logs).
This issue was found by a rare failure of the `FollowStatsIT#testFollowStatsApiIncludeShardFollowStatsWithRemovedFollowerIndex` test.
Closes#38779
* During fetching remote mapping if remote client is missing then
`NoSuchRemoteClusterException` was not handled.
* When adding remote connection, check that it is really connected
before continue-ing to run the tests.
Relates to #38695
This commit is the first step in integrating shard history retention
leases with CCR. In this commit we integrate shard history retention
leases with recovery from remote. Before we start transferring files, we
take out a retention lease on the primary. Then during the file copy
phase, we repeatedly renew the retention lease. Finally, when recovery
from remote is complete, we disable the background renewing of the
retention lease.
This commit adds a `ListenerTimeouts` class that will wrap a
`ActionListener` in a listener with a timeout scheduled on the generic
thread pool. If the timeout expires before the listener is completed,
`onFailure` will be called with an `ElasticsearchTimeoutException`.
Timeouts for the get ccr file chunk action are implemented using this
functionality. Additionally, this commit attempts to fix#38027 by also
blocking proxied get ccr file chunk actions. This test being un-muted is
useful to verify the timeout functionality.
Today when processing an operation on a replica engine (or the
following engine), we first add it to Lucene, then add it to translog,
then finally marks its seq_no as completed. If a flush occurs after step1,
but before step-3, the max_seq_no in the commit's user_data will be
smaller than the seq_no of some documents in the Lucene commit.
We verify seq_no_stats is aligned between copies at the end of some
disruption tests. Sometimes, the assertion `assertSeqNos` is tripped due
to a lagged global checkpoint on replicas. The global checkpoint on
replicas is lagged because we sync the global checkpoint 30 seconds (by
default) after the last replication operation. This change reduces the
global checkpoint sync-internal to 1s in the disruption tests.
Closes#38318Closes#36789
The CCR REST tests that rely on these assertions are flaky. They are
flaky since the introduction of recovery from the remote.
The underlying problem is this: these tests are making assertions about
the number of operations read by the shard following task. However, with
recovery from remote, we no longer have guarantees that the assumptions
these tests were relying on hold. Namely, these tests were assuming that
the only way that a document could land in the follower index is via the
shard following task. With recovery from remote, there is another way,
which is via the files that are copied over during the recovery
phase. Most of the time this will not be a problem because with the
small number of documents that we are indexing in these tests, it is
usally not the case that a flush would occur and so there would not be
any documents in the files copied over. However, a flush can occur any
time at which point all of the indexed documents could end up in a safe
commit and copied over during recovery from remote. This commit modifies
these assertions to ones that are not prone to this issue, yet still
validate the health of the follower shard.
The previous logic for concurrent file chunk fetching did not allow for multiple chunks from the same
file to be fetched in parallel. The parallelism only allowed to fetch chunks from different files in
parallel. This required complex logic on the follower to be aware from which file it was already
fetching information, in order to ensure that chunks for the same file would be fetched in sequential
order. During benchmarking, this exhibited throughput issues when recovery came towards the end,
where it would only be sequentially fetching chunks for the same largest segment file, with
throughput considerably going down in a high-latency network as there was no parallelism anymore.
The new logic here follows the peer recovery model more closely, and sends multiple requests for
the same file in parallel, and then reorders the results as necessary. Benchmarks show that this
leads to better overall throughput and the implementation is also simpler.
The should fix the following NPE:
```
[2019-02-11T23:27:48,452][WARN ][o.e.p.PersistentTasksNodeService] [node_s_0] task kD8YzUhHTK6uKNBNQI-1ZQ-0 failed with an exception
1> java.lang.NullPointerException: null
1> at org.elasticsearch.xpack.ccr.action.ShardFollowTasksExecutor.lambda$fetchFollowerShardInfo$7(ShardFollowTasksExecutor.java:305) ~[main/:?]
1> at org.elasticsearch.action.ActionListener$1.onResponse(ActionListener.java:61) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.TransportAction$1.onResponse(TransportAction.java:68) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.TransportAction$1.onResponse(TransportAction.java:64) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$AsyncAction.onCompletion(TransportBroadcastByNodeAction.java:383) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$AsyncAction.onNodeResponse(TransportBroadcastByNodeAction.java:352) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$AsyncAction$1.handleResponse(TransportBroadcastByNodeAction.java:324) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$AsyncAction$1.handleResponse(TransportBroadcastByNodeAction.java:314) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.TransportService$ContextRestoreResponseHandler.handleResponse(TransportService.java:1108) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.TransportService$DirectResponseChannel.processResponse(TransportService.java:1189) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.TransportService$DirectResponseChannel.sendResponse(TransportService.java:1169) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.TaskTransportChannel.sendResponse(TaskTransportChannel.java:54) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$BroadcastByNodeTransportRequestHandler.messageReceived(TransportBroadcastByNodeAction.java:417) [elasticsearch-8.0.0-SNAP
SHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.action.support.broadcast.node.TransportBroadcastByNodeAction$BroadcastByNodeTransportRequestHandler.messageReceived(TransportBroadcastByNodeAction.java:391) [elasticsearch-8.0.0-SNAP
SHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.RequestHandlerRegistry.processMessageReceived(RequestHandlerRegistry.java:63) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.transport.TransportService$7.doRun(TransportService.java:687) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.common.util.concurrent.ThreadContext$ContextPreservingAbstractRunnable.doRun(ThreadContext.java:751) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at org.elasticsearch.common.util.concurrent.AbstractRunnable.run(AbstractRunnable.java:37) [elasticsearch-8.0.0-SNAPSHOT.jar:8.0.0-SNAPSHOT]
1> at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) [?:1.8.0_202]
1> at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) [?:1.8.0_202]
1> at java.lang.Thread.run(Thread.java:748) [?:1.8.0_202]
```
Relates to #38779
* Add rolling upgrade multi cluster test module (#38277)
This test starts 2 clusters, each with 3 nodes.
First the leader cluster is started and tests are run against it and
then the follower cluster is started and tests execute against this two cluster.
Then the follower cluster is upgraded, one node at a time.
After that the leader cluster is upgraded, one node at a time.
Every time a node is upgraded tests are ran while both clusters are online.
(and either leader cluster has mixed node versions or the follower cluster)
This commit only tests CCR index following, but could be used for CCS tests as well.
In particular for CCR, unidirectional index following is tested during a rolling upgrade.
During the test several indices are created and followed in the leader cluster before or
while the follower cluster is being upgraded.
This tests also verifies that attempting to follow an index in the upgraded cluster
from the not upgraded cluster fails. After both clusters are upgraded following the
index that previously failed should succeed.
Relates to #37231 and #38037
* Filter out upgraded version index settings when starting index following (#38838)
The `index.version.upgraded` and `index.version.upgraded_string` are likely
to be different between leader and follower index. In the event that
a follower index gets restored on a upgraded node while the leader index
is still on non-upgraded nodes.
Closes#38835
Currently we index documents concurrently to attempt to ensure that we
update mappings during the restore process. However, this does not
actually test that the mapping will be correct and is dangerous as it
can lead to a misalignment between the max sequence number and the local
checkpoint. If these are not aligned, peer recovery cannot be completed
without initiating following which this test does not do. That causes
teardown assertions to fail.
This commit removes the concurrent indexing and flushes after the
documents are indexed. Additionally it modifies the mapping specific
test to ensure that there is a mapping update when the restore session
is initiated. This mapping update is picked up at the end of the restore
by the follower.
There were two documents (seq=2 and seq=103) missing on the follower in
one of the failures of `testFailOverOnFollower`. I spent several hours
on that failure but could not figure out the reason. I adjust log and
unmute this test so we can collect more information.
Relates #38633
The Close Index API has been refactored in 6.7.0 and it now performs
pre-closing sanity checks on shards before an index is closed: the maximum
sequence number must be equals to the global checkpoint. While this is a
strong requirement for regular shards, we identified the need to relax this
check in the case of CCR following shards.
The following shards are not in charge of managing the max sequence
number or global checkpoint, which are pulled from a leader shard. They
also fetch and process batches of operations from the leader in an unordered
way, potentially leaving gaps in the history of ops. If the following shard lags
a lot it's possible that the global checkpoint and max seq number never get
in sync, preventing the following shard to be closed and a new PUT Follow
action to be issued on this shard (which is our recommended way to
resume/restart a CCR following).
This commit allows each Engine implementation to define the specific
verification it must perform before closing the index. In order to allow
following/frozen/closed shards to be closed whatever the max seq number
or global checkpoint are, the FollowingEngine and ReadOnlyEngine do
not perform any check before the index is closed.
Co-authored-by: Martijn van Groningen <martijn.v.groningen@gmail.com>
Adds the ability to fetch chunks from different files in parallel, configurable using the new `ccr.indices.recovery.max_concurrent_file_chunks` setting, which defaults to 5 in this PR.
The implementation uses the parallel file writer functionality that is also used by peer recoveries.
In #38333 and #38350 we moved away from the `discovery.zen` settings namespace
since these settings have an effect even though Zen Discovery itself is being
phased out. This change aligns the documentation and the names of related
classes and methods with the newly-introduced naming conventions.
This is related to #35975. We do not want a slow master to fail a
recovery from remote process due to a slow put mappings call. This
commit increases the master node timeout on this call to 30 mins.
Currently the snapshot/restore process manually sets the global
checkpoint to the max sequence number from the restored segements. This
does not work for Ccr as this will lead to documents that would be
recovered in the normal followering operation from being recovered.
This commit fixes this issue by setting the initial global checkpoint to
the existing local checkpoint.
With this change we no longer support pluggable discovery implementations. No
known implementations of `DiscoveryPlugin` actually override this method, so in
practice this should have no effect on the wider world. However, we were using
this rather extensively in tests to provide the `test-zen` discovery type. We
no longer need a separate discovery type for tests as we no longer need to
customise its behaviour.
Relates #38410
Jason Tedor
Ryan Ernst
Armin Braun
Nhat Nguyen
Martijn van Groningen
Alpar Torok
Yannick Welsch
Gordon Brown
Mark Vieira
David Turner
Luca Cavanna
Simon Willnauer
Tal Levy
Ryan Ernst
Hicham Mallah
David Kyle
Christoph Büscher
Adrien Grand
Julie Tibshirani
Tim Brooks
alex101101
Like
Henning Andersen
Jake Landis
Daniel Mitterdorfer
Tanguy Leroux
Benjamin Trent
iverase
Przemyslaw Gomulka