If a shard follow-task hits a non-retryable error and stops, then we
should also stop the retention-leases renewal process associated with
that follow-task.
Today this test fails because the sizes of the snapshot
shards are only kept in a very short period of time in
the InternalSnapshotsInfoService and are not
guaranteed to exist once the shards are correctly
assigned.
closes#64167
If the deleted index has N shards, then ShardFollowTaskCleaner can send
N*(N-1)/2 requests to remove N shard-follow tasks. I think that's fine
as the implementation is straightforward. The test failed when the
deleted index has 8 shards. This commit increases the timeout in the
test.
Closes#64311
This test failed sometimes for various reasons: an empty bulk request
that can't be validated, a background force-merge that completes after
the store stats were collected and finally an assertBusy() that waits
10 seconds while we usually wait 60s on the follower cluster in CCR
tests.
Closes#64167
Determines the shard size of shards before allocating shards that are
recovering from snapshots. It ensures during shard allocation that the
target node that is selected as recovery target will have enough free
disk space for the recovery event. This applies to regular restores,
CCR bootstrap from remote, as well as mounting searchable snapshots.
The InternalSnapshotInfoService is responsible for fetching snapshot
shard sizes from repositories. It provides a getShardSize() method
to other components of the system that can be used to retrieve the
latest known shard size. If the latest snapshot shard size retrieval
failed, the getShardSize() returns
ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE. While
we'd like a better way to handle such failures, returning this value
allows to keep the existing behavior for now.
Note that this PR does not address an issues (we already have today)
where a replica is being allocated without knowing how much disk
space is being used by the primary.
Co-authored-by: Yannick Welsch <yannick@welsch.lu>
If the connection between clusters is disconnected or the leader cluster
is offline, then CCR shard-follow tasks can stop with "no seed node
left". CCR should retry on this error.
This commit adds a dedicated threadpool for system index write
operations. The dedicated resources for system index writes serves as
a means to ensure that user activity does not block important system
operations from occurring such as the management of users and roles.
Backport of #61655
This commit adds the `index.routing.allocation.prefer._tier` setting to the
`DataTierAllocationDecider`. This special-purpose allocation setting lets a user specify a
preference-based list of tiers for an index to be assigned to. For example, if the setting were set
to:
```
"index.routing.allocation.prefer._tier": "data_hot,data_warm,data_content"
```
If the cluster contains any nodes with the `data_hot` role, the decider will only allow them to be
allocated on the `data_hot` node(s). If there are no `data_hot` nodes, but there are `data_warm` and
`data_content` nodes, then the index will be allowed to be allocated on `data_warm` nodes.
This allows us to specify an index's preference for tier(s) without causing the index to be
unassigned if no nodes of a preferred tier are available.
Subsequent work will change the ILM migration to make additional use of this setting.
Relates to #60848
CCR shard follow task can hit CircuitBreakingException on the leader
cluster (read changes requests) or the follower cluster (bulk requests).
CCR should retry on CircuitBreakingException as it's a transient error.
If the master node of the follower cluster is busy, then the
auto-follower will fail to initialize the following process. This also
occurs when an auto-follow pattern matches multiple indices. We should
set the timeout of put-follow requests issued by the auto-follower to
unbounded to avoid this problem.
Closes#56891
This commit adds the functionality to allocate newly created indices on nodes in the "hot" tier by
default when they are created.
This does not break existing behavior, as nodes with the `data` role are considered to be part of
the hot tier. Users that separate their deployments by using the `data_hot` (and `data_warm`,
`data_cold`, `data_frozen`) roles will have their data allocated on the hot tier nodes now by
default.
This change is a little more complicated than changing the default value for
`index.routing.allocation.include._tier` from null to "data_hot". Instead, this adds the ability to
have a plugin inject a setting into the builder for a newly created index. This has the benefit of
allowing this setting to be visible as part of the settings when retrieving the index, for example:
```
// Create an index
PUT /eggplant
// Get an index
GET /eggplant?flat_settings
```
Returns the default settings now of:
```json
{
"eggplant" : {
"aliases" : { },
"mappings" : { },
"settings" : {
"index.creation_date" : "1597855465598",
"index.number_of_replicas" : "1",
"index.number_of_shards" : "1",
"index.provided_name" : "eggplant",
"index.routing.allocation.include._tier" : "data_hot",
"index.uuid" : "6ySG78s9RWGystRipoBFCA",
"index.version.created" : "8000099"
}
}
}
```
After the initial setting of this setting, it can be treated like any other index level setting.
This new setting is *not* set on a new index if any of the following is true:
- The index is created with an `index.routing.allocation.include.<anything>` setting
- The index is created with an `index.routing.allocation.exclude.<anything>` setting
- The index is created with an `index.routing.allocation.require.<anything>` setting
- The index is created with a null `index.routing.allocation.include._tier` value
- The index was created from an existing source metadata (shrink, clone, split, etc)
Relates to #60848
* Faster `equals` for `BytesArray` which is nice since with this change we use it for the search cache
* Lighter `StreamInput` for `BytesArray` that should save memory and some indirection relative to the one on the abstract bytes reference
* Lighter `writeTo` implementation
* Build a `BytesArray` instead of a PagedBytesReference whenever possible to save indirection and memory
The test failed because the leader was taking a lot of CPUs to process
many mapping updates. This commit reduces the mapping updates, increases
timeout, and adds more debug info.
Closes#59832
CCR will stop functioning if the master node is on 7.8, but data nodes
are before that version because the master node considers that all data
nodes do not have the remote cluster client role. This commit allows CCR
work on data nodes with legacy roles only.
Relates #54146
Relates #59375
If a primary shard of a follower index is being relocated, then we
will fail to create a follow-task. This validation is too restricted.
We should ensure that all primaries of the follower index are active
instead.
Closes#59625
Today, a follow task will fail if the master node of the follower
cluster is temporarily overloaded and unable to process master node
requests (such as update mapping, setting, or alias) from a follow-task
within the default timeout. This error is transient, and follow-tasks
should not abort. We can avoid this problem by setting the timeout of
master node requests on the follower cluster to unbounded.
Closes#56891
It can take more than 10 seconds to auto-follow and create a follow-task
on a slow CI. This commit increases timeout in AutoFollowIT by replacing
assertBusy with assertLongBusy.
Closes#59952
Enables fully concurrent snapshot operations:
* Snapshot create- and delete operations can be started in any order
* Delete operations wait for snapshot finalization to finish, are batched as much as possible to improve efficiency and once enqueued in the cluster state prevent new snapshots from starting on data nodes until executed
* We could be even more concurrent here in a follow-up by interleaving deletes and snapshots on a per-shard level. I decided not to do this for now since it seemed not worth the added complexity yet. Due to batching+deduplicating of deletes the pain of having a delete stuck behind a long -running snapshot seemed manageable (dropped client connections + resulting retries don't cause issues due to deduplication of delete jobs, batching of deletes allows enqueuing more and more deletes even if a snapshot blocks for a long time that will all be executed in essentially constant time (due to bulk snapshot deletion, deleting multiple snapshots is mostly about as fast as deleting a single one))
* Snapshot creation is completely concurrent across shards, but per shard snapshots are linearized for each repository as are snapshot finalizations
See updated JavaDoc and added test cases for more details and illustration on the functionality.
Some notes:
The queuing of snapshot finalizations and deletes and the related locking/synchronization is a little awkward in this version but can be much simplified with some refactoring. The problem is that snapshot finalizations resolve their listeners on the `SNAPSHOT` pool while deletes resolve the listener on the master update thread. With some refactoring both of these could be moved to the master update thread, effectively removing the need for any synchronization around the `SnapshotService` state. I didn't do this refactoring here because it's a fairly large change and not necessary for the functionality but plan to do so in a follow-up.
This change allows for completely removing any trickery around synchronizing deletes and snapshots from SLM and 100% does away with SLM errors from collisions between deletes and snapshots.
Snapshotting a single index in parallel to a long running full backup will execute without having to wait for the long running backup as required by the ILM/SLM use case of moving indices to "snapshot tier". Finalizations are linearized but ordered according to which snapshot saw all of its shards complete first
Many of the parameters we pass into this method were only used to
build the `SnapshotInfo` instance to write.
This change simplifies the signature. Also, it seems less error prone to build
`SnapshotInfo` in `SnapshotsService` isntead of relying on the fact that each repository
implementation will build the correct `SnapshotInfo`.
This PR introduces two new fields in to `RepositoryData` (index-N) to track the blob name of `IndexMetaData` blobs and their content via setting generations and uuids. This is used to deduplicate the `IndexMetaData` blobs (`meta-{uuid}.dat` in the indices folders under `/indices` so that new metadata for an index is only written to the repository during a snapshot if that same metadata can't be found in another snapshot.
This saves one write per index in the common case of unchanged metadata thus saving cost and making snapshot finalization drastically faster if many indices are being snapshotted at the same time.
The implementation is mostly analogous to that for shard generations in #46250 and piggy backs on the BwC mechanism introduced in that PR (which means this PR needs adjustments if it doesn't go into `7.6`).
Relates to #45736 as it improves the efficiency of snapshotting unchanged indices
Relates to #49800 as it has the potential of loading the index metadata for multiple snapshots of the same index concurrently much more efficient speeding up future concurrent snapshot delete
Currently we combine coordinating and primary bytes into a single bucket
for indexing pressure stats. This makes sense for rejection logic.
However, for metrics it would be useful to separate them.
The primary shards of follower indices during the bootstrap need to be
on nodes with the remote cluster client role as those nodes reach out to
the corresponding leader shards on the remote cluster to copy Lucene
segment files and renew the retention leases. This commit introduces a
new allocation decider that ensures bootstrapping follower primaries are
allocated to nodes with the remote cluster client role.
Co-authored-by: Jason Tedor <jason@tedor.me>
We have recently added internal metrics to monitor the amount of
indexing occurring on a node. These metrics introduce back pressure to
indexing when memory utilization is too high. This commit exposes these
stats through the node stats API.
Today, we send operations in phase2 of peer recoveries batch by batch
sequentially. Normally that's okay as we should have a fairly small of
operations in phase 2 due to the file-based threshold. However, if
phase1 takes a lot of time and we are actively indexing, then phase2 can
have a lot of operations to replay.
With this change, we will send multiple batches concurrently (defaults
to 1) to reduce the recovery time.
Backport of #58018
This is a follow-up to #57573. This commit combines coordinating and
primary bytes under the same "write" bucket. Double accounting is
prevented by only accounting the bytes at either the reroute phase or
the primary phase. TransportBulkAction calls execute directly, so the
operations handler is skipped and the bytes are not double accounted.
When the documents are large, a follower can receive a partial response
because the requesting range of operations is capped by
max_read_request_size instead of max_read_request_operation_count. In
this case, the follower will continue reading the subsequent ranges
without checking the remaining size of the buffer. The buffer then can
use more memory than max_write_buffer_size and even causes OOM.
Backport of #58620
The checks on the license state have a singular method, isAllowed, that
returns whether the given feature is allowed by the current license.
However, there are two classes of usages, one which intends to actually
use a feature, and another that intends to return in telemetry whether
the feature is allowed. When feature usage tracking is added, the latter
case should not count as a "usage", so this commit reworks the calls to
isAllowed into 2 methods, checkFeature, which will (eventually) both
check whether a feature is allowed, and keep track of the last usage
time, and isAllowed, which simply determines whether the feature is
allowed.
Note that I considered having a boolean flag on the current method, but
wanted the additional clarity that a different method name provides,
versus a boolean flag which is more easily copied without realizing what
the flag means since it is nameless in call sites.
Restoring from a snapshot (which is a particular form of recovery) does not currently take recovery throttling into account
(i.e. the `indices.recovery.max_bytes_per_sec` setting). While restores are subject to their own throttling (repository
setting `max_restore_bytes_per_sec`), this repository setting does not allow for values to be configured differently on a
per-node basis. As restores are very similar in nature to peer recoveries (streaming bytes to the node), it makes sense to
configure throttling in a single place.
The `max_restore_bytes_per_sec` setting is also changed to default to unlimited now, whereas previously it was set to
`40mb`, which is the current default of `indices.recovery.max_bytes_per_sec`). This means that no behavioral change
will be observed by clusters where the recovery and restore settings were not adapted.
Relates https://github.com/elastic/elasticsearch/issues/57023
Co-authored-by: James Rodewig <james.rodewig@elastic.co>
Today the disk-based shard allocator accounts for incoming shards by
subtracting the estimated size of the incoming shard from the free space on the
node. This is an overly conservative estimate if the incoming shard has almost
finished its recovery since in that case it is already consuming most of the
disk space it needs.
This change adds to the shard stats a measure of how much larger each store is
expected to grow, computed from the ongoing recovery, and uses this to account
for the disk usage of incoming shards more accurately.
Backport of #58029 to 7.x
* Picky picky
* Missing type
Today we have individual settings for configuring node roles such as
node.data and node.master. Additionally, roles are pluggable and we have
used this to introduce roles such as node.ml and node.voting_only. As
the number of roles is growing, managing these becomes harder for the
user. For example, to create a master-only node, today a user has to
configure:
- node.data: false
- node.ingest: false
- node.remote_cluster_client: false
- node.ml: false
at a minimum if they are relying on defaults, but also add:
- node.master: true
- node.transform: false
- node.voting_only: false
If they want to be explicit. This is also challenging in cases where a
user wants to have configure a coordinating-only node which requires
disabling all roles, a list which we are adding to, requiring the user
to keep checking whether a node has acquired any of these roles.
This commit addresses this by adding a list setting node.roles for which
a user has explicit control over the list of roles that a node has. If
the setting is configured, the node has exactly the roles in the list,
and not any additional roles. This means to configure a master-only
node, the setting is merely 'node.roles: [master]', and to configure a
coordinating-only node, the setting is merely: 'node.roles: []'.
With this change we deprecate the existing 'node.*' settings such as
'node.data'.
* Add support for snapshot and restore to data streams (#57675)
This change adds support for including data streams in snapshots.
Names are provided in indices field (the same way as in other APIs), wildcards are supported.
If rename pattern is specified it renames both data streams and backing indices.
It also adds test to make sure SLM works correctly.
Closes#57127
Relates to #53100
* version fix
* compilation fix
* compilation fix
* remove unused changes
* compilation fix
* test fix
Today when creating a follower index via the put follow API, or via an
auto-follow pattern, it is not possible to specify settings overrides
for the follower index. Instead, we copy all of the leader index
settings to the follower. Yet, there are cases where a user would want
some different settings on the follower index such as the number of
replicas, or allocation settings. This commit addresses this by allowing
the user to specify settings overrides when creating follower index via
manual put follower calls, or via auto-follow patterns. Note that not
all settings can be overrode (e.g., index.number_of_shards) so we also
have detection that prevents attempting to override settings that must
be equal between the leader and follow index. Note that we do not even
allow specifying such settings in the overrides, even if they are
specified to be equal between the leader and the follower
index. Instead, the must be implicitly copied from the leader index, not
explicitly set by the user.
The leases issued by CCR keep one extra operation around on the leader shards. This is not
harmful to the leader cluster, but means that there's potentially one delete that can't be
cleaned up.
Merging logic is currently split between FieldMapper, with its merge() method, and
MappedFieldType, which checks for merging compatibility. The compatibility checks
are called from a third class, MappingMergeValidator. This makes it difficult to reason
about what is or is not compatible in updates, and even what is in fact updateable - we
have a number of tests that check compatibility on changes in mapping configuration
that are not in fact possible.
This commit refactors the compatibility logic so that it all sits on FieldMapper, and
makes it called at merge time. It adds a new FieldMapperTestCase base class that
FieldMapper tests can extend, and moves the compatibility testing machinery from
FieldTypeTestCase to here.
Relates to #56814
Nhat Nguyen
Tanguy Leroux
Armin Braun
Howard
Jay Modi
Lee Hinman
Yannick Welsch
Tim Brooks
Daniel Mitterdorfer
Dan Hermann
Ryan Ernst
David Turner
Jason Tedor
Przemko Robakowski
Stuart Tettemer
Alan Woodward