The autoGeneratedTimestamp field is internally used to speed up indexing of operations with
auto-ids, as we can rule out duplicates. Setting this field externally can make the index
inconsistent, resulting in duplicate documents with same id.
Adds support for handling auto-id requests with optype CREATE. Also simplifies the code
handling this by using the standard indexing path when dealing with possible retry conflicts.
Relates #47169
Bulk requests currently do not allow adding "create" actions with auto-generated IDs.
This commit allows using the optype CREATE for append-only indexing operations. This is
mainly the user facing aspect of it.
Synonym queries (when two tokens/paths start at the same position) use the alias field instead
of the concrete field to build Lucene queries. This commit fixes this bug by resolving the alias field upfront in order to provide the concrete field to the actual query parser.
Today, we don't clear the shard info of the primary shard when a new
node joins; then we might risk of making replica allocation decisions
based on the stale information of the primary. The serious problem is
that we can cancel the current recovery which is more advanced than the
copy on the new node due to the old info we have from the primary.
With this change, we ensure the shard info from the primary is not older
than any node when allocating replicas.
Relates #46959
This work was done by Henning in #42518.
Co-authored-by: Henning Andersen <henning.andersen@elastic.co>
This commit adjusts randomization for the cluster shard limit tests so
that there is often more of a gap left between the limit and the size of
the first index. This allows the same randomization to be used for all
tests, and alleviates flakiness in
`testIndexCreationOverLimitFromTemplate`.
Today the comment boldly claims that this line of code keeps nodes above the
10-byte low watermark when in fact this is not true at all. This change fixes
this so that it really does keep nodes above the low watermark.
Fixes#45338. Again.
This is a preliminary of #46250 making the snapshot
delete work by doing all the metadata updates first
and then bulk deleting all of the now unreferenced
blobs.
Before this change, the metadata updates for each shard
and subsequent deletion of the blobs that have become unreferenced
due to the delete would happen sequentially shard-by-shard
parallelising only over all the indices in the snapshot.
This change makes it so the all the metadata updates
happen in parallel on a shard level first.
Once all of the updates of shard-level metadata have finished,
all the now unreferenced blobs are deleted in bulk.
This has two benefits (outside of making #46250 a smaller change):
* We have a lower likelihood of failing to update shard level metadata because
it happens with priority and a higher degree of parallelism
* Deleting of unreferenced data in the shards should go much faster in many cases (rolling indices, large number of indices with many unchanged shards) as well because a number of small bulk deletions (just two blobs for `index-N` and `snap-` for each unchanged shard) are grouped into larger bulk deletes of `100-1000` blobs depending on Cloud provider (even though the final bulk deletes are happening sequentially this should be much faster in almost all cases as you'd parallelism of 50 (GCS) to 500 (S3) snapshot threads to achieve the same delete rates when deleting from unchanged shards).
We cancel ongoing peer recoveries if a node joins the cluster with a completely
up-to-date copy of a shard, because we can use such a copy to recover a replica
instantly. However, today we only look for recoveries to cancel while there are
unassigned shards in the cluster. This means that we do not contemplate the
cancellation of the last few recoveries since recovering shards are not
unassigned. It might take much longer for these recoveries to complete than
would be necessary if they were cancelled.
This commit fixes this by checking for cancellable recoveries even if all
shards are assigned.
As a result of #45689 snapshot finalization started to
take significantly longer than before. This may be a
little unfortunate since it increases the likelihood
of failing to finalize after having written out all
the segment blobs.
This change parallelizes all the metadata writes that
can safely run in parallel in the finalization step to
speed the finalization step up again. Also, this will
generally speed up the snapshot process overall in case
of large number of indices.
This is also a nice to have for #46250 since we add yet
another step (deleting of old index- blobs in the shards
to the finalization.
The method Setting#getRaw leaks implementation details about settings,
namely that they are backed by strings. We do not want code to rely upon
this, so this commit makes Setting#getRaw private as a first step
towards hiding the implementaton details of settings from the rest of
the codebase.
Today plugins may provide upgraders for custom metadata and index metadata, but
these upgraders are bypassed during a rolling restart. Fortunately this
extension mechanism is unused by all known plugins. This commit removes these
extension points.
Relates #47297
Today if you create an index in a cluster without any data nodes then it will
report yellow health because it never attempts to assign any shards if there
are no data nodes, so the new shards remain at `AllocationStatus.NO_ATTEMPT`.
This commit moves the new primaries to `AllocationStatus.DECIDERS_NO` in this
situation, causing the cluster health to move to red.
Fixes#41073
Fixes a bug related to how "closed replicated indices" (introduced in 7.2) interact with the index
metadata storage mechanism, which has special handling for closed indices (but incorrectly
handles replicated closed indices). On non-master-eligible data nodes, it's possible for the
node's manifest file (which tracks the relevant metadata state that the node should persist) to
become out of sync with what's actually stored on disk, leading to an inconsistency that is then
detected at startup, refusing for the node to start up.
Closes#47276
Currently DateMathParser with roundUp = true is relying on the DateFormatter build with
combined optional sub parsers with defaulted fields (depending on the formatter).
That means that for yyyy-MM-dd'T'HH:mm:ss||yyyy-MM-dd'T'HH:mm:ss.SSS
Java.time implementation expects optional parsers in order from most specific to
least specific (reverse in the example above).
It is causing a problem because the first parsing succeeds but does not consume the full input.
The second parser should be used.
We can work around this with keeping a list of RoundUpParsers and iterate over them choosing
the one that parsed full input. The same approach we used for regular (non date math) in
relates #40100
The jdk is not considering this to be a bug https://bugs.openjdk.java.net/browse/JDK-8188771
Those below will expect this change first
relates #46242
relates #45284
backport #46654
Backport of #45794 to 7.x. Convert most `awaitBusy` calls to
`assertBusy`, and use asserts where possible. Follows on from #28548 by
@liketic.
There were a small number of places where it didn't make sense to me to
call `assertBusy`, so I kept the existing calls but renamed the method to
`waitUntil`. This was partly to better reflect its usage, and partly so
that anyone trying to add a new call to awaitBusy wouldn't be able to find
it.
I also didn't change the usage in `TransportStopRollupAction` as the
comments state that the local awaitBusy method is a temporary
copy-and-paste.
Other changes:
* Rework `waitForDocs` to scale its timeout. Instead of calling
`assertBusy` in a loop, work out a reasonable overall timeout and await
just once.
* Some tests failed after switching to `assertBusy` and had to be fixed.
* Correct the expect templates in AbstractUpgradeTestCase. The ES
Security team confirmed that they don't use templates any more, so
remove this from the expected templates. Also rewrite how the setup
code checks for templates, in order to give more information.
* Remove an expected ML template from XPackRestTestConstants The ML team
advised that the ML tests shouldn't be waiting for any
`.ml-notifications*` templates, since such checks should happen in the
production code instead.
* Also rework the template checking code in `XPackRestTestHelper` to give
more helpful failure messages.
* Fix issue in `DataFrameSurvivesUpgradeIT` when upgrading from < 7.4
This commit replaces some uses of Setting#getRaw in the throttling
allocation decider settings. Instead, these settings should be using
fallback settings.
This commit removes some leniency that exists in getting the allow
rebalance setting. Fortunately, that leniency is dead code, this can
never happen. The reason this can never happen is because the settings
infrastructure will not allow setting an invalid value for this
setting. If you try to set this in the elasticsearch.yml, then the node
will fail to start, since parsing the setting will fail. If you try to
set this via an update settings API call, then parsing the setting will
fail and the settings update will be rejected. Therefore, this leniency
can never be activated, so we remove it.
This commit is the first of a few in an attempt to remove the public
uses of Setting#getRaw.
This is the Java side of https://github.com/elastic/ml-cpp/pull/593
with a fallback so that ml-cpp bundles with either the
new or old directory structure work for the time being.
A few days after merging the C++ changes a followup to
this change will be made that removes the fallback.
This change merges the `ShardSearchTransportRequest` and `ShardSearchLocalRequest`
into a single `ShardSearchRequest` that can be used to create a SearchContext.
Relates #46523
Backport of #46241
This PR changes the ingest executing to be non blocking
by adding an additional method to the Processor interface
that accepts a BiConsumer as handler and changing
IngestService#executeBulkRequest(...) to ingest document
in a non blocking fashion iff a processor executes
in a non blocking fashion.
This is the second PR that merges changes made to server module from
the enrich branch (see #32789) into the master branch.
The plan is to merge changes made to the server module separately from
the pr that will merge enrich into master, so that these changes can
be reviewed in isolation.
This change originates from the enrich branch and was introduced there
in #43361.
Today if metadata persistence is excessively slow on a master-ineligible node
then the `ClusterApplierService` emits a warning indicating that the
`GatewayMetaState` applier was slow, but gives no further details. If it is
excessively slow on a master-eligible node then we do not see any warning at
all, although we might see other consequences such as a lagging node or a
master failure.
With this commit we emit a warning if metadata persistence takes longer than a
configurable threshold, which defaults to `10s`. We also emit statistics that
record how much index metadata was persisted and how much was skipped since
this can help distinguish cases where IO was slow from cases where there are
simply too many indices involved.
Backport of #47005.
Currently the logic to check if a connection to a remote discovery node
exists and otherwise create a proxy connection is mixed with the
collect nodes, cluster connection lifecycle, and other
RemoteClusterConnection logic. This commit introduces a specialized
RemoteConnectionManager class which handles the open connections.
Additionally, it reworks the "round-robin" proxy logic to create the list
of potential connections at connection open/close time, opposed to each
time a connection is requested.
We can have a large number of shard copies in this test. For example,
the two recent failures have 24 and 27 copies respectively and all
replicas have to copy segment files as their stores are corrupted. Our
CI needs more than 30 seconds to start all these copies.
Note that in two recent failures, the cluster was green just after the
cluster health timed out.
Closes#41899
* Wait for snapshot completion in SLM snapshot invocation
This changes the snapshots internally invoked by SLM to wait for
completion. This allows us to capture more snapshotting failure
scenarios.
For example, previously a snapshot would be created and then registered
as a "success", however, the snapshot may have been aborted, or it may
have had a subset of its shards fail. These cases are now handled by
inspecting the response to the `CreateSnapshotRequest` and ensuring that
there are no failures. If any failures are present, the history store
now stores the action as a failure instead of a success.
Relates to #38461 and #43663
We should not be quietly ignoring a corrupted shard-level index-N
blob. Simply creating a new empty shard-level index-N and moving
on means that all snapshots of that shard show `SUCESS` as their
state at the repository root but are in fact broken.
This change at least makes it visible to the user that they can't
snapshot the given shard any more and forces the user to move on
to a new repository since the current one is broken and will not
allow snapshotting the inconsistent shard again.
Also, this change stops the delete action for shards with broken
index-N blobs instead of simply deleting all blobs in the path
containing the broken index-N. This prevents a temporarily
broken/missing index-N blob from corrupting all snapshots of
that shard.
Simplify `SnapshotResiliencyTests` to more closely
match the structure of `AbstractCoordinatorTestCase` and allow for
future drying up between the two classes:
* Make the test cluster nodes a nested-class in the test cluster itself
* Remove the needless custom network disruption implementation and
simply track disconnected node ids like `AbstractCoordinatorTestCase`
does
Today we log and swallow exceptions during cluster state application, but such
an exception should not occur. This commit adds assertions of this fact, and
updates the Javadocs to explain it.
Relates #47038
We emit a debug log message whenever a child circuit breaker trips (in
`ChildMemoryCircuitBreaker#circuitBreak(String, long)`) but we never
emit a log message when the parent circuit breaker trips. As this is
more likely to happen with the real memory circuit breaker it is not
possible to detect this in the logs. With this commit we add a log
message on the same log level (debug) when the parent circuit breaker
trips.
We speculatively added support for `regexp` queries on the `_index` field in
#34089 (this functionality was not actually requested by a user). Supporting
regex logic adds complexity to the `_index` field for not much gain, so we
would like to remove it.
From an end-to-end test it turns out this functionality never even worked in
the first place because of an error in how regex flags were interpreted! For
this reason, we can remove support for `regexp` on `_index` without a
deprecation period.
Relates to #46640.
Currently in the ConnectionManager we lock around the node id. This is
odd because we key connections by the ephemeral id. Upon further
investigation it appears to me that we do not need the locking. Using
the concurrent map, we can ensure that only one connection attempt
completes. There is a very small chance that a new connection attempt
will proceed right as another connection attempt is completing. However,
since the whole process is asynchronous and event oriented
(lightweight), that does not seem to be an issue.
Due to recent changes in the nio transport, a failure to bind the server
channel has started to be logged at an error level. This exception leads
to an automatic retry on a different port, so it should only be logged
at a trace level.
Today the `LeaderChecker` rejects checks from nodes that are not in the current
cluster with the exception message `leader check from unknown node` which
offers no information about why the node is unknown. In fact the node must have
been in the cluster in the recent past, so it might help guide the user to a
more useful log message if we describe it as a `removed node` instead of an
`unknown node`. This commit changes the exception message like this, and also
tidies up a few other loose ends in the `LeaderChecker`.
Today `GatewayMetaState` implements `PersistedState` but it's an error to use
it as a `PersistedState` before it's been started, or if the node is
master-ineligible. It also holds some fields that are meaningless on nodes that
do not persist their states. Finally, it takes responsibility for both loading
the original cluster state and some of the high-level logic for writing the
cluster state back to disk.
This commit addresses these concerns by introducing a more specific
`PersistedState` implementation for use on master-eligible nodes which is only
instantiated if and when it's appropriate. It also moves the fields and
high-level persistence logic into a new `IncrementalClusterStateWriter` with a
more appropriate lifecycle.
Follow-up to #46326 and #46532
Relates #47001
Previously, queries on the _index field were not able to specify index aliases.
This was a regression in functionality compared to the 'indices' query that was
deprecated and removed in 6.0.
Now queries on _index can specify an alias, which is resolved to the concrete
index names when we check whether an index matches. To match a remote shard
target, the pattern needs to be of the form 'cluster:index' to match the
fully-qualified index name. Index aliases can be specified in the following query
types: term, terms, prefix, and wildcard.
This commit replaces the SearchContext used in AbstractQueryTestCase with
a QueryShardContext in order to reduce the visibility of search contexts.
Relates #46523
`testLogsWarningPeriodicallyIfClusterNotFormed` simulates a leader failure and
waits for long enough that a failing leader check is scheduled. However it does
not wait for the failing check to actually fail, which requires another two
actions and therefore might take up to 200ms more. Unlucky timing would result
in this test failing, for instance:
./gradle ':server:test' \
--tests "org.elasticsearch.cluster.coordination.CoordinatorTests.testLogsWarningPeriodicallyIfClusterNotFormed" \
-Dtests.jvm.argline="-Dhppc.bitmixer=DETERMINISTIC" \
-Dtests.seed=F18CDD0EBEB5653:E9BC1A8B062E697A
This commit adds the extra delay needed for the leader failure to complete as
expected.
Fixes#46920
This at a very low rate and with the force merge in place
before checking the cache size it's not clear why the cache
is not of size `0` -> seems something else
must be happening here that is unexpected.
-> add debug logging to this test to find out
Relates #46701
We were repeatedly trying to send shard state updates for aborted
snapshots on every cluster state update.
This is simply dead-code since those updates are already safely
sent in the callbacks passed to `SnapshotShardsService#snapshot`.
On master failover, we ensure that the status update is resent
via `SnapshotShardsService#syncShardStatsOnNewMaster`.
=> there is no need for trying to send updates here over and over
and this logic can safely be removed
This assertion was added during the development of required
pipelines. In the initial version of that work, the notion of whether or
not a request was forwarded from the coordinating node to an ingest node
was introduced. It was realized later that instead we needed to track
whether or not the pipeline for the request was resolved. When that
change was made, this assertion, while not incorrect, was left behind
and only applied if the coordnating node was forwarding the
request. Instead, the assertion applies whether or not the request is
being forwarded. This commit addresses that by moving the assertion and
renaming some variables.
This commit adds the ability to require an ingest pipeline on an
index. Today we can have a default pipeline, but that could be
overridden by a request pipeline parameter. This commit introduces a new
index setting index.required_pipeline that acts similarly to
index.default_pipeline, except that it can not be overridden by a
request pipeline parameter. Additionally, a default pipeline and a
request pipeline can not both be set. The required pipeline can be set
to _none to ensure that no pipeline ever runs for index requests on that
index.
This test is broken with a very low failure rate after recent changes.
Particularly after #45689 which does not check for duplicate snapshot
uuids during snapshot finalization any more.
The check for duplicate uuids during finalization was removed
conciously since it lead to problems during master failover.
This test fails because it increments the repository state id in an
unexpected manner now, starting from the impossible situation of having
the same snapshot UUID for two different repository state ids.
This situation can't normally be reached, but we manually crafted it here.
This test didn't do anything before though, because the manually crafted
cluster state would simply result in an error during finalization before
and nothing but a normal snapshot delete would be tested.
=> removing this test here, it doesn't test anything.
Closes#46843
When using auto-generated IDs + the ingest drop processor (which looks to be used by filebeat
as well) + coordinating nodes that do not have the ingest processor functionality, this can lead
to a NullPointerException.
The issue is that markCurrentItemAsDropped() is creating an UpdateResponse with no id when
the request contains auto-generated IDs. The response serialization is lenient for our
REST/XContent format (i.e. we will send "id" : null) but the internal transport format (used for
communication between nodes) assumes for this field to be non-null, which means that it can't
be serialized between nodes. Bulk requests with ingest functionality are processed on the
coordinating node if the node has the ingest capability, and only otherwise sent to a different
node. This means that, in order to reproduce this, one needs two nodes, with the coordinating
node not having the ingest functionality.
Closes#46678
In #46250 we will have to move to two different delete
paths for BwC. Both paths will share the initial
reads from the repository though so I extracted/separated the
delete logic away from the initial reads to significantly simplify
the diff against #46250 here.
Also, I added some JavaDoc from #46250 here as well which makes the
code a little easier to follow even ignoring #46250 I think.
Suppress the reasonable-history check in this test to guarantee we're always getting ops based recovery even after a background sync.
Closes#45953
Co-Authored-By: David Turner <david.turner@elastic.co>
In the case that an ingest processor factory relies on other configuration
in the cluster state in order to construct a processor instance then
it is currently undetermined if a processor facotry can be notified about
a change if multiple cluster state updates are bundled together and
if a processor implement `ClusterStateApplier` interface.
(IngestService implements this interface too)
The idea with ingest cluster state listener is that it is guaranteed to
update the processor factory first before the ingest service creates
a pipeline with their respective processor instances.
Currently this concept is used in the enrich branch:
https://github.com/elastic/elasticsearch/blob/enrich/x-pack/plugin/enrich/src/main/java/org/elasticsearch/xpack/enrich/EnrichProcessorFactory.java#L21
In this case it a processor factory is interested in enrich indices' _meta
mapping fields.
This is the third PR that merges changes made to server module from
the enrich branch (see #32789) into the master branch.
Changes to the server module are merged separately from the pr that will
merge enrich into master, so that these changes can be reviewed in isolation.
Reenable this test since it was fixed by #45689 in production
code (specifically, the fact that we write the `snap-` blobs
without overwrite checks now).
Only required adding the assumed blocking on index file writes
to test code to properly work again.
* Closes#25281
* Write metadata during snapshot finalization after segment files to prevent outdated metadata in case of dynamic mapping updates as explained in #41581
* Keep the old behavior of writing the metadata beforehand in the case of mixed version clusters for BwC reasons
* Still overwrite the metadata in the end, so even a mixed version cluster is fixed by this change if a newer version master does the finalization
* Fixes#41581
Without replicas we won't actually get any relocations
going when removing the node constraints in this test.
Adjusted the code to force relocations by forbidding
nodes that hold primaries instead.
Also, fixed the timeouts and asserted that we actually
get relocations.
Fixes#46276
* [ILM] Add date setting to calculate index age
Add the `index.lifecycle.origination_date` to allow users to configure a
custom date that'll be used to calculate the index age for the phase
transmissions (as opposed to the default index creation date).
This could be useful for users to create an index with an "older"
origination date when indexing old data.
Relates to #42449.
* [ILM] Don't override creation date on policy init
The initial approach we took was to override the lifecycle creation date
if the `index.lifecycle.origination_date` setting was set. This had the
disadvantage of the user not being able to update the `origination_date`
anymore once set.
This commit changes the way we makes use of the
`index.lifecycle.origination_date` setting by checking its value when
we calculate the index age (ie. at "read time") and, in case it's not
set, default to the index creation date.
* Make origination date setting index scope dynamic
* Document orignation date setting in ilm settings
(cherry picked from commit d5bd2bb77ee28c1978ab6679f941d7c02e389d32)
Signed-off-by: Andrei Dan <andrei.dan@elastic.co>
* Parallelize Repository Cleanup Actions
Deleting root blobs and unreferenced indices can safely happen in parallel,
no need to have both operations run sequentially when they preclude all other
repository operations.
We added docs for proxy mode in #40281 but on reflection we should not be
documenting this setting since it does not play well with all proxies and we
can't recommend its use. This commit removes those docs and expands its Javadoc
instead.
We renew the CCR retention lease at a fixed interval, therefore it's
possible to have more than one in-flight renewal requests at the same
time. If requests arrive out of order, then the assertion is violated.
Closes#46416Closes#46013
Synced-flush consists of three steps: (1) force-flush on every active
copy; (2) check for ongoing indexing operations; (3) seal copies if
there's no change since step 1. If some indexing operations are
completed on the primary but not replicas, then Lucene commits from step
1 on replicas won't be the same as the primary's. And step 2 would pass
if it's executed when all pending operations are done. Once step 2
passes, we will incorrectly emit the "out of sync" warning message
although nothing wrong here.
Relates #28464
Relates #30244
There's nothing wrong in the logs from these failures. I think 30
seconds might not be enough to relocate shards with many documents as CI
is quite slow. This change increases the timeout to 60 seconds for these
relocation tests. It also dumps the hot threads in case of timed out.
Closes#46526Closes#46439
This change delays the creation of the SubSearchContext for nested and parent/child inner_hits
to the fetch sub phase in order to ensure that a SearchContext can built entirely from a
QueryShardContext. This commit also adds a validation step to the inner hits builder that ensures that we fail the request early if the inner hits path is invalid.
Relates #46523
Fixes that way linestrings that are crossing the antimeridian are
indexed due to a normalization bug these lines were decomposed into
a line segment that was stretching entire globe.
Fixes#43775
This makes the AllocatedPersistentTask#init() method protected so that
implementing classes can perform their initialization logic there,
instead of the constructor. Rollup's task is adjusted to use this
init method.
It also slightly refactors the methods to se a static logger in the
AllocatedTask instead of passing it in via an argument. This is
simpler, logged messages come from the task instead of the
service, and is easier for tests
This commit replaces the `SearchContext` with the `QueryShardContext` when building aggregator factories. Aggregator factories are part of the `SearchContext` so they shouldn't require a `SearchContext` to create them.
The main changes here are the signatures of `AggregationBuilder#build` that now takes a `QueryShardContext` and `AggregatorFactory#createInternal` that passes the `SearchContext` to build the `Aggregator`.
Relates #46523
* Fix Path comparisons for Windows tests
The test NodeEnvironmentTests#testCustonDataPaths worked just fine on
Darwin and Linux, but the comparison was breaking in Windows because one
path had the "C:\" prefix and the other one didn't. The simple fix is to
compare absolute paths rather than potentially relative ones.
Currently we allow `_field_names` fields to be disabled explicitely, but since
the overhead is negligible now we decided to keep it turned on by default and
deprecate the `enable` option on the field type. This change adds a deprecation
warning whenever this setting is used, going forward we want to ignore and finally
remove it.
Closes#27239
* More Efficient Ordering of Shard Upload Execution (#42791)
* Change the upload order of of snapshots to work file by file in parallel on the snapshot pool instead of merely shard-by-shard
* Inspired by #39657
* Cleanup BlobStoreRepository Abort and Failure Handling (#46208)
This commit replaces the `SearchContext` with the `QueryShardContext` when building collapsing conteext
Collapse context is part of the `SearchContext` so it shouldn't require a `SearchContext` to create one.
Relates #46523
This change adds an IndexSearcher and the node's BigArrays in the QueryShardContext.
It's a spin off of #46527 as this change is required to allow aggregation builder to solely use the
query shard context.
Relates #46523
Obviously we have to run the status request again to busy wait for
the `STARTED` state, just busy waiting on an existing response
won't do anything.
Closes#45917
* Add retention to Snapshot Lifecycle Management (#46407)
This commit adds retention to the existing Snapshot Lifecycle Management feature (#38461) as described in #43663. This allows a user to configure SLM to automatically delete older snapshots based on a number of criteria.
An example policy would look like:
```
PUT /_slm/policy/snapshot-every-day
{
"schedule": "0 30 2 * * ?",
"name": "<production-snap-{now/d}>",
"repository": "my-s3-repository",
"config": {
"indices": ["foo-*", "important"]
},
// Newly configured retention options
"retention": {
// Snapshots should be deleted after 14 days
"expire_after": "14d",
// Keep a maximum of thirty snapshots
"max_count": 30,
// Keep a minimum of the four most recent snapshots
"min_count": 4
}
}
```
SLM Retention is run on a scheduled configurable with the `slm.retention_schedule` setting, which supports cron expressions. Deletions are run for a configurable time bounded by the `slm.retention_duration` setting, which defaults to 1 hour.
Included in this work is a new SLM stats API endpoint available through
``` json
GET /_slm/stats
```
That returns statistics about snapshot taken and deleted, as well as successful retention runs, failures, and the time spent deleting snapshots. #45362 has more information as well as an example of the output. These stats are also included when retrieving SLM policies via the API.
* Add base framework for snapshot retention (#43605)
* Add base framework for snapshot retention
This adds a basic `SnapshotRetentionService` and `SnapshotRetentionTask`
to start as the basis for SLM's retention implementation.
Relates to #38461
* Remove extraneous 'public'
* Use a local var instead of reading class var repeatedly
* Add SnapshotRetentionConfiguration for retention configuration (#43777)
* Add SnapshotRetentionConfiguration for retention configuration
This commit adds the `SnapshotRetentionConfiguration` class and its HLRC
counterpart to encapsulate the configuration for SLM retention.
Currently only a single parameter is supported as an example (we still
need to discuss the different options we want to support and their
names) to keep the size of the PR down. It also does not yet include version serialization checks
since the original SLM branch has not yet been merged.
Relates to #43663
* Fix REST tests
* Fix more documentation
* Use Objects.equals to avoid NPE
* Put `randomSnapshotLifecyclePolicy` in only one place
* Occasionally return retention with no configuration
* Implement SnapshotRetentionTask's snapshot filtering and delet… (#44764)
* Implement SnapshotRetentionTask's snapshot filtering and deletion
This commit implements the snapshot filtering and deletion for
`SnapshotRetentionTask`. Currently only the expire-after age is used for
determining whether a snapshot is eligible for deletion.
Relates to #43663
* Fix deletes running on the wrong thread
* Handle missing or null policy in snap metadata differently
* Convert Tuple<String, List<SnapshotInfo>> to Map<String, List<SnapshotInfo>>
* Use the `OriginSettingClient` to work with security, enhance logging
* Prevent NPE in test by mocking Client
* Allow empty/missing SLM retention configuration (#45018)
Semi-related to #44465, this allows the `"retention"` configuration map
to be missing.
Relates to #43663
* Add min_count and max_count as SLM retention predicates (#44926)
This adds the configuration options for `min_count` and `max_count` as
well as the logic for determining whether a snapshot meets this criteria
to SLM's retention feature.
These options are optional and one, two, or all three can be specified
in an SLM policy.
Relates to #43663
* Time-bound deletion of snapshots in retention delete function (#45065)
* Time-bound deletion of snapshots in retention delete function
With a cluster that has a large number of snapshots, it's possible that
snapshot deletion can take a very long time (especially since deletes
currently have to happen in a serial fashion). To prevent snapshot
deletion from taking forever in a cluster and blocking other operations,
this commit adds a setting to allow configuring a maximum time to spend
deletion snapshots during retention. This dynamic setting defaults to 1
hour and is best-effort, meaning that it doesn't hard stop a deletion
at an hour mark, but ensures that once the time has passed, all
subsequent deletions are deferred until the next retention cycle.
Relates to #43663
* Wow snapshots suuuure can take a long time.
* Use a LongSupplier instead of actually sleeping
* Remove TestLogging annotation
* Remove rate limiting
* Add SLM metrics gathering and endpoint (#45362)
* Add SLM metrics gathering and endpoint
This commit adds the infrastructure to gather metrics about the different SLM actions that a cluster
takes. These actions are stored in `SnapshotLifecycleStats` and perpetuated in cluster state. The
stats stored include the number of snapshots taken, failed, deleted, the number of retention runs,
as well as per-policy counts for snapshots taken, failed, and deleted. It also includes the amount
of time spent deleting snapshots from SLM retention.
This commit also adds an endpoint for retrieving all stats (further commits will expose this in the
SLM get-policy API) that looks like:
```
GET /_slm/stats
{
"retention_runs" : 13,
"retention_failed" : 0,
"retention_timed_out" : 0,
"retention_deletion_time" : "1.4s",
"retention_deletion_time_millis" : 1404,
"policy_metrics" : {
"daily-snapshots2" : {
"snapshots_taken" : 7,
"snapshots_failed" : 0,
"snapshots_deleted" : 6,
"snapshot_deletion_failures" : 0
},
"daily-snapshots" : {
"snapshots_taken" : 12,
"snapshots_failed" : 0,
"snapshots_deleted" : 12,
"snapshot_deletion_failures" : 6
}
},
"total_snapshots_taken" : 19,
"total_snapshots_failed" : 0,
"total_snapshots_deleted" : 18,
"total_snapshot_deletion_failures" : 6
}
```
This does not yet include HLRC for this, as this commit is quite large on its own. That will be
added in a subsequent commit.
Relates to #43663
* Version qualify serialization
* Initialize counters outside constructor
* Use computeIfAbsent instead of being too verbose
* Move part of XContent generation into subclass
* Fix REST action for master merge
* Unused import
* Record history of SLM retention actions (#45513)
This commit records the deletion of snapshots by the retention component
of SLM into the SLM history index for the purposes of reviewing operations
taken by SLM and alerting.
* Retry SLM retention after currently running snapshot completes (#45802)
* Retry SLM retention after currently running snapshot completes
This commit adds a ClusterStateObserver to wait until the currently
running snapshot is complete before proceeding with snapshot deletion.
SLM retention waits for the maximum allowed deletion time for the
snapshot to complete, however, the waiting time is not factored into
the limit on actual deletions.
Relates to #43663
* Increase timeout waiting for snapshot completion
* Apply patch
From 2374316f0d.patch
* Rename test variables
* [TEST] Be less strict for stats checking
* Skip SLM retention if ILM is STOPPING or STOPPED (#45869)
This adds a check to ensure we take no action during SLM retention if
ILM is currently stopped or in the process of stopping.
Relates to #43663
* Check all actions preventing snapshot delete during retention (#45992)
* Check all actions preventing snapshot delete during retention run
Previously we only checked to see if a snapshot was currently running,
but it turns out that more things can block snapshot deletion. This
changes the check to be a check for:
- a snapshot currently running
- a deletion already in progress
- a repo cleanup in progress
- a restore currently running
This was found by CI where a third party delete in a test caused SLM
retention deletion to throw an exception.
Relates to #43663
* Add unit test for okayToDeleteSnapshots
* Fix bug where SLM retention task would be scheduled on every node
* Enhance test logging
* Ignore if snapshot is already deleted
* Missing import
* Fix SnapshotRetentionServiceTests
* Expose SLM policy stats in get SLM policy API (#45989)
This also adds support for the SLM stats endpoint to the high level rest client.
Retrieving a policy now looks like:
```json
{
"daily-snapshots" : {
"version": 1,
"modified_date": "2019-04-23T01:30:00.000Z",
"modified_date_millis": 1556048137314,
"policy" : {
"schedule": "0 30 1 * * ?",
"name": "<daily-snap-{now/d}>",
"repository": "my_repository",
"config": {
"indices": ["data-*", "important"],
"ignore_unavailable": false,
"include_global_state": false
},
"retention": {}
},
"stats": {
"snapshots_taken": 0,
"snapshots_failed": 0,
"snapshots_deleted": 0,
"snapshot_deletion_failures": 0
},
"next_execution": "2019-04-24T01:30:00.000Z",
"next_execution_millis": 1556048160000
}
}
```
Relates to #43663
* Rewrite SnapshotLifecycleIT as as ESIntegTestCase (#46356)
* Rewrite SnapshotLifecycleIT as as ESIntegTestCase
This commit splits `SnapshotLifecycleIT` into two different tests.
`SnapshotLifecycleRestIT` which includes the tests that do not require
slow repositories, and `SLMSnapshotBlockingIntegTests` which is now an
integration test using `MockRepository` to simulate a snapshot being in
progress.
Relates to #43663Resolves#46205
* Add error logging when exceptions are thrown
* Update serialization versions
* Fix type inference
* Use non-Cancellable HLRC return value
* Fix Client mocking in test
* Fix SLMSnapshotBlockingIntegTests for 7.x branch
* Update SnapshotRetentionTask for non-multi-repo snapshot retrieval
* Add serialization guards for SnapshotLifecyclePolicy
Today we load the metadata from disk while constructing the node. However there
is no real need to do so, and this commit moves that code to run later while
the node is starting instead.
Registering two different http methods on the same path using different
wildcard names would result in the last wildcard name being active only.
Now throw an exception instead.
Closes#46482
The cuckoofilters could be randomly created with too small of
capacity or precision, which means that they can only absorb a few
values before collisions start to make all filters look identical.
This increases the size of filters we generate (capacity >> than
the test cases) and lower fpp rate.
Today when the membership of the cluster changes we log messages that describe
the change like this:
added {{node-1}{OPdaTIGmSxaEXXOyg3o96w}{127.0.0.1}{127.0.0.1:9301}{di},}
The trailing comma suggests there is some missing string that might contain
extra information, but in fact it's an artefact of how these messages are
constructed. This commit removes the trailing comma from these lists.
I couldn't find a test for this, as it seems we only get
into this error handler on a bug. Regardless, we are
executing the snapshot finalization on the master update
thread here which shouldn't happen and will make
debugging a production issue resulting from this
trickier than it has to be (because we probably also
get a cluster state apply is slow warning in addition
to the original bug).
Used the generic pool here instead of the snapshot pool
because we're resolving the user callback here as
well and the generic pool seemed like the safer bet for
that.
Previously, we ignore replication for noop updates because they do not
have sequence numbers. Since #44603, we started assigning sequence
numbers to noop updates leading them to be replicated to replicas.
This bug occurs only on 8.0 for it requires #41065 and #44603.
Closes#46366
Following performance optimisations to the adjacency_matrix aggregation we no longer require this setting. Marked as deprecated and due for removal in 8.0
Related #46324
Yannick Welsch
Jim Ferenczi
Nhat Nguyen
Gordon Brown
David Turner
Armin Braun
Colin Goodheart-Smithe
Howard
Ignacio Vera
Jason Tedor
Gaurav614
Przemyslaw Gomulka
Martijn van Groningen
Rory Hunter
David Roberts
Tanguy Leroux
Tim Brooks
Lee Hinman
Daniel Mitterdorfer
Julie Tibshirani
Eray
Christos Soulios
Andrei Dan
Zachary Tong
Igor Motov
Ryan Ernst
William Brafford
Christoph Büscher
Mayya Sharipova
Henning Andersen
markharwood