Now that we're consistently using `cat_match` to filter which shards we
run on we can get this confusing case:
1. You have a search with, say, a range and a sub-agg.
2. That search has a query that `can_match` can recognize will match no
docs. On *any* shard.
3. So we dutifully run it on a single shard so it can produce the
"empty" aggs.
4. The shard we pick happens to not have the target of the range mapped.
5. This kicks in the special range aggregator that doesn't collect any
documents.
6. Before this commit, that range aggregator *also* never produced any
sub-aggs.
So, without this change, it was quite possible for a search that
happened to match no documents to "throw away" the sub-aggs of a range
and a few other aggs.
We've had this problem for a long, long time but it is more confusing
now because `can_match` is really kicking in and causing us to see cases
where it looks like you are targeting a lot of shards but you really are
only targeting a couple. It used to be that to get the "no sub-aggs"
behavior you had to explicitly target only shards that didn't map the
target field of the `range` agg. And, like, in that case it isn't too
bad because you targeted a sort of degenerate shard. But now that
`can_match` is doing its thing you can end up with the confusing steps
above. It took me several hours to track down what what happening I know
how the individual pieces of all of this works. It took four hours to
figure out how they fit together in this case....
Anyway! This replaces all the aggregator implementations that throw out
the sub-aggregators with ones that keep them. I think this'll be less
confusing in the future.
Closes#64142
This commit adds logging to indicate whether or not we are using the
bundled JDK. We distinguish between using a distribution that bundles
the JDK versus using a distribution that does not bundle the JDK.
In 7.x we can't just by default generate this setting as it might not be
supported by data nodes that are assigned shards for an older version in mixed version
clusters.
Closes#64152
This commit fixes an issue with the detection on macOS for whether or
not the bundled JDK is being used. The logic between macOS and non-macOS
is different because the JDK has a different directory structure on
macOS versus non-macOS. However, due to notarization issues, we changed
the top-level directory from jdk to jdk.app, yet never updated this
detection logic to account for that.
Ideally, we would have a packaging test that asserts that we have the
behavior here correct, and it maintains over time. Alas, we do not
currently have packaging tests on macOS.
With this change, we will always return the same point in time in a
search response as its input until we implement the retry mechanism
for the point in times.
The formatting of the global bottom value does not take the resolution of the provided
numeric_type into account. This change fixes this bug by providing the resolution
directly in the doc value format if the numeric_type is provided as `date_nanos`.
Closes#63719
We must not remove the snapshot from the initializing set
in the `timeout` getter. This was a plain oversight/mistake
and went unnoticed. It can lead to the removal of a valid
snapshot clone from the cluster state in rare circumstances
(e.g. when a node concurrently joins the cluster or a routing
change happens as it did in the linked test failure).
Closes#64115
If we run into a background merge between creating the snapshot and closing the index
then with compound files we could be in a situation where we get zero file reuse
on restore.
Force merging before the snapshot gives us a single segment that won't change down the line
so the restore always sees file reuse from the close index.
Closes#63476
Assuming the clone failed when the request failed is not sufficient.
There are failure modes where the request fails but the clone still works out
because the data node resent the requeest after the first clone had already been
failed and removed from the cluster state when master was restarted.
Closes#63473
We have to wait for no more operations here not for `1`. This mostly worked
because the test thread would add the listener quickly enough so that it sees the
state where either the snapshot or clone but not both have already finished
but randomly the test thread would be slow and time out on a state without snaphots in it.
testHealthOnMasterFailover could timeout on some of the health requests
in the case where an index is added, since the recovery leads to
extended test run time.
Closes#62690
We had and an error when serializing fully reduced scripted metrics.
Small typo and sever lack of tests..... Anyway, this fixed the one
character typo and adds a bunch more tests.
This commit adds a test in DiskThresholdDeciderTests that verifies
the allocation of a snapshot recovery source based shard in the
situation where the snapshot shard size was successfully provided
by the SnapshotInfoService introduced in #61906 and when the
service failed to provide the size.
Relates #61906
In #57892 I broke *some* sub-aggregations inside of the `parent` and
`child` aggregator, specifically any sub-aggregations that do work in
the `postCollect` phase. This fixes it by delaying the post collect
phase of aggs under `parent` and `child` until `beforeBuildingBuckets`
because, well, we haven't done *any* collection until after that phase.
Currently if distance_feature query contains boost,
it incorrectly gets applied twice: in AbstractQueryBuilder::toQuery and
we also pass this boost to Lucene's LongPoint.newDistanceFeatureQuery.
As a result we get incorrect scores.
This fixes this error to ensure that boost is applied only once.
Closes#63691
This commit fixes the UpdateThreadPoolSettingsTests to be aware of the
hard limit on the maximum size of the system_write executor. This
executor has a hard limit that matches the write executor, which is
the number of allocated processors.
Closes#63131
Backport #63700
Today indexing to a shard with 2147483519 documents will fail that
shard. We should check the number of documents and reject the write
requests instead.
Closes#51136
This fixes a gap in testing and a bug that can occur in various forms:
When we would start a snapshot or clone related to a shard that was done
snapshotting/cloning but its overall operation was not yet finalized
at the time of starting the operation, we would base the operation off of
the wrong generation. This would not cause a corrupted repo, but would
cause the operation to be `PARTIAL`.
This commit fixes the state machine to take into account the correct generation
in this case.
Closes#63498
This PR implements value fetching for the following field types:
* `text` phrase and prefix subfields
* `search_as_you_type`, plus its subfields
* `token_count`, which is implemented by fetching doc values
Supporting these types helps ensure that retrieving all fields through
`"fields": ["*"]` doesn't fail because of unsupported value fetchers.
Currently we flush the Translog buffer when a new operation causes the
buffer to breach 1MB. This introduces a scenario where an exception is
thrown AFTER the writer has accepted the operation. To avoid this, this
commit flushes the Translog in an #add call before adding a new
operation.
This fixes#63299.
This PR adds factory methods for the most common implementations:
* `SourceValueFetcher.identity` to pass through the source value untouched.
* `SourceValueFetcher.toString` to simply convert the source value to a string.
#63214 made TypeFieldType a constant field, and fixed things so that it always
emits deprecation warnings whenever it is referenced in a query or aggregation.
However, it also emits warnings when it is used to build a type filter through
the search context; this is unnecessary, as warnings are already emitted by
the REST layer when types are specified as part of the URL, and it is causing
failures in some BWC tests.
This commit adds a specialised typeFilter method to TypeFieldType to handle
this case without emitted any extra warnings. It also removes an unused duplicate
TypeFieldType class that resulted from a backport merge error.
Fixes#63366
As a result of this, we can remove a chunk of code from TypeParsers as well. Tests
for search/index mode analyzers have moved into their own file. This commit also
rationalises the serialization checks for parameters into a single SerializerCheck
interface that takes the values includeDefaults, isConfigured and the value
itself.
Relates to #62988
We were not consistent in checking for node roles before adding listeners.
In some cases we did check the necessity of a CS listener and in others we did not.
This commit fixes a number of cases of redundant listeners that don't apply to all node roles.
In #61906 we agreed on always providing the default value
ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE
when the SnasphotInfoService failed to retrieve the exact
size for a given snapshot shard. The motivation was to
allow the shard allocation to move forward in case of
failures (so that the unassigned shard does not get stuck
in an unassigned state for too long) while relying on the
fallback values for shard sizes.
Sadly a bug in the
SnapshotShardSizeInfo#getShardSize(ShardRouting, long)
makes the default value to be ignored when the snapshot
shard size retrieval previously failed, returning
ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE
instead of the provided default value. With DiskThresholdDecider
also not relying on the provided default value this triggers
some assertion like in #63376 which helped us to spot the bug.
Closes ##63376
The first refreshDiskUsage() refreshes the ClusterInfo update which in turn
calls listeners like DiskThreshMonitor. This one triggers a reroute as
expected and turns an internal checkInProgress flag before submitting
a cluster state update to relocate shards (the internal flag is toggled
again once the cluster state update is processed).
In the test I suspect that the second refreshDiskUsage() may complete
before DiskThreshMonitor's internal flag is set back to its initial state,
resulting in the second ClusterInfo update to be ignored and message
like "[node_t0] skipping monitor as a check is already in progress" to
be logged. Adding another wait for languid events to be processed
before executing the second refreshDiskUsage() should help here.
Closes#62326
Nik Everett
Jason Tedor
Armin Braun
Nhat Nguyen
Jim Ferenczi
Marios Trivyzas
Przemyslaw Gomulka
markharwood
Henning Andersen
Ignacio Vera
Tanguy Leroux
Mayya Sharipova
Ioannis Kakavas
Jay Modi
James Rodewig
Lee Hinman
Julie Tibshirani
Tim Brooks
Dan Hermann
Christoph Büscher
Alan Woodward