This commit enables concurrent search execution in the DFS phase, which is going to improve resource usage as well as performance of knn queries which benefit from both concurrent rewrite and collection.
We will enable concurrent execution for the query phase in a subsequent commit. While this commit does not introduce parallelism for the query phase, it introduces offloading sequential computation to the newly introduced executor. This is true both for situations where a single slice needs to be searched, as well as scenarios where a specific request does not support concurrency (currently only DFS phase does regardless of the request). Sequential collection is not offloaded only if the request includes aggregations that don't support offloading: composite, nested and cardinality as their post collection method must be executed in the same thread as the collection or we'll trip a lucene assertion that verifies that doc_values are pulled and consumed from the same thread.
## Technical details
This commit introduces a secondary executor, used exclusively to execute the concurrent bits of search. The search threads are still the ones that coordinate the search (where the caller search will originate from), but the actual work will be offloaded to the newly introduced executor.
We are offloading not only parallel execution but also sequential execution, to make the workload more predictable, as it would be surprising to have bits of search executed in either of the two thread pools. Also, that would introduce the possibility to suddenly run a higher amount of heavy operations overall (some in the caller thread and some in the separate threads), which could overload the system as well as make sizing of thread pools more difficult.
Note that fetch, together with other actions, is still executed in the search thread pool. This commit does not make the search thread pool merely a coordinating only thread pool, It does so only for what concerns the IndexSearcher#search operation itself, which is though a big portion of the different phases of search API execution.
Given that the searcher blocks waiting for all tasks to be completed, we take a simple approach of introducing a thread pool executor that has the same size as the existing search thread pool but relies on an unbounded queue. This simplifies handling of thread pool queue and rejections. In fact, we'd like to guarantee that the secondary thread pool won't reject, and delegate queuing entirely to the search thread pool which is the entry point for every search operation anyway. The principle behind this is that if you got a slot in the search thread pool, you should be able to complete your search, and rather quickly.
As part of this commit we are also introducing the ability to cancel tasks that have not started yet, so that if any task throws an exception, other tasks are prevented from starting needless computation.
Relates to #80693
Relates to #90700
Today by default the `SEARCH_COORDINATION` pool is sized at half the
allocated processors, or five if there are more than ten CPUs. Yet, if
we scale up a node to have more than ten CPUs, we probably want to scale
up the number of search coordination threads to match. This commit
removes the limit of five threads.
Discovery, like cluster membership, can also be affected by network-like
issues (e.g. GC/VM pauses, dropped packets and blocked threads) so this
commit duplicates the troubleshooting info across both places.
A completely idle `transport_worker` thread is reported as `0.0%` idle,
which is confusing. Moreover the docs on the network threading model do
not reflect the changes made in #90482. This commit fixes both of those
things.
Suggest calling `jstack` every 15s to ensure that at least one capture
shows a stuck thread. Also adds a link to this guide to the list on the
troubleshooting overview page.
Explains why you should remove `cluster.initial_master_nodes`, and
rewords some of the other sections a little for (subjectively) improved
readability.
* Fixes CORS headers needed by Elastic clients
Updates the default value for the `http.cors.allow-headers`
setting to include headers used by Elastic client libraries.
Also adds the `access-control-expose-headers` header to responses to
CORS requests so that clients can successfully perform their product
check.
In #92309 we have aligned the size of the `search` and the `get` thread
pool but the docs still contain the prior `get` thread pool size. With
this commit we also align the docs.
Relates #92309
We sometimes see a `ShardLockObtainFailedException` when a shard failed
to shut down as fast as we expected, often because a node left and
rejoined the cluster. Sometimes this is because it was held open by
ongoing scrolls or PITs, but other times it may be because the shutdown
process itself is too slow. With this commit we add the ability to
capture and log a thread dump at the time of the failure to give us more
information about where the shutdown process might be running slowly.
Relates #93226
If debug logging is enabled then the lag detector will capture and
report the hot threads of a lagging node. In some cases the resulting
log message can be very large, exceeding 10kiB, which means it is
truncated in most logging setups. The relevant thread(s) may be waiting
on I/O, which is not considered "hot" and therefore may not appear in
the first 10kiB.
This commit adjusts this logging mechanism to split the message into
chunks of size at most 2kiB (after compression and base64-encoding) to
ensure that the entire hot threads output can be faithfully
reconstructed from these logs.
Closes#88126
Today to troubleshoot an unstable cluster we ask the users to parse the
rather complex `node-join` and `node-left` messages emitted by the
`MasterService`. These messages may refer to many nodes, may be
truncated, and are generally pretty hard to work with.
With this commit we start to emit a simplified log message about each
node added and removed. It also renames the respective executor classes:
- `JoinTaskExecutor` -> `NodeJoinExecutor`
- `NodeRemovalClusterStateTaskExecutor` -> `NodeLeftExecutor`
This brings their names in line with each other, and the messages that
they emit, whilst preserving the older `node-join` and `node-left`
terminology as reported by the `MasterService`.
Finally, it updates the troubleshooting logs to reflect these new and
simplified logs.
Relates #92741
* Update search-settings documentation to reflect the fact that the indices.query.bool.max_clause_count setting has been deprecated
* Fix indentation
* Replace Elasticsearch with {es}
* Add deprecation entry to release notes
Co-authored-by: Abdon Pijpelink <abdon.pijpelink@elastic.co>
Currently the documentation on network threading suggests that we still
use a model where we have individual workers dedicated to server
sockets. That is no longer true and server sockets are assigned to
normal workers. This commit updates the documentation.
* Update threadpool.asciidoc
Starting from 8.0 the value of the `node.processors` setting is bounded by the number of available
processors https://github.com/elastic/elasticsearch/pull/44894
* Update docs/reference/modules/threadpool.asciidoc
Co-authored-by: Adam Locke <adam.locke@elastic.co>
When parsing queries on the coordinating node, there is currently no way to share state between the different parsing methods (`fromXContent`). The only query that supports a parse context is bool query, which uses the context to track nested depth of queries, added with #66204. Such nested depth tracking mechanism is not 100% accurate as it tracks bool queries only, while there's many more query types that can hold other queries hence potentially cause stack overflow when deeply nested.
This change removes the parsing context that's specific to bool query, introduced with #66204, in favour of generalizing the nested depth tracking to all query types.
The generic tracking is introduced by wrapping the parser and overriding the method that parses named objects through the xcontent registry. Another way would have been to require a context argument when parsing queries, which would mean adding a context argument to all the QueryBuilder#fromXContent static methods. That would be a breaking change for plugins that provide custom queries, hence I went for trying out a different approach.
One aspect that this change requires and introduces is the distinction between parsing a top level query (which will wrap the parser, or it would create the context if we had one), as opposed to parsing an inner query, which goes ahead with the given parser and context. We already have this distinction as we have two different static methods in `AbstractQueryBuilder` but in practice only bool query makes the distinction being the only context-aware query.
In addition to generalizing tracking nested depth when parsing queries, we should be able to adopt this same strategy to track queries usage as part #90176 .
Given that the depth check is now more restrictive, as it counts all compound queries and not only bool, we have decided to raise the default limit to `30` to ensure that users are not going to hit the limit due to this change.
Adds to the docs a note that the `100mb` default for
`http.max_content_length` is the recommended maximum, along with
suggestions for what to do when hitting this limit.
Introduce max headroom settings for the low, high, and flood disk watermark stages, similar to the existing max headroom setting for the flood stage of the frozen tier. Introduce new max headrooms in HealthMetadata and in ReactiveStorageDeciderService. Add multiple tests in DiskThresholdDeciderUnitTests, DiskThresholdDeciderTests and DiskThresholdMonitorTests. Moreover, addition & subtraction for ByteSizeValue, and min.
This commit adds support for floating point node.processors setting.
This is useful when the nodes run in an environment where the CPU
time assigned to the ES node process is limited (i.e. using cgroups).
With this change, the system would be able to size the thread pools
accordingly, in this case it would round up the provided setting
to the closest integer.
The docs for `transport.ping_schedule` note that the transport client
defaults to a 5s ping schedule, but this is no longer relevant. This
commit drops this from the docs, and also moves the docs for this
setting further down the page to reflect its relative unimportance.
Today we say that voting-only nodes require a "low-latency" network.
This term has a specific meaning in some operating environments which is
different from our intended meaning. To avoid this confusion this commit
removes the absolute term "low-latency" in favour of describing the
requirements relative to the user's own performance goals.
Clean up network setting docs
- Add types for all params
- Remove mention of JDKs before 11
- Clarify some wording
Co-authored-by: Stef Nestor <steffanie.nestor@gmail.com>
This change ensures that existing read_only_allow_delete blocks that
are placed on indices when the flood_stage watermark threshold is
exceeded, are removed when the disk threshold monitoring is disabled.
This is done by changing how InternalClusterInfoService behaves when
disabled. With this change, it will keep calling the registered
listeners periodically, but with an empty ClusterInfo.
Closes#86383
Our current default for the http.max_header_size setting is 8kb. This
is lower than the current default for Kibana (16kb in 8.x), and the ESS
proxy (1mb based on the Go http library default). To align with the
current convention of other Elastic components, this PR increases the
ES header size setting default to 16kb.
Closes#88501
* Convert disk watermarks to RelativeByteSizeValues
Similar to the existing watermark setting for the frozen tier.
Pre-requisite for PR 88639 that plans to introduce max headroom
settings for the disk watermarks, similar to the frozen tier max
headroom setting.
* Add changelog
* Revert 20gb to 20GB
* Make formatNoTrailingZerosPercent non static
* ByteSizeValue.MINUS_ONE
* Remove getMinimumTotalSizeForBelowWatermark
* Remove comment
* Fix minor stuff
* Make parsing of RelativeByteSizeValue faster
Mimicks older definitelyNotPercentage function
* Remove Locale from Strings.format
* More MINUS_ONE
* Adding discovery troubleshooting link
* Add tags to pull in discovery troubleshooting content
* Move discovery troubleshooting to separate page and add redirects
Co-authored-by: Adam Locke <adam.locke@elastic.co>
In #85074 we added docs on discovery troubleshooting that really only
talked about troubleshooting master elections. There's also the case
where the master is elected fine but some other node can't join it. This
commit adds troubleshooting docs about that too.
Co-authored-by: Adam Locke <adam.locke@elastic.co>
