Because concurrent sync requests from a primary to its replicas could be
in flight, it can be the case that an older retention leases collection
arrives and is processed on the replica after a newer retention leases
collection has arrived and been processed. Without a defense, in this
case the replica would overwrite the newer retention leases with the
older retention leases. This commit addresses this issue by introducing
a versioning scheme to retention leases. This versioning scheme is used
to resolve out-of-order processing on the replica. We persist this
version into Lucene and restore it on recovery. The encoding of
retention leases is starting to get a little ugly. We can consider
addressing this in a follow-up.
The semantics of the API changed considerably since the documentation was written.
The main change is to remove references to memory reduction (this is related to refresh).
Instead, flush refers to recovery times. I also removed the references to trimming the translog
as the translog may be required for other purposes (operation history for ops based recovery
and complement ongoing file based recoveries).
Closes#32869
This commit changes the default out-of-the-box configuration for the
number of shards from five to one. We think this will help address a
common problem of oversharding. For users with time-based indices that
need a different default, this can be managed with index templates. For
users with non-time-based indices that find they need to re-shard with
the split API in place they no longer need to resort only to
reindexing.
Since this has the impact of changing the default number of shards used
in REST tests, we want to ensure that we still have coverage for issues
that could arise from multiple shards. As such, we randomize (rarely)
the default number of shards in REST tests to two. This is managed via a
global index template. However, some tests check the templates that are
in the cluster state during the test. Since this template is randomly
there, we need a way for tests to skip adding the template used to set
the number of shards to two. For this we add the default_shards feature
skip. To avoid having to write our docs in a complicated way because
sometimes they might be behind one shard, and sometimes they might be
behind two shards we apply the default_shards feature skip to all docs
tests. That is, these tests will always run with the default number of
shards (one).
As follow up to #28245 , this PR removes the logic for selecting the
right start commit from the Engine constructor in favor of explicitly
trimming them in the Store, before the engine is opened. This makes the
constructor in engine follow standard Lucene semantics and use the last
commit.
Relates #28245
Relates #29156
#28245 has introduced the utility class`EngineDiskUtils` with a set of methods to prepare/change
translog and lucene commit points. That util class bundled everything that's needed to create and
empty shard, bootstrap a shard from a lucene index that was just restored etc.
In order to safely do these manipulations, the util methods acquired the IndexWriter's lock. That
would sometime fail due to concurrent shard store fetching or other short activities that require the
files not to be changed while they read from them.
Since there is no way to wait on the index writer lock, the `Store` class has other locks to make
sure that once we try to acquire the IW lock, it will succeed. To side step this waiting problem, this
PR folds `EngineDiskUtils` into `Store`. Sadly this comes with a price - the store class doesn't and
shouldn't know about the translog. As such the logic is slightly less tight and callers have to do the
translog manipulations on their own.
Currently we have a fairly complicated logic in the engine constructor logic to deal with all the
various ways we want to mutate the lucene index and translog we're opening.
We can:
1) Create an empty index
2) Use the lucene but create a new translog
3) Use both
4) Force a new history uuid in all cases.
This leads complicated code flows which makes it harder and harder to make sure we cover all the
corner cases. This PR tries to take another approach. Constructing an InternalEngine always opens
things as they are and all needed modifications are done by static methods directly on the
directory, one at a time.
Currently the Translog constructor is capable both of opening an existing translog and creating a
new one (deleting existing files). This PR separates these two into separate code paths. The
constructors opens files and a dedicated static methods creates an empty translog.
The new ops based recovery, introduce as part of #10708, is based on the assumption that all operations below the global checkpoint known to the replica do not need to be synced with the primary. This is based on the guarantee that all ops below it are available on primary and they are equal. Under normal operations this guarantee holds. Sadly, it can be violated when a primary is restored from an old snapshot. At the point the restore primary can miss operations below the replica's global checkpoint, or even worse may have total different operations at the same spot. This PR introduces the notion of a history uuid to be able to capture the difference with the restored primary (in a follow up PR).
The History UUID is generated by a primary when it is first created and is synced to the replicas which are recovered via a file based recovery. The PR adds a requirement to ops based recovery to make sure that the history uuid of the source and the target are equal. Under normal operations, all shard copies will stay with that history uuid for the rest of the index lifetime and thus this is a noop. However, it gives us a place to guarantee we fall back to file base syncing in special events like a restore from snapshot (to be done as a follow up) and when someone calls the truncate translog command which can go wrong when combined with primary recovery (this is done in this PR).
We considered in the past to use the translog uuid for this function (i.e., sync it across copies) and thus avoid adding an extra identifier. This idea was rejected as it removes the ability to verify that a specific translog really belongs to a specific lucene index. We also feel that having a history uuid will serve us well in the future.
The response is attempting to illustrate the sync_id marker, but in
the test the index is too "fresh" to have a sync marker. So the test
needs to execute a sync flush behind the scenes so that the marker
is present
Today when parsing a stats request, Elasticsearch silently ignores
incorrect metrics. This commit removes lenient parsing of stats requests
for the nodes stats and indices stats APIs.
Relates #21417
Adds infrastructure so `gradle :docs:check` will extract tests from
snippets in the documentation and execute the tests. This is included
in `gradle check` so it should happen on CI and during a normal build.
By default each `// AUTOSENSE` snippet creates a unique REST test. These
tests are executed in a random order and the cluster is wiped between
each one. If multiple snippets chain together into a test you can annotate
all snippets after the first with `// TEST[continued]` to have the
generated tests for both snippets joined.
Snippets marked as `// TESTRESPONSE` are checked against the response
of the last action.
See docs/README.asciidoc for lots more.
Closes#12583. That issue is about catching bugs in the docs during build.
This catches *some* bugs in the docs during build which is a good start.
To better distribute the memory allocating to indexing, the IndexingMemoryController periodically checks the different shard for their last indexing activity. If no activity has happened for a while, the controller marks the shards as in active and allocated it's memory buffer budget (but a small minimal budget) to other active shards. The recently added synced flush feature (#11179, #11336) uses this inactivity trigger to attempt as a trigger to attempt adding a sync id marker (which will speed up future recoveries).
We wait for 30m before declaring a shard inactive. However, these days the operation just requires a refresh and is light. We can be stricter (and 5m) increase the chance a synced flush will be triggered.
Closes#11479
#10032 introduced the notion of sealing an index by marking it with a special read only marker, allowing for a couple of optimization to happen. The most important one was to speed up recoveries of shards where we know nothing has changed since they were online by skipping the file based sync phase. During the implementation we came up with a light notion which achieves the same recovery benefits but without the read only aspects which we dubbed synced flush. The fact that it was light weight and didn't put the index in read only mode, allowed us to do it automatically in the background which has great advantage. However we also felt the need to allow users to manually trigger this operation.
The implementation at #11179 added the sync flush internal logic and the manual (rest) rest API. The name of the API was modeled after the sealing terminology which may end up being confusing. This commit changes the API name to match the internal synced flush naming, namely `{index}/_flush/synced'.
On top of that it contains a couple other changes:
- Remove all java client API. This feature is not supposed to be called programtically by applications but rather by admins.
- Improve rest responses making structure similar to other (flush) API
- Change IndexShard#getOperationsCount to exclude the internal +1 on open shard . it's confusing to get 1 while there are actually no ongoing operations
- Some minor other clean ups
The `full` option and `FlushType.NEW_WRITER` only exists to allow
realtime changes to two settings (`index.codec` and `index.concurrency`).
Those settings are very expert and don't really need to be updateable
in realtime.
This commit adds the ability to force blocking on the flush operaition
to make sure all files have been written and synced to disk. Without
this option a flush might be executing at the same time causing the
current flush to fail and return before all files being synced.
Closes#6996
* Clean up s/ElasticSearch/Elasticsearch on docs/*
* Clean up s/ElasticSearch/Elasticsearch on src/* bin/* & pom.xml
* Clean up s/ElasticSearch/Elasticsearch on NOTICE.txt and README.textile
Closes#4634
Refresh flag in flush is problematic, since the shards refresh is allowed to execute on is different compared to the flush shards. In order to do flush and then refresh, they should be executed as separate APIs when needed.
closes#3689
Jason Tedor
Boaz Leskes
Josh Soref
Nhat Nguyen
Zachary Tong
Zachary Tong
Nik Everett
Clinton Gormley
Dongjoon Hyun
Simon Willnauer
Shay Banon