Today index names are often resolved lazily, only when they are really
needed. This can be problematic especially when it gets to mapping updates
etc. when a node sends a mapping update to the master but while the request
is in-flight the index changes for whatever reason we would still apply the update
since we use the name of the index to identify the index in the clusterstate.
The problem is that index names can be reused which happens in practice and sometimes
even in a automated way rendering this problem as realistic.
In this change we resolve the index including it's UUID as early as possible in places
where changes to the clusterstate are possible. For instance mapping updates on a node use a
concrete index rather than it's name and the master will fail the mapping update iff
the index can't be found by it's <name, uuid> tuple.
Closes#17048
Changes:
- no more option to configure eager/lazy loading of the norms (useless now
that orms are disk-based)
- only the `string`, `text` and `keyword` fields support the `norms` setting
- the `norms` setting takes a boolean that decides whether norms should be
stored in the index but old options are still supported to give users time
to upgrade
- setting a `boost` no longer implicitely enables norms (for new indices only,
this is still needed for old indices)
Today, certain bootstrap properties are set and read via system
properties. This action-at-distance way of managing these properties is
rather confusing, and completely unnecessary. But another problem exists
with setting these as system properties. Namely, these system properties
are interpreted as Elasticsearch settings, not all of which are
registered. This leads to Elasticsearch failing to startup if any of
these special properties are set. Instead, these properties should be
kept as local as possible, and passed around as method parameters where
needed. This eliminates the action-at-distance way of handling these
properties, and eliminates the need to register these non-setting
properties. This commit does exactly that.
Additionally, today we use the "-D" command line flag to set the
properties, but this is confusing because "-D" is a special flag to the
JVM for setting system properties. This creates confusion because some
"-D" properties should be passed via arguments to the JVM (so via
ES_JAVA_OPTS), and some should be passed as arguments to
Elasticsearch. This commit changes the "-D" flag for Elasticsearch
settings to "-E".
To make API's output more easy to read we are suppressing stack traces (#12991) unless explicitly requested by setting `error_trace=true` on the request. To compensate we are logging the stacktrace into the logs so people can look it up even the error_trace wasn't enabled. Currently we do so using the `INFO` level which can be verbose if an api is called repeatedly by some automation. For example, if someone tries to read from an index that doesn't exist we will respond with a 404 exception and log under info every time. We should reduce the level to `DEBUG` as we do with other API driven errors. Internal errors (rest codes >=500) are logged as WARN.
Closes#16627
This change adds the infrastructure to run the rest tests on a multi-node
cluster that users 2 different minor versions of elasticsearch. It doesn't implement
any dedicated BWC tests but rather leverages the existing REST tests.
Since we don't have a real version to test against, the tests uses the current version
until the first minor / RC is released to ensure the infrastructure works.
Relates to #14406Closes#17072
When unzipping a plugin zip, the zip entries are resolved relative to
the directory being unzipped into. However, there are currently no
checks that the entry name was not absolute, or relatively points
outside of the plugin dir. This change adds a check for those two cases.
This adds methods and tests to ScriptSortBuilder that
makes it implement NamedWritable and adds the fromXContent
method needed to read itseld from xContent.
Test failures showed problems with passing down
the same collate parameter map reference from the
phrase suggestion builder to the context where.
This changes the collate parameter setters to
make a shallow copy of the map passed in.
This commit adds fields bytes_recovered and files_recovered to the cat
recovery API. These fields, respectively, indicate the total number of
bytes and files recovered. Additionally, for consistency, some totals
fields and translog recovery fields have been renamed.
Closes#17064
This tries to remove friction to upgrade to 5.0 that would be caused by mapping
changes:
- old ways to specify mapping settings (eg. store: yes instead of store:true)
will still work but a deprecation warning will be logged
- string mappings that only use the most common options will be upgraded
automatically to text/keyword
The change to move dynamic mapping handling to the end of document
parsing has an edge case which can cause dynamic mappings to fail
document parsing. If field a.b is added as an as part of the root update,
followed by a.c.d, then we need to expand the mappers on the stack,
since a is hidden inside the root update which exists on the stack.
This change adds a test for this case, as well as tries to better
document how the logic works for building up the stack before adding a
dynamic mapper.
Sequence of events that lead to the NPE:
- avg metric returns NaN for buckets
- Movavg skips NaN or null buckets, and simply re-uses the existing bucket (e.g. doesn't add
a 'movavg' field)
- Derivative references Movavg, the bucket resolution returns null because Movavg wasn't added
to the bucket, NPE when trying to subtract null values
The ingest stats include the following statistics:
* `ingest.total.count`- The total number of document ingested during the lifetime of this node
* `ingest.total.time_in_millis` - The total time spent on ingest preprocessing documents during the lifetime of this node
* `ingest.total.current` - The total number of documents currently being ingested.
* `ingest.total.failed` - The total number ingest preprocessing operations failed during the lifetime of this node
Also these stats are returned on a per pipeline basis.
Currently, the cluster service is tightly coupled to the transport service by both managing node connections and requiring the bound address in order to create the local disco node. This commit introduces a new NodeConnectionsService which is in charge of node connection management and makes it possible to remove all network related calls from the cluster service. The local DiscoNode is now created by DiscoveryNodeService and is set both the cluster service and the transport service during node start up.
Closes#16788Closes#16872
