Adding a small value to the threshold prevents weight deltas that are
very very close to the threshold to not trigger relocations. These
deltas can be rounding errors that lead to unnecessary relocations. In
practice this might only happen under very rare circumstances.
In general it's a good idea for the shard allocator to be a bit
more conversavtive in terms of rebalancing since in general relocation
costs are pretty high.
Closes#4630
Norms can be eagerly loaded on a per-field basis by setting norms.loading to
`eager` instead of the default `lazy`:
```
"my_string_field" : {
"type": "string",
"norms": {
"loading": "eager"
}
}
```
In case this behavior should be applied to all fields, it is possible to change
the default value by setting `index.norms.loading` to `eager`.
Close#4079
First, this breaks backwards compatibility!
* Removed /_cluster/nodes/stats endpoint
* Excpect the stats types not as parameters, but as part of the URL
* Returning all indices stats by default, returning all nodes stats by default
* Supporting groups & types in nodes stats now as well
* Updated documentation & tests accordingly
* Allow level parameter for "shards" and "indices" (cluster does not make sense here)
Closes#4057
Note: This breaks backward compatibility
* Removed clear/all parameters, now all stats are returned by default
* Made the metrics part of the URL
* Removed a lot of handlers
* Added shards/indices/cluster level paremeter to change response serialization
* Returning translog statistics in IndicesStats
* Added TranslogStats class
* Added IndexShard.translogStats() method to get the stats from concrete implementation
* Updated documentation
Closes#4054
The reason to not start packages on installation is to allow to configure
them before starting up (setting heap, cluster.name etc)
Also the documentation was updated in order to show, which statements need
to be executed.
In addition, these statements are also printed out when the package is
installed, depending on whether chkconfig, system or update-rc.d is used.
Closes#3722
When testing plugin manager with real downloads, it could happen that the test run forever. Fortunately, test suite will be interrupted after 20 minutes, but it could be useful not to fail the whole test suite but only warn in that case.
By default, plugin manager still wait indefinitely but it can be modified using new `--timeout` option:
```sh
bin/plugin --install elasticsearch/kibana --timeout 30s
bin/plugin --install elasticsearch/kibana --timeout 1h
```
Closes#4603.
Closes#4600.
Today during restart scenarios it is possible that we recover from a node that
has already been upgraded to version N+1. The node that we relocate to is
on version N and might not be able to read the index format from the node
we relocate from. This causes `IndexFormatToNewException` during
recovery but only after recovery has finished which can cause large
load spikes during the upgrade period.
Closes#4588
The node we need to lookup for attribute colelction might not be part
of the `DiscoveryNodes` anymore due to node failure or shutdown. This
commit adds a check and removes the shard from the iteration.
Closes#4589
This adds the field data circuit breaker, which is used to estimate
the amount of memory required to load field data before loading it. It
then raises a CircuitBreakingException if the limit is exceeded.
It is configured with two parameters:
`indices.fielddata.cache.breaker.limit` - the maximum number of bytes
of field data to be loaded before circuit breaking. Defaults to
`indices.fielddata.cache.size` if set, unbounded otherwise.
`indices.fielddata.cache.breaker.overhead` - a contast for all field
data estimations to be multiplied with before aggregation. Defaults to
1.03.
Both settings can be configured dynamically using the cluster update
settings API.
Today we try to detect if we need to generate the mapping or not in
the all mapper. This is error prone since it misses conditions if not
explicitly added. We should rather similate the generation instead.
This commit also adds a random test to check if the settings
of the all field mapper are correctly applied.
Closes#4579Closes#4581
today if a specific feature is disabled for term vectors with something
like 'store_term_vector_positions = false' term vectors might be disabeled
alltogether even if 'store_term_vectors=true' in the mapping. This depends on the
order of the values in the mapping since the more specific one might override
the less specific on.
Closes#4582
Currently there are two get aliases apis that both have the same functionality, but have a different response structure. The reason for having 2 apis is historic.
The GET _alias api was added in 0.90.x and is more efficient since it only sends the needed alias data from the cluster state between the master node and the node that received the request. In the GET _aliases api the complete cluster state is send to the node that received the request and then the right information is filtered out and send back to the client.
The GET _aliases api should be removed in favour for the alias api
Closes to #4539
* `ignore_unavailable` - Controls whether to ignore if any specified indices are unavailable, this includes indices that don't exist or closed indices. Either `true` or `false` can be specified.
* `allow_no_indices` - Controls whether to fail if a wildcard indices expressions results into no concrete indices. Either `true` or `false` can be specified. For example if the wildcard expression `foo*` is specified and no indices are available that start with `foo` then depending on this setting the request will fail. This setting is also applicable when `_all`, `*` or no index has been specified.
* `expand_wildcards` - Controls to what kind of concrete indices wildcard indices expression expand to. If `open` is specified then the wildcard expression if expanded to only open indices and if `closed` is specified then the wildcard expression if expanded only to closed indices. Also both values (`open,closed`) can be specified to expand to all indices.
Closes to #4436
Results in less data being sent over the wire, as the Cat API does not
need to have the whole cluster state.
Also added matchers for hasKey() for immutable open map (I think we should
add more of those to have map style assertions).
Closes#4455
Nik Everett
Simon Willnauer
Honza Král
Martijn van Groningen
Adrien Grand
Luca Cavanna
Alexander Reelsen
Lee Hinman
Igor Motov
Andrew Raines
David Pilato
sam
Britta Weber
Florian Schilling
Shay Banon
Dawid Weiss