During query parsing if a filter is encountered that extends from NoCacheFilter then the filter will not be given to the filter cache (also not wrapped in FilterCacheFilterWrapper).
Also if a filter directly or indirectly wraps a NoCacheFilter then that filter will also not be cached.
Relates to #4757
Although the empty paths are still registered in the java RestActions, that is only to return a meaningful error in case the index is not provided.
Index/indices are now mandatory.
Add transport.publish_port setting to allow users to specify the port
other cluster members should use when connecting to an instance. This
is needed for systems such as OpenShift, where cluster communication
needs to use a publicly accessibly proxy port, because the normal port
(9300) is bound to a private loopback IP address.
If a thread is not alive getting ThreadMXBean#getThreadInfo(long[], int)
places null elemnents in the returned array which are not repected
in the HotTheards API.
Closes#4775
`omit_term_freq_and_positions` was deprecated in `0.20` and
is not documented anymore. We should reject indices that are
created with this option in the future.
Closes#4722
With JUnit up to 4.10 there's no way to distinguish between what's shown in IDEs and test ids that need to be unique. A test Description is identified by just a string, that needs to be unique and is shown by IDEs. IDEs have slightly different behaviours when it comes to showing tests and suites titles. Some IDE (e.g. IntelliJ) strips the description on '.', which is why the '.' was replaced in the first place, in order to obtain the same behaviour on all IDEs. On the other hand the information printed out by RestReproduceInfoPrinter was wrong as the file path contained a '_' instead of a '.', which made the string to reproduce a failure useless in some cases. At the end of the day it seems better to just keep the dots and accept slightly different behaviours that are IDE dependent.
If the thread pools of an elasticsearch node cannot be shutdown
immediately, a wait of 10 seconds is added. This clashes with the
RPM scripts, as by default the init functions wait for 3 seconds
for a service to shutdown before a KILL signal is sent, resulting
in an unclean shutdown - not from an elasticsearch point of view,
but from init system point of view, as some lock files are left
around.
In order to prevent this the init script as well as the systemd
configuration now feature the same timeout than the debian package,
which is 20 seconds.
The await statement, which causes the 10 second delay can be found in
InternalNode.close()
Clinton Gormley
Martijn van Groningen
Simon Willnauer
Alexander Reelsen
Luca Cavanna
Honza Král
Adrien Grand
Andy Goldstein
b.hwang
mrsolo
Costin Leau
Andrew Raines
David Pilato
Alexander Reelsen