Previously, HealthMvcEndpoint stored the cached Health and its last
access time in two separate fields. Neither field was volatile and
no synchronization was used. This meant that there were potential
visibility problems. In a possible worst case scenario one field may
see the updated access time but an old health so it would incorrectly
believe that the old health was up-to-date and return it.
This commit reworks the endpoint to store the cached health and the
time at which it was created in a single, volatile field. This ensures
that the cached health and its creation time will be visible across
threads. Note that a race between threads when the cache is stale is
still possible. This race may result in multiple calls to the
delegate but these should be harmless.
Closes gh-9454
In Gradle 4.x, SourceSetOutput now has multiple classes directories
and getClassesDir() has been deprecated. This commit introduces the
use of reflection to use getClassesDirs() when it's available rather
than getClassesDir().
Closes gh-9559
This commit makes sure that if the `register-mbeans` property of the
Hikary datasource config is set, Spring Boot doesn't attempt to expose
the mbean again.
Closes gh-5114
This commit add the support for CustomConversions in
spring-data-cassandra. To customize, bean just need to be declared and
it will be auto-configured.
See gh-8534
Elastic have announced [1] that embedded Elasticsearch is no longer
supported. This commit brings us into line with that announcement by
removing the auto-configuration that would create an Elasticsearch
Node and NodeClient.
To use the Elasticsearch auto-configuration, a user must now provide
the address of one or more cluster nodes
(via the spring.elastisearch.cluster-nodes property) which will then
be used to create a TransportClient.
See gh-9374
[1] https://www.elastic.co/blog/elasticsearch-the-server
Previously, RedisTestServer only supported Jedis and would blow up if
only Lettuce was on the classpath. This commit defensively checks which
driver is available and chose the appropriate one, defaulting to Jedis.
Closes gh-9524
Reducing the default to 100ms is a good compromise to retain a quiet time
in for parallel execution and optimize for default, single-threaded
execution (such as test execution or regular application shutdown). The
shutdown timeout can be adjusted to fit specific application needs.
See gh-9526
Stephane Nicoll
Johnny Lim
Eddú Meléndez
Andy Wilkinson
Huang YunKun
Brian Clozel
Dave Syer
Madhura Bhave
Mark Paluch