ringles commented on a change in pull request #6296:
URL: https://github.com/apache/geode/pull/6296#discussion_r617664346
##########
File path:
geode-apis-compatible-with-redis/src/test/java/org/apache/geode/redis/internal/data/RedisHashTest.java
##########
@@ -258,6 +258,13 @@ public void hscanSnaphots_shouldExpireAfterExpiryPeriod() {
});
}
+ @Test
+ public void
hashSizeOverhead_shouldNotBeChanged_withoutForethoughtAndTesting() {
+ assertThat(RedisHash.PER_OBJECT_OVERHEAD).isEqualTo(8);
+
assertThat(RedisHash.getPerStringOverhead()).isEqualTo(RedisHash.PER_OBJECT_OVERHEAD
+ 46);
+
assertThat(RedisHash.getPerHashOverhead()).isEqualTo(RedisHash.PER_OBJECT_OVERHEAD
+ 116);
+ }
Review comment:
This turns out to be hard. We can create a ReflectionObjectSizer in the
constructor, and get the size of the object without any members added.
Then we'd have to examine at least one member being added, and attempt to
derive the per-object overhead as we added it. Except that it turns out that an
empty HashSet doesn't allocate some storage. When you add the first object, you
add both the object and the internal storage. So we could then add a second
object, and based on the length attempt to figure out what the 'internal
storage overhead' is...
Anyway, it's not trivial. But we might have to do it because it looks like -
no surprise and as feared - JVM v11 doesn't have the same overhead as v8.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
