Trying to choose the most appropriate persistent message queue for the following needs

gistfile1.txt

1) It should be persistent at the most deep meaning of that word. 
All events need to be available forever. 
Persistent storage will never be cleaned up in future.

2) Several producers will generate approximately 10 millions new events per day

3) When system starts there will be some default number of different 
consumers which will be fed with some filtered stream of those events.

4) In runtime user will have capablity to define its own custom consumers 
which will start eating events from the very beginning. The starting point 
for consumer could be some timestamp mark. Performance of reading batches of 
events should not depend on choosen timestamp

5) ofcourse events have to be ordered by creation time.

6) each message is simply just set of [id, create_timestamp, json_body]. 
The message body size could vary.

I've already considered RabbitMQ and apache kafka but unfortunately none 
of them perfectly fits to my needs. May be I should look at some modern 
database which will be suitable enough for this kind of tasks. 

At this time I'm thinking about using PostgreSQL table but i'm guessing 
this decision is so far from the right choice. So any advice will be appreciated.