Not 100% sure, but yes, I believe this is correct. One of the servers
would get 0-99, the other 100-199. The server to use up that batch of
100 values would then request 200-299, etc. Setting the cache to be 0
would likely impact the performance of Phoenix.
Using some external system to perform the global ordering may result in
better performance. Some testing would probably be needed :)
For the final rowKeyOrderSaltedTable question, yes, I believe that query
would work as you intend.
I am using Phoenix 4.8.1 with HBase 1.2.3 and the Phoenix Query Server.
I want to use a sequence to generate a monotonically increasing id for
each row. Since the documentation states that 100 sequence numbers are
cached by default in the client (in my case, I assume the caching would
happen in the query server), what is the behavior if I have 2 query
servers (load-balanced)? Does this mean Server A would generate numbers
starting from 0, and Server B would generate numbers starting from 100?
I need to make sure that the id is in order on a global basis for the
whole table. Would setting the CACHE to 0 be the best of achieving this?
Also, as the ID is monotonically increasing, I plan to salt the table
using something like (no manual split points):
CREATE TABLE mytable (id BIGINT NOT NULL PRIMARY KEY, VARCHAR)
SALT_BUCKETS = 20;
Without setting phoenix.query.rowKeyOrderSaltedTable to true, would I
still be able to get my records in order if I select them using
something like this?
SELECT * FROM mytable WHERE id > 5 && id < 100 ORDER BY id