This also optimizes ZREVRANGEBYSCORE for pathological cases where a
sorted set contains many elements with the same score. Previously,
it would traverse the list from back to front in such a case.
This avoids the extra allocation of sizeof(double) for storing the score
of a zset entry in the hash table. Saves sizeof(double) + malloc
overhead = approx. 16 bytes per zset entry.
In order to trigger the bug what's needed is to call ZINTERSTORE
resulting into an empty set created, bug against a key that already
existed. The command was not propagated, so the replica ended with the
key that the master removed. Sequence of command to reproduce:
redis-cli hset 446 34 905
redis-cli hset 446 393 911
redis-cli zadd 966 0.085412045980529885 652
redis-cli zadd 645 0.25081839284432045 280
redis-cli zinterstore 446 2 966 645
networking related stuff moved into networking.c
moved more code
more work on layout of source code
SDS instantaneuos memory saving. By Pieter and Salvatore at VMware ;)
cleanly compiling again after the first split, now splitting it in more C files
moving more things around... work in progress
split replication code
splitting more
Sets split
Hash split
replication split
even more splitting
more splitting
minor change
