Skip to content

Fix sharding- we need to divide by max_shards / shard, not by shard

Jcrespo requested to merge T334069 into main

We implemented sharding wrongly- however it worked because max_shards / # shards == # shards ( 16 / 4 = 4). However, we need to do it right now that we have an additional sharding (5 in total).

This is the effective change:

0 -> backup1005 -> backup1005 0 1 -> backup1005 -> backup1005 0 2 -> backup1005 -> backup1005 0 3 -> backup1005 -> backup1006 1 4 -> backup1006 -> backup1006 1 5 -> backup1006 -> backup1006 1 6 -> backup1006 -> backup1007 2 7 -> backup1006 -> backup1007 2 8 -> backup1007 -> backup1007 2 9 -> backup1007 -> backup1008 3 10 -> backup1007 -> backup1008 3 11 -> backup1007 -> backup1008 3 12 -> backup1008 -> backup1011 4 13 -> backup1008 -> backup1011 4 14 -> backup1008 -> backup1011 4 15 -> backup1008 -> backup1011 4

Note this affects default writing shards, not existing data, which means old data will be available on the previous location, unless resharded.

Bug: T334069

Merge request reports