external
/
citus
mirror of https://github.com/citusdata/citus.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Apply suggestions from code review 

Co-authored-by: Steven Sheehy <17552371+steven-sheehy@users.noreply.github.com>
Co-authored-by: Marco Slot <marco.slot@gmail.com>
pull/7638/head
Jelte Fennema-Nio 2024-06-27 09:48:00 +02:00 committed by GitHub
parent d59466fce3
commit c3d1ad3f9d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/backend/distributed/README.md
View File

@ -2368,12 +2368,12 @@ necessary to change one of them to add a feature/fix a bug.
The rebalancing algorithm tries to find an optimal placement of shard groups
across nodes. This is not an easy job, because this is a [co-NP-complete
problem](https://en.wikipedia.org/wiki/Knapsack_problem). So instead going for
problem](https://en.wikipedia.org/wiki/Knapsack_problem). So instead of going for
the fully optimal solution it uses a greedy approach to reach a local
optimimum, which so far has proved effective in getting to a pretty optimal
optimum, which so far has proved effective in getting to a pretty optimal
solution.
Eventhough it won't result in the perfect balance, the greedy aproach has two
Even though it won't result in the perfect balance, the greedy approach has two
important practical benefits over a perfect solution:
1. It's relatively easy to understand why the algorithm decided on a certain move.
2. Every move makes the balance better. So if the rebalance is cancelled midway
@ -2395,7 +2395,7 @@ main usage for "Is a shard group allowed on a certain node?" is to be able to pi
specific shard group to a specific node.
There is one last definition that you should know to understand the algorithm
and that is "utilization". Utilization is total cost of all shard groups
and that is "utilization". Utilization is the total cost of all shard groups
divided by capacity. In practice this means that utilization is almost always
the same as cost because as explained above capacity is almost always 1. So if
you see "utilization" in the algorithm, for all intents and purposes you can
@ -2457,7 +2457,7 @@ moving shards around isn't free.
most 10% above or 10% below the average utilization then no moves are
necessary anymore (i.e. the nodes are balanced enough). The main reason for
this threshold is that these small differences in utilization are not
necessarily problematic and might very well resolve automatically over time
necessarily problematic and might very well resolve automatically over time. For example, consider a scenario in which
one shard gets mostly written in during the weekend, while another one during
the week. Moving shards on monday and that you then have to move back on
friday is not very helpful given the overhead of moving data around.