honeymoose
/
OpenSearch
mirror of https://github.com/honeymoose/OpenSearch.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Make shard balancing deterministic if weights are identical 

It happens to be the case that the iteration order of a HashMaps
keyset might be different across runs. This can cause undeterministic
results in shard balancing if weights are identical and multiple shards
of the same index are eligable for relocation. This commit adds
a tie-breaker based on the shard ID to prioritise the lowest shard
ID. This also makes `AddIncrementallyTests#testAddNodesAndIndices`
reproducible.

Closes #4867
This commit is contained in:
Simon Willnauer 2014-01-23 15:45:24 +01:00
parent 3158776438
commit 592a411b2c
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/main/java/org/elasticsearch/cluster/routing/allocation/allocator/BalancedShardsAllocator.java
View File

@ -804,7 +804,10 @@ public class BalancedShardsAllocator extends AbstractComponent implements Shards
if ((srcDecision = maxNode.removeShard(shard)) != null) {
minNode.addShard(shard, srcDecision);
final float delta = weight.weight(operation, this, minNode, idx) - weight.weight(operation, this, maxNode, idx);
if (delta < minCost) {
if (delta < minCost ||
(candidate != null && delta == minCost && candidate.id() > shard.id())) {
/* this last line is a tie-breaker to make the shard allocation alg deterministic
* otherwise we rely on the iteration order of the index.getAllShards() which is a set.*/
minCost = delta;
candidate = shard;
decision = new Decision.Multi().add(allocationDecision).add(rebalanceDecision);