WriteLoadConstraintMonitor will not call reroute if no nodes are below threshold #136925
Conversation
DiannaHohensee
added
>non-issue
:Distributed Coordination/Allocation
|
Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination) |
…w threshold Closes ES-13237
DiannaHohensee
force-pushed
the
2025/10/20/ES-13237
branch
from
cd4b73a to
3270aa5
Compare
|
I'm starting to wonder whether we'd want to check for index vs search nodes in this.. We could have a cluster with 1 index node and 1 search node, and try to rebalance the hot-spot. I was just discussing DiscoveryNodeRole with Zhubo for the IndexBalanceAllocationDecider, and there are SEARCH_ROLE and INDEX_ROLE roles defined in stateful. So I believe it is doable. |
...r/src/main/java/org/elasticsearch/cluster/routing/allocation/WriteLoadConstraintMonitor.java
Show resolved
Hide resolved
|
I've got the index vs search roles handled now. Ready for another look. |
| * state contains #(numberOfIndexNodes) nodes with {@link DiscoveryNodeRole#INDEX_ROLE}, assigning the primary shards to those nodes, | ||
| * and #(numberOfSearchNodes) nodes with {@link DiscoveryNodeRole#SEARCH_ROLE}. | ||
| */ | ||
| public static ClusterState buildServerlessRoleNodes( |
There was a problem hiding this comment.
@zhubotang-wq this new utility method might be useful in your work, too, if you didn't already build something.
| public static final Map<String, ThreadPoolUsageStats> ZERO_USAGE_THREAD_POOL_USAGE_MAP = Map.of( | ||
| ThreadPool.Names.WRITE, | ||
| new NodeUsageStatsForThreadPools.ThreadPoolUsageStats(5, 0, 0) | ||
| ); |
There was a problem hiding this comment.
This looks like test code in production classes?
There was a problem hiding this comment.
Moved the to the test file 👍
| int numberOfSearchNodes, | ||
| int numberOfHotSpottingNodes | ||
| ) { | ||
| final long queueLatencyThresholdMillis = randomLongBetween(1000, 5000); |
There was a problem hiding this comment.
Nit: should we assert that numberOfHotSpottingNodes <= numberOfIndexNodes
| return state.build(); | ||
| } | ||
|
|
||
| public record IndexState(IndexMetadata indexMetadata, IndexRoutingTable.Builder indexRoutingTableBuilder) {} |
There was a problem hiding this comment.
Nit: could this record be private?
DiannaHohensee
added
the
auto-merge-without-approval
3 participants
The write load monitor will no longer attempt to address
hot-spots with a reroute request if there are no nodes
below the queue latency threshold to receive relocated
shards.
Excludes search role nodes when considering hot-spotting
nodes and relocation target nodes.
Closes ES-13237