I configured coordinator’s
PT10S, but coordinator apply LoadRule too slow.
As looks like below coordinator log, it’s more than 5 minutes (much higher than 10 seconds)
2023-01-05T02:28:44,944 INFO [qtp1522549999-137] org.apache.druid.metadata.IndexerSQLMetadataStorageCoordinator - Published segments to DB: [test-juhong-segment-loading_2023-01-01T00:00:00.000Z_2023-01-02T00:00:00.000Z_2023-01-05T02:28:43.948Z]
2023-01-05T02:35:22,039 INFO [Coordinator-Exec--0] org.apache.druid.server.coordinator.rules.LoadRule - Assigning 'primary' for segment [test-juhong-segment-loading_2023-01-01T00:00:00.000Z_2023-01-02T00:00:00.000Z_2023-01-05T02:28:43.948Z] to server [ip-172-31-201-235.ap-northeast-1.compute.internal:8083] in tier [_default_tier]
is the source test-juhong fully available in 10s after ingestion?
I didn’t understand your question.
coordinator/global/time metrics are very high.
org.apache.druid.server.coordinator.duty.BalanceSegments is notably high (more than 500s).
I tried to increase dynamic config
balancerComputeThreads 4 to 16 but it didn’t work.
my question was to try to find out if there is some network issue between the cluster and deep storage which slowing the fetching of segments. How many segments are created by the ingestion and what is the total size?
I found emitBalancingStats config is root cause.
coordinator/global/time metrics and it was so high.
BalanceSegments duty spent more than 500 seconds.
When I disable emit balancing stats,
BalanceSegments duty spent less than 1 second, and segment loading also be faster.