[ https://issues.apache.org/jira/browse/IGNITE-14003?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Ivan Daschinskiy updated IGNITE-14003: -------------------------------------- Description: Scenario 1. Start replicated cache on ignite node, memory region approx 6 Gb, heap 1Gb 2. Load significant amount of data to cache with values approx 200Kb (~20K kv pairs) 3. Start another node First node (supplier) will crash while initializing rebalance iterator with OOM Main reason -- all values, to whon pointed from leaf of BTree, are all loaded to buffer in BPlusTree#ForwardCursor. For replicated cache, 512 iterators for each partition are created at once. Reproducer is attached, run it with *-Xmx1G* was: Scenario 1. Start replicated cache on ignite node, memory region approx 6 Gb, heap 1Gb 2. Load significant amount of data to cache with values approx 200Kb (~20K kv pairs) 3. Start another node First node (supplier) will crash while initializing rebalance iterator with OOM Main reason -- all values, to whon pointed from leaf of BTree, are all loaded to buffer in BPlusTree#ForwardCursor. For replicated cache, 512 iterators for each partition are created at once. Reproducer is attached. > OOM on creating rebalance iterator while rebalancing cache with large values. > ----------------------------------------------------------------------------- > > Key: IGNITE-14003 > URL: https://issues.apache.org/jira/browse/IGNITE-14003 > Project: Ignite > Issue Type: Bug > Affects Versions: 2.9, 2.8.1, 2.9.1 > Reporter: Ivan Daschinskiy > Priority: Critical > Attachments: ReplicatedCacheRebalanceTest.java > > > Scenario > 1. Start replicated cache on ignite node, memory region approx 6 Gb, heap 1Gb > 2. Load significant amount of data to cache with values approx 200Kb (~20K kv > pairs) > 3. Start another node > First node (supplier) will crash while initializing rebalance iterator with > OOM > Main reason -- all values, to whon pointed from leaf of BTree, are all loaded > to buffer in BPlusTree#ForwardCursor. For replicated cache, 512 iterators for > each partition are created at once. > Reproducer is attached, run it with *-Xmx1G* -- This message was sent by Atlassian Jira (v8.3.4#803005)