I mean that you have accidentally insert string key, not UUID. During insertion, ignite checks types of key during updating index B+ tree and this check currently lead to CorruptedTreeException.
ср, 9 июн. 2021 г. в 17:37, Ivan Daschinsky <ivanda...@gmail.com>: > > Hi Marcus. Yes, it is a bug on ignite, but I suppose that you insert > UUID instead of String > Here is the actual cause of CorruptedTreeException: > > Caused by: java.lang.UnsupportedOperationException: Value type doesn't > match: exp=20, act=13 > at > org.apache.ignite.internal.processors.query.h2.database.inlinecolumn.AbstractInlineIndexColumn.ensureValueType(AbstractInlineIndexColumn.java:193) > > 20 -- is alias for UUID, 13 is alias for String. Could you please show > your code that inserts data? > > > ср, 9 июн. 2021 г. в 09:18, Lo, Marcus <marcus...@citi.com>: > > > > Hi, > > > > > > > > We have a 4 node ignite (v.2.10.0) cluster setup, running on redhat 7 with > > 60G non-heap cache and 10G heap, with persistence enabled. We find that all > > nodes would crash with CorruptedTreeException when we do the following: > > > > > > > > Start all nodes as background process, and activate the cluster > > Create a table and index via SQL: > > > > CREATE TABLE IF NOT EXISTS LIMIT_VIEW_STATUS > > > > ( > > > > viewId UUID, > > > > batchId INT, > > > > instance VARCHAR, > > > > nodes VARCHAR, > > > > status VARCHAR, > > > > eqtgContext VARCHAR, > > > > lastUpdateTime TIMESTAMP, > > > > PRIMARY KEY (status, viewId) > > > > ) > > > > with "template=RiskDataTemplate, affinity_key=viewId, > > cache_name=LimitViewStatusCache"; > > > > > > > > CREATE INDEX IF NOT EXISTS limit_view_status_batchid_instance_idx ON > > LIMIT_VIEW_STATUS (batchId,instance); > > > > Insert data into the table concurrently by 5 thin clients. > > Inactivate the cluster while thin clients are still inserting data. > > Stop the nodes one by one by killing the ignite process (normal SIGTERM > > kill). > > Restart all nodes one by one. > > Insert data again into the table concurrently by 5 thin clients, and then > > all the nodes would crash. > > > > > > > > Please find the attached for the config, log file (for one node) and zipped > > wal/pds files (for one node). I have changed the wal file size to 512Kb > > just to make the wal files small – the issue can be reproduced using > > default wal size (64Mb), or even very large wal size (slightly less than > > 2Gb). > > > > > > > > Can anyone shed some light on the issue? Thanks. > > > > > > > > Regards, > > > > Marcus > > > > > > > > -- > Sincerely yours, Ivan Daschinskiy -- Sincerely yours, Ivan Daschinskiy
