You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@ignite.apache.org by "Mitchell Rathbun (BLOOMBERG/ 731 LEX)" <mr...@bloomberg.net> on 2020/09/23 16:34:11 UTC

OutOfMemoryException with Persistence and Eviction Enabled

We currently have a cache that is structured with a key of record type and a value that is a map from field id to field. So to update this cache, which has persistence enabled, we need to atomically load the value map for a key, add to that map, and write the map back to the cache. This can be done using invokeAll and a CacheEntryProcessor. However, when I test with a higher load (100k records with 50 fields each), I run into an OOM exception that I will post below. The cause of the exception is reported to be the failure to find a page to evict. However, even when I set the DataRegion's eviction threshold to .5 and the page eviction mode to RANDOM_2_LRU, I am still getting the same error. I have 2 main questions from this:

1. Why is it failing to evict a page even with a lower threshold and eviction enabled? Is it failing to reach the threshold somehow? Are non-data pages like metadata and index pages taken into account when determining if the threshold has been reached?

2. We don't have this issue when using IgniteDataStreamer to write large amounts of data to the cache, we just can't get transactional support at the same time. Why is this OOME an issue with regular cache puts but not with IgniteDataStreamer? I would think that any issues with checkpointing and eviction would also occur with IgniteDataStreamer.