[IGNITE-12530] Pages list caching can cause IgniteOOME when checkpoint is triggered by "too many dirty pages" reason

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.8
• Fix Version/s: 2.8, 2.9
• Component/s: persistence
• Assignee: Aleksey Plekhanov
• Reporter: Aleksey Plekhanov
• Created: 13/Jan/20 14:16
• Updated: 24/Jan/20 13:22
• Resolved: 20/Jan/20 15:23

Description

When a checkpoint is triggered, we need some amount of page memory to store pages list on-heap cache.

If data region is too small, a checkpoint is triggered by "too many dirty pages" reason and pages list cache is rather big, we can get IgniteOutOfMemoryException.

Reproducer:

@Override protected IgniteConfiguration getConfiguration(String name) throws Exception {
    IgniteConfiguration cfg = super.getConfiguration(name);

    cfg.setDataStorageConfiguration(new DataStorageConfiguration()
        .setDefaultDataRegionConfiguration(new DataRegionConfiguration()
            .setPersistenceEnabled(true)
            .setMaxSize(50 * 1024 * 1024)
        ));

    return cfg;
}

@Test
public void testUpdatesNotFittingIntoMemoryRegion() throws Exception {
    IgniteEx ignite = startGrid(0);

    ignite.cluster().active(true);

    ignite.getOrCreateCache(DEFAULT_CACHE_NAME);

    try (IgniteDataStreamer<Object, Object> streamer = ignite.dataStreamer(DEFAULT_CACHE_NAME)) {
        for (int i = 0; i < 100_000; i++)
            streamer.addData(i, new byte[i % 2048]);
    }
}

Issue Links

• causes: IGNITE-12562 FreeListCachingTest.testPageListCacheLimit failing in Disk Page Compressions suite
• is caused by: IGNITE-6930 Optionally to do not write free list updates to WAL
• links to: GitHub Pull Request #7245