You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Vladimir Steshin (Jira)" <ji...@apache.org> on 2023/08/01 08:36:00 UTC

[jira] [Commented] (IGNITE-19904) Assertion in defragmentation

    [ https://issues.apache.org/jira/browse/IGNITE-19904?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17749563#comment-17749563 ] 

Vladimir Steshin commented on IGNITE-19904:
-------------------------------------------

Caused by concurrent default checkpointer which clears shared 
{code:java}
CheckpointProgress#clearCounters()
{code}
and rises hidden NPE in `evictedPagesCounter().`:
{code:java}
@Override public void CheckpointProgressImpl#updateEvictedPages(int delta) {
        A.ensure(delta > 0, "param must be positive");

        if (evictedPagesCounter() != null)
            evictedPagesCounter().addAndGet(delta);
    }
{code}
while flushing replaced in `PageMemoryImpl#allocatePage(int grpId, int partId, byte flags)`



> Assertion in defragmentation
> ----------------------------
>
>                 Key: IGNITE-19904
>                 URL: https://issues.apache.org/jira/browse/IGNITE-19904
>             Project: Ignite
>          Issue Type: Bug
>    Affects Versions: 2.12
>            Reporter: Vladimir Steshin
>            Priority: Major
>              Labels: ise
>         Attachments: default-config.xml, failure2.16_with_thread_dump.log, ignite.log, jvm.opts
>
>
> Defragmentaion fails with:
> {code:java}
> java.lang.AssertionError: Invalid state. Type is 0! pageId = 0001000d00024cbf
> 		at org.apache.ignite.internal.processors.cache.persistence.pagemem.PageMemoryImpl.copyPageForCheckpoint(PageMemoryImpl.java:1359) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> 		at org.apache.ignite.internal.processors.cache.persistence.pagemem.PageMemoryImpl.checkpointWritePage(PageMemoryImpl.java:1277) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> 		at org.apache.ignite.internal.processors.cache.persistence.checkpoint.CheckpointPagesWriter.writePages(CheckpointPagesWriter.java:208) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> 		at org.apache.ignite.internal.processors.cache.persistence.checkpoint.CheckpointPagesWriter.run(CheckpointPagesWriter.java:150) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> {code}
> Difficult to write a test. Can't reproduce on my computers :(. Flackly appears on a server (4 core x 4 cpu) with 100G of the test cache data and million+ pages to checkpoint during defragmentation. More often, this occurs with pageSize 1024 (to produce more pages).
> Regarding my diagnostic build, I suppose that a fresh, empty page is caught in defragmentation. Here is a page dump with test-expented PAGE_OVERHEAD (=64) and same error a bit before copyPageForCheckpoint():
> {code:java}
> org.apache.ignite.IgniteException: Wrong page type in checkpointWritePage1. Page: Data region = 'defragPartitionsDataRegion'.
>  FullPageId [pageId=281878703760205, effectivePageId=403727049549, grpId=-1368047378].
>  PageDump = page_id: 281878703760205, rel_id: 48603, cache_id: -1368047378, pin: 0, lock: 65536, tmp_buf: 72057594037927935, test_val: 1. data_hex: 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
> 		at org.apache.ignite.internal.processors.cache.persistence.pagemem.PageMemoryImpl.checkpointWritePage(PageMemoryImpl.java:1240) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> 		at org.apache.ignite.internal.processors.cache.persistence.checkpoint.CheckpointPagesWriter.writePages(CheckpointPagesWriter.java:208) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> 		at org.apache.ignite.internal.processors.cache.persistence.checkpoint.CheckpointPagesWriter.run(CheckpointPagesWriter.java:150) ~[ignite-core-2.16.0-SNAPSHOT.jar:2.16.0-SNAPSHOT]
> {code}
> 'test_val' is my diagnostic page prefix extension. Various numbers are assigned where tmp_buf is assigned (by `PageHeader::tempBufferPointer()`). '1' comes from PageHeader::initNew(long absPtr, long relative):
> {code:java}
>     public static void PageHeader::initNew(long absPtr, long relative) {
>         relative(absPtr, relative);
>         tempBufferPointer(absPtr, PageMemoryImpl.INVALID_REL_PTR);
>         GridUnsafe.putLong(absPtr, PAGE_MARKER);
>         GridUnsafe.putInt(absPtr + PAGE_PIN_CNT_OFFSET, 0);
>         // For diagnostic purposes.
>         PageUtils.setTestTmpValue(absPtr, 1);
>     }
>     
>     public static void PageUtils::setTestTmpValue(long absPtr, int val) {
>         // Next to page overhead
>         putInt(absPtr, PAGE_OVERHEAD - 4, val);
>     }
>     
>     public static final int PageMemoryImpl#PAGE_OVERHEAD = 64;
> {code}



--
This message was sent by Atlassian Jira
(v8.20.10#820010)