Another ArrayIndexOutOfBoundsException about qora

[catbref]

Working backwards from innermost stack frame:

mapdb/Volume.java:434

        public final DataInput2 getDataInput(long offset, int size) {
            return new DataInput2(chunks[(int)(offset >>> chunkShift)], (int) (offset&chunkSizeModMask));
        }

offset >>> chunkShift equates to 136, which is the non-existent array index in chunks.
chunkShift seems to be 20 so chunks are 1MB in size?

So the offset is wrong when passed at mapdb/StoreWAL.java:344:

            DataInput2 in = (DataInput2) log.getDataInput(r[0]&LOG_MASK_OFFSET, size);

r[0] comes from the modified transaction HashMap (lines 333-334):

        //check if record was modified in current transaction
        long[] r = modified.get(ioRecid);

Is modified somehow out of sync?

• cleared in reloadIndexFile but this shouldn't cause the issue above where there is a record
• set to PREALLOC in preallocate() but also shouldn't cause issue above
• set in put() line 228
• set in update() line 416
• set in compareAndSwap() line 477
• "tombstoned" in delete() line 528

Is the log ByteBufferVol corrupted/reset outside of a write-lock by another thread?

• logReset() only checks for structuralLock
• logReset() called by:
  • rollback() - under full write locks
  • replayLogFile() twice - called by commit() - under full write locks
  • StoreWAL constructor

Does this situation only occur on Windows and hence something to do with mmap/Windows caching?

Is it possible to store more than chunkSize in a chunk when ByteBufferVol should put extra bytes into next chunk? (Maybe add assert()s to mapdb/Volume.java to test this)