[IGNITE-15147] Possible leak in metrics in PageLockTracker

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.10
• Component/s: persistence
• Ignite Flags: Release Notes Required
• Assignee: Unassigned
• Reporter: Sergey Chugunov
• Created: 19/Jul/21 14:17
• Updated: 05/Nov/23 22:25

Description

In one of PageHandler#readPage methods there is the following code:

long pageAddr = readLock(pageMem, cacheId, pageId, page, lsnr);

if (pageAddr == 0L)
    return lockFailed;
try {
    PageIO io = pageIoRslvr.resolve(pageAddr);
    return h.run(cacheId, pageId, page, pageAddr, io, null, arg, intArg, statHolder);
}
finally {
    readUnlock(pageMem, cacheId, pageId, page, pageAddr, lsnr);
}

Here we obtain a read lock on a page by calling readLock method, its implementation is as following:

lsnr.onBeforeReadLock(cacheId, pageId, page);

long pageAddr = pageMem.readLock(cacheId, pageId, page);

lsnr.onReadLock(cacheId, pageId, page, pageAddr);

return pageAddr;

And here is a problem: in readLock we always call onReadLock for the page but onReadUnlock is called only if lock was acquired successfully.
Otherwise lock counters end up in incorrect state: onReadLock called although no lock was actually acqured.