Search in sources :

Example 1 with InterruptDetectedException

use of org.apache.derby.iapi.util.InterruptDetectedException in project derby by apache.

the class FileContainer method newPage.

/**
 *	  Create a new page in the container.
 *
 *	  <BR> MT - thread aware - It is assumed that our caller (our super class)
 *	  has already arranged a logical lock on page allocation to only allow a
 *	  single thread through here.
 *
 *	  Adding a new page involves 2 transactions and 2 pages.
 *	  The User Transaction (UT) initiated the addPage call and expects a
 *	  latched page (owns by the UT) to be returned.
 *	  The Nested Top Transaction (NTT) is the transaction started by RawStore
 *	  inside an addPage call.  This NTT is committed before the page is
 *	  returned.  The NTT is used to accessed high traffic data structure such
 *	  as the AllocPage.
 *
 *	  This is outline of the algorithm used in adding a page:
 *	  1) find or make an allocPage which can handle the addding of a new page.
 *		Latch the allocPage with the NTT.
 *	  2) invalidate the allocation information cached by the container.
 *		Without the cache no page can be gotten from the container.  Pages
 *		already in the page cache is not affected.  Thus by latching the
 *		allocPage and invalidating the allocation cache, this NTT blocks out
 *		all page gets from this container until it commits.
 *	  3) the allocPage determines which page can be allocated, mark that in its
 *		data structure (the alloc extent) and returns the page number of the
 *		new page.  This change is associated with the NTT.
 *	  4) the NTT gets or creates the new page in the page cache (bypassing the
 *		lookup of the allocPage since that is already latched by the NTT and
 *		will deadlock).
 *	  5) the NTT initializes the page (mark it is being a VALID page).
 *	  6) the page latch is transfered to the UT from the NTT.
 *	  7) the new page is returned, latched by UT
 *
 *	  If we use an NTT, the caller has to commit the NTT to release the
 *	  allocPage latch.  If we don't use an NTT, the allocPage latch is released
 *	  as this routine returns.
 *
 *	  @param userHandle - the container handle opened by the user transaction,
 *						use this to latch the new user page
 *	  @param ntt - the nested top transaction for the purpose of allocating the new page
 *						If ntt is null, use the user transaction for allocation.
 *	  #param allocHandle - the container handle opened by the ntt,
 *						use this to latch the alloc page
 *
 *	  @exception StandardException Standard Derby error policy
 */
protected BasePage newPage(BaseContainerHandle userHandle, RawTransaction ntt, BaseContainerHandle allocHandle, boolean isOverflow) throws StandardException {
    // NOTE: we are single threaded thru this method, see MT comment
    boolean useNTT = (ntt != null);
    // if ntt is null, use user transaction
    if (!useNTT)
        ntt = userHandle.getTransaction();
    // last allocated page
    long lastPage;
    // last pre-allcated page
    long lastPreallocPage;
    long pageNumber = // init to appease compiler
    ContainerHandle.INVALID_PAGE_NUMBER;
    // the page number of the new page
    // the identity of the new page
    PageKey pkey;
    // if true, we are trying to reuse a page
    boolean reuse;
    /* in case the page recommeded by allocPage is not committed yet, may
		/* need to retry a couple of times */
    boolean retry;
    int numtries = 0;
    int maxTries = InterruptStatus.MAX_INTERRUPT_RETRIES;
    long startSearch = lastAllocatedPage;
    // the alloc page
    AllocPage allocPage = null;
    // the new page
    BasePage page = null;
    try {
        do {
            // we don't expect we need to retry
            retry = false;
            synchronized (allocCache) {
                if (SanityManager.DEBUG) {
                    SanityManager.ASSERT(ntt.getId().equals(allocHandle.getTransaction().getId()));
                    if (useNTT)
                        SanityManager.ASSERT(!ntt.getId().equals(userHandle.getTransaction().getId()));
                }
                /* find an allocation page that can handle adding a new 
                     * page.
                     *
                     * allocPage is unlatched when the ntt commits. The new 
                     * page is initialized by the ntt but the latch is 
                     * transfered to the user transaction before the allocPage 
                     * is unlatched.  The allocPage latch prevents almost any 
                     * other reader or writer from finding the new page until 
                     * the ntt is committed and the new page is latched by the
                     * user transaction.
                     *
                     * (If the page is being reused, it is possible for another
                     * xact which kept a handle on the reused page to find the 
                     * page during the transfer UT -> NTT. If this unlikely 
                     * even occurs and the transfer fails [see code relating 
                     * to transfer below], we retry from the beginning.)
                     *
                     * After the NTT commits a reader (getNextPageNumber) may 
                     * get the page number of the newly allocated page and it 
                     * will wait for the new page and latch it when the user 
                     * transaction commits, aborts or unlatches the new page. 
                     * Whether the user transaction commits or aborts, the new 
                     * page stay allocated.
                     *
                     * RESOLVE: before NTT rolls back (or commits) the latch is
                     * released.  To repopulate the allocation cache, need to 
                     * get either the container lock on add page, or get a per 
                     * allocation page lock.
                     *
                     * This blocks all page read (getPage) from accessing this 
                     * alloc page in this container until the alloc page is 
                     * unlatched.  Those who already have a page handle into 
                     * this container are unaffected.
                     *
                     * In other words, allocation blocks out reader (of any 
                     * page that is managed by this alloc page) by the latch 
                     * on the allocation page.
                     *
                     * Note that write page can proceed as usual.
                     */
                try {
                    allocPage = findAllocPageForAdd(allocHandle, ntt, startSearch);
                } catch (InterruptDetectedException e) {
                    // hold. We release it when we do "continue" below.
                    if (--maxTries > 0) {
                        // Clear firstAllocPageNumber, i.e. undo side
                        // effect of makeAllocPage, so retry will work
                        firstAllocPageNumber = ContainerHandle.INVALID_PAGE_NUMBER;
                        retry = true;
                        // needs) and retry writeRAFHeader.
                        try {
                            Thread.sleep(InterruptStatus.INTERRUPT_RETRY_SLEEP);
                        } catch (InterruptedException ee) {
                            // This thread received an interrupt as
                            // well, make a note.
                            InterruptStatus.setInterrupted();
                        }
                        continue;
                    } else {
                        throw StandardException.newException(SQLState.FILE_IO_INTERRUPTED, e);
                    }
                }
                allocCache.invalidate(allocPage, allocPage.getPageNumber());
            }
            if (SanityManager.DEBUG) {
                if (allocPage == null)
                    allocCache.dumpAllocationCache();
                SanityManager.ASSERT(allocPage != null, "findAllocPageForAdd returned a null alloc page");
            }
            // 
            // get the next free page's number.
            // for case 1, page number > lastPreallocPage
            // for case 2, page number <= lastPage
            // for case 3, lastPage < page number <= lastPreallocPage
            // 
            pageNumber = allocPage.nextFreePageNumber(startSearch);
            // need to distinguish between the following 3 cases:
            // 1) the page has not been allocate or initalized.
            // Create it in the page cache and sync it to disk.
            // 2) the page is being re-allocated.
            // We need to read it in to re-initialize it
            // 3) the page has been preallocated.
            // Create it in the page cache and don't sync it to disk
            // 
            // first find out the current last initialized page and
            // preallocated page before the new page is added
            lastPage = allocPage.getLastPagenum();
            lastPreallocPage = allocPage.getLastPreallocPagenum();
            reuse = pageNumber <= lastPage;
            // no address translation necessary
            pkey = new PageKey(identity, pageNumber);
            if (reuse) {
                // if re-useing a page, make sure the deallocLock on the new
                // page is not held.  We only need a zero duration lock on
                // the new page because the allocPage is latched and this
                // is the only thread which can be looking at this
                // pageNumber.
                RecordHandle deallocLock = BasePage.MakeRecordHandle(pkey, RecordHandle.DEALLOCATE_PROTECTION_HANDLE);
                if (!getDeallocLock(allocHandle, deallocLock, false, /* nowait */
                true)) {
                    // until we get a brand new page.
                    if (numtries == 0) {
                        startSearch = ContainerHandle.INVALID_PAGE_NUMBER;
                        lastAllocatedPage = pageNumber;
                    } else
                        // continue from where we were
                        startSearch = pageNumber;
                    numtries++;
                    // We have to unlatch the allocPage so that if that
                    // transaction rolls back, it won't deadlock with this
                    // transaction.
                    allocPage.unlatch();
                    allocPage = null;
                    retry = true;
                } else {
                    // we got the lock, next time start from there
                    lastAllocatedPage = pageNumber;
                }
            } else {
                // deallocated pages
                if (numtries > 0)
                    lastAllocatedPage = ContainerHandle.INVALID_PAGE_NUMBER;
                else
                    lastAllocatedPage = pageNumber;
            }
            // Retry from the beginning if necessary.
            if (retry)
                continue;
            // If we get past here must have (retry == false)
            if (SanityManager.DEBUG) {
                SanityManager.ASSERT(retry == false);
            }
            if (SanityManager.DEBUG) {
                // ASSERT lastPage <= lastPreallocPage
                if (lastPage > lastPreallocPage) {
                    SanityManager.THROWASSERT("last page " + lastPage + " > lastPreallocPage " + lastPreallocPage);
                }
            }
            // No I/O at all if this new page is requested as part of a
            // create and load statement or this new page is in a temporary
            // container.
            // 
            // In the former case, BaseContainer will allow the
            // MODE_UNLOGGED bit to go thru to the nested top transaction
            // alloc handle.  In the later case, there is no nested top
            // transaction and the alloc handle is the user handle, which
            // is UNLOGGED.
            boolean noIO = (allocHandle.getMode() & ContainerHandle.MODE_UNLOGGED) == ContainerHandle.MODE_UNLOGGED;
            // or in a create container.
            if (!noIO && (bulkIncreaseContainerSize || (pageNumber > lastPreallocPage && pageNumber > PreAllocThreshold))) {
                allocPage.preAllocatePage(this, PreAllocThreshold, PreAllocSize);
            }
            // update last preAllocated Page, it may have been changed by
            // the preAllocatePage call.  We don't want to do the sync if
            // preAllocatePage already took care of it.
            lastPreallocPage = allocPage.getLastPreallocPagenum();
            boolean prealloced = pageNumber <= lastPreallocPage;
            // Argument to the create is an array of ints.
            // The array is only used for new page creation or for creating
            // a preallocated page, not for reuse.
            // 0'th element is the page format
            // 1'st element is whether or not to sync the page to disk
            // 2'nd element is pagesize
            // 3'rd element is spareSpace
            PageCreationArgs createPageArgs = new PageCreationArgs(StoredPage.FORMAT_NUMBER, prealloced ? 0 : (noIO ? 0 : CachedPage.WRITE_SYNC), pageSize, spareSpace, minimumRecordSize, 0);
            // RESOLVE: right now, there is no re-mapping of pages, so
            // pageOffset = pageNumber*pageSize
            long pageOffset = pageNumber * pageSize;
            try {
                page = initPage(allocHandle, pkey, createPageArgs, pageOffset, reuse, isOverflow);
            } catch (StandardException se) {
                if (SanityManager.DEBUG) {
                    SanityManager.DEBUG_PRINT("FileContainer", "got exception from initPage:" + "\nreuse = " + reuse + "\nsyncFlag = " + createPageArgs.syncFlag + "\nallocPage = " + allocPage);
                }
                allocCache.dumpAllocationCache();
                throw se;
            }
            if (SanityManager.DEBUG) {
                SanityManager.ASSERT(page != null, "initPage returns null page");
                SanityManager.ASSERT(page.isLatched(), "initPage returns unlatched page");
            }
            // allocate the page in the allocation page bit map
            allocPage.addPage(this, pageNumber, ntt, userHandle);
            if (useNTT) {
                // transfer the page latch from NTT to UT.
                // 
                // after the page is unlatched by NTT, it is still
                // protected from being found by almost everybody else
                // because the alloc page is still latched and the alloc
                // cache is invalidated.
                // 
                // However it is possible for the page to be
                // found by threads who specifically ask for this
                // pagenumber (e.g. HeapPostCommit).
                // We may find that such a thread has latched the page.
                // We shouldn't wait for it because we have the alloc page
                // latch, and this could cause deadlock (e.g.
                // HeapPostCommit might call removePage and this would wait
                // on the alloc page).
                // 
                // We may instead find that we can latch the page, but that
                // another thread has managed to get hold of it during the
                // transfer and either deallocated it or otherwise change it
                // (add rows, delete rows etc.)
                // 
                // Since this doesn't happen very often, we retry in these
                // 2 cases (we give up the alloc page and page and we start
                // this method from scratch).
                // 
                // If the lock manager were changed to allow latches to be
                // transferred between transactions, wouldn't need to
                // unlatch to do the transfer, and would avoid having to
                // retry in these cases (DERBY-2337).
                page.unlatch();
                page = null;
                // need to find it in the cache again since unlatch also
                // unkept the page from the cache
                page = (BasePage) pageCache.find(pkey);
                page = latchPage(userHandle, page, false);
                if (page == null || // rows (including deleted rows)
                page.recordCount() != 0 || page.getPageStatus() != BasePage.VALID_PAGE) {
                    retry = true;
                    if (page != null) {
                        page.unlatch();
                        page = null;
                    }
                    allocPage.unlatch();
                    allocPage = null;
                }
            }
        // if ntt is null, no need to transfer.  Page is latched by user
        // transaction already.  Will be no need to retry.
        // the alloc page is unlatched in the finally block.
        } while (retry == true);
        // At this point, should have a page suitable for returning
        if (SanityManager.DEBUG)
            SanityManager.ASSERT(page.isLatched());
    } catch (StandardException se) {
        if (page != null)
            page.unlatch();
        page = null;
        // rethrow error
        throw se;
    } finally {
        if (!useNTT && allocPage != null) {
            allocPage.unlatch();
            allocPage = null;
        }
    // NTT is committed by the caller
    }
    if (SanityManager.DEBUG)
        SanityManager.ASSERT(page.isLatched());
    // at a time in the future.
    if (bulkIncreaseContainerSize) {
        bulkIncreaseContainerSize = false;
        PreAllocSize = DEFAULT_PRE_ALLOC_SIZE;
    }
    if (!isOverflow && page != null)
        setLastInsertedPage(pageNumber);
    // logging, this is an estimate only
    if (estimatedPageCount >= 0)
        estimatedPageCount++;
    if (!this.identity.equals(page.getPageId().getContainerId())) {
        if (SanityManager.DEBUG) {
            SanityManager.THROWASSERT("just created a new page from a different container" + "\n this.identity = " + this.identity + "\n page.getPageId().getContainerId() = " + page.getPageId().getContainerId() + "\n userHandle is: " + userHandle + "\n allocHandle is: " + allocHandle + "\n this container is: " + this);
        }
        throw StandardException.newException(SQLState.DATA_DIFFERENT_CONTAINER, this.identity, page.getPageId().getContainerId());
    }
    // return the newly added page
    return page;
}
Also used : StandardException(org.apache.derby.shared.common.error.StandardException) PageKey(org.apache.derby.iapi.store.raw.PageKey) RecordHandle(org.apache.derby.iapi.store.raw.RecordHandle) InterruptDetectedException(org.apache.derby.iapi.util.InterruptDetectedException)

Aggregations

PageKey (org.apache.derby.iapi.store.raw.PageKey)1 RecordHandle (org.apache.derby.iapi.store.raw.RecordHandle)1 InterruptDetectedException (org.apache.derby.iapi.util.InterruptDetectedException)1 StandardException (org.apache.derby.shared.common.error.StandardException)1